2026
Espinosa-Campos, Pablo; Liebe, Lucas; Nguyen, Thanh-Tung; Tau, Nhat-Quang; Wu, Yuheng; Dongman, Lee
ADAS-LORA: Efficient Federated LLM Tuning in Edge Systems by Decoupling Compute and Network Conference Forthcoming
Proceedings of the 46th IEEE International Conference on Distributed Computing Systems (ICDCS), Federated Foundation Models at the Edge Workshop, 2026, Forthcoming.
@conference{nokey,
title = {ADAS-LORA: Efficient Federated LLM Tuning in Edge Systems by Decoupling Compute and Network},
author = {Pablo Espinosa-Campos and Lucas Liebe and Thanh-Tung Nguyen and Nhat-Quang Tau and Yuheng Wu and Lee Dongman},
year = {2026},
date = {2026-06-25},
urldate = {2026-06-25},
booktitle = {Proceedings of the 46th IEEE International Conference on Distributed Computing Systems (ICDCS), Federated Foundation Models at the Edge Workshop, 2026},
keywords = {},
pubstate = {forthcoming},
tppubtype = {conference}
}
Yoon, Junho; Jeong, Jaemo; Kim, Hyunju; Lee, Dongman
DETACH : Decomposed Spatio-Temporal Alignment for Exocentric Video and Ambient Sensors with Staged Learning Conference Forthcoming
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2026 Forthcoming.
@conference{yoon2025detachb,
title = {DETACH : Decomposed Spatio-Temporal Alignment for Exocentric Video and Ambient Sensors with Staged Learning},
author = {Junho Yoon and Jaemo Jeong and Hyunju Kim and Dongman Lee},
editor = {jaemo jeong},
year = {2026},
date = {2026-06-05},
urldate = {2026-06-05},
organization = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2026},
keywords = {},
pubstate = {forthcoming},
tppubtype = {conference}
}
Wu, Yuheng; Gao, Xiangbo; Tau, Quang; Tu, Zhengzhong; Lee, Dongman
Background Fades, Foreground Leads: Curriculum-Guided Background Pruning for Efficient Foreground-Centric Collaborative Perception Conference Forthcoming
IEEE International Conference on Robotics & Automation, Forthcoming.
@conference{nokey,
title = {Background Fades, Foreground Leads: Curriculum-Guided Background Pruning for Efficient Foreground-Centric Collaborative Perception},
author = {Wu, Yuheng and Gao, Xiangbo and Tau, Quang and Tu, Zhengzhong and Lee, Dongman},
year = {2026},
date = {2026-02-01},
urldate = {2026-02-01},
booktitle = {IEEE International Conference on Robotics & Automation},
abstract = {Collaborative perception enhances the reliability and spatial coverage of autonomous vehicles by sharing complementary information across vehicles, offering a promising solution to long-tail scenarios that challenge single-vehicle perception. However, the bandwidth constraints of vehicular networks make transmitting the entire feature map impractical. Recent methods, therefore, adopt a foreground-centric paradigm, transmitting only predicted foreground-region features while discarding the background, which encodes essential context. We propose FadeLead, a foreground-centric framework that overcomes this limitation by learning to encapsulate background context into compact foreground features during training. At the core of our design is a curricular learning strategy that leverages background cues early on but progressively prunes them away, forcing the model to internalize context into foreground representations without transmitting background itself. Extensive experiments on both simulated and real-world benchmarks show that FadeLead outperforms prior methods under different bandwidth settings, underscoring the effectiveness of context-enriched foreground sharing.
},
keywords = {},
pubstate = {forthcoming},
tppubtype = {conference}
}
Collaborative perception enhances the reliability and spatial coverage of autonomous vehicles by sharing complementary information across vehicles, offering a promising solution to long-tail scenarios that challenge single-vehicle perception. However, the bandwidth constraints of vehicular networks make transmitting the entire feature map impractical. Recent methods, therefore, adopt a foreground-centric paradigm, transmitting only predicted foreground-region features while discarding the background, which encodes essential context. We propose FadeLead, a foreground-centric framework that overcomes this limitation by learning to encapsulate background context into compact foreground features during training. At the core of our design is a curricular learning strategy that leverages background cues early on but progressively prunes them away, forcing the model to internalize context into foreground representations without transmitting background itself. Extensive experiments on both simulated and real-world benchmarks show that FadeLead outperforms prior methods under different bandwidth settings, underscoring the effectiveness of context-enriched foreground sharing.
2025
Cheng, Jinghan; Nguyen, Thanh-Tung; Liebe, Lucas; Wu, Yuheng; Tau, Nhat-Quang; Espinosa, Pablo; Lee, Dongman
OctoCross: Workload-Aware Request Offloading Scheduling in Cross-Camera Collaboration Proceedings Article
In: Service-Oriented Computing, pp. 365–380, Springer Nature Singapore, 2025, ISBN: 978-981-95-5012-8.
@inproceedings{Cheng2025OctoCross,
title = {OctoCross: Workload-Aware Request Offloading Scheduling in Cross-Camera Collaboration},
author = {Jinghan Cheng and Thanh-Tung Nguyen and Lucas Liebe and Yuheng Wu and Nhat-Quang Tau and Pablo Espinosa and Dongman Lee},
url = {https://link.springer.com/chapter/10.1007/978-981-95-5012-8_27},
doi = {10.1007/978-981-95-5012-8_27},
isbn = {978-981-95-5012-8},
year = {2025},
date = {2025-12-04},
urldate = {2026-12-04},
booktitle = {Service-Oriented Computing},
pages = {365--380},
publisher = {Springer Nature Singapore},
abstract = {Cross-camera collaboration is emerging as a crucial strategy to support dynamic and heterogeneous real-world workloads in real-time video analytics (VA) systems. However, existing offloading methods lack the capability to jointly model spatial and temporal dynamics, which limits their offloading accuracy, leading to low system performances. We propose OctoCross, a spatiotemporal offloading framework for request offloading in mutli-camera VA systems. OctoCross introduces a novel spatiotemporal model that can jointly learns spatial and temporal dynamics to produce an actionable offloading plan in an end-to-end manner. Extensive evaluations on real-world multi-camera datasets show that OctoCross achieves up to 5.8× higher throughput and 3.2× lower latency compared to baselines, while maintaining high SLO compliance across diverse deployment scenarios.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Cross-camera collaboration is emerging as a crucial strategy to support dynamic and heterogeneous real-world workloads in real-time video analytics (VA) systems. However, existing offloading methods lack the capability to jointly model spatial and temporal dynamics, which limits their offloading accuracy, leading to low system performances. We propose OctoCross, a spatiotemporal offloading framework for request offloading in mutli-camera VA systems. OctoCross introduces a novel spatiotemporal model that can jointly learns spatial and temporal dynamics to produce an actionable offloading plan in an end-to-end manner. Extensive evaluations on real-world multi-camera datasets show that OctoCross achieves up to 5.8× higher throughput and 3.2× lower latency compared to baselines, while maintaining high SLO compliance across diverse deployment scenarios.
Kim, Hyunju; Lee, Dongman
Self-supervised Dual-view Framework with Tailored Negative Sampling for New Activity Detection Proceedings Article
In: Proceedings of the 34th ACM International Conference on Information and Knowledge Management, pp. 1313–1323, Association for Computing Machinery, New York, NY, USA, 2025, ISBN: 9798400720406.
@inproceedings{kim2025self,
title = {Self-supervised Dual-view Framework with Tailored Negative Sampling for New Activity Detection},
author = {Hyunju Kim and Dongman Lee},
url = {https://dl.acm.org/doi/10.1145/3746252.3761163},
doi = {10.1145/3746252.3761163},
isbn = {9798400720406},
year = {2025},
date = {2025-11-10},
urldate = {2025-11-10},
booktitle = {Proceedings of the 34th ACM International Conference on Information and Knowledge Management},
pages = {1313–1323},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
series = {CIKM '25},
abstract = {With the growing adoption of IoT, analyzing multi-modal sensor time series for human activity recognition has become crucial for intelligent and context-aware applications. While existing approaches assume a fixed set of known activities, real-world deployments encounter new activities not present in the training data. Detecting them is challenging due to overlapping patterns between known and new activities, high intra-class variability, and sensor heterogeneity across datasets. To address these challenges, we propose CLAN, a novel self-supervised dual-view framework for new activity detection. It employs a two-tower architecture that extracts discriminative representations from both time and frequency domains. By treating multiple strongly augmented views of known activity samples as negatives, the model learns invariant representations that effectively distinguish new activities. In addition, a dataset-aware augmentation selection mechanism adaptively determines transformations tailored to each dataset's characteristics, thereby enhancing generalization across diverse sensor environments. Extensive experiments on five real-world human activity datasets demonstrate that CLAN consistently outperforms new activity detection baselines, achieving up to 9.24% improvement in AUROC.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
With the growing adoption of IoT, analyzing multi-modal sensor time series for human activity recognition has become crucial for intelligent and context-aware applications. While existing approaches assume a fixed set of known activities, real-world deployments encounter new activities not present in the training data. Detecting them is challenging due to overlapping patterns between known and new activities, high intra-class variability, and sensor heterogeneity across datasets. To address these challenges, we propose CLAN, a novel self-supervised dual-view framework for new activity detection. It employs a two-tower architecture that extracts discriminative representations from both time and frequency domains. By treating multiple strongly augmented views of known activity samples as negatives, the model learns invariant representations that effectively distinguish new activities. In addition, a dataset-aware augmentation selection mechanism adaptively determines transformations tailored to each dataset's characteristics, thereby enhancing generalization across diverse sensor environments. Extensive experiments on five real-world human activity datasets demonstrate that CLAN consistently outperforms new activity detection baselines, achieving up to 9.24% improvement in AUROC.
Wu, Yuheng; Nguyen, Thanh-Tung; Liebe, Lucas; Tau, Nhat-Quang; Campos, Pablo Espinosa; Cheng, Jinghan; Lee, Dongman
How2Compress: Scalable and Efficient Edge Video Analytics via Adaptive Granular Video Compression Proceedings Article
In: Proceedings of the 33rd ACM International Conference on Multimedia, pp. 12006–12015, Association for Computing Machinery, New York, NY, USA, 2025, ISBN: 9798400720352.
@inproceedings{nokey,
title = {How2Compress: Scalable and Efficient Edge Video Analytics via Adaptive Granular Video Compression},
author = {Yuheng Wu and Thanh-Tung Nguyen and Lucas Liebe and Nhat-Quang Tau and Pablo Espinosa Campos and Jinghan Cheng and Dongman Lee},
url = {https://doi.org/10.1145/3746027.3754922},
doi = {10.1145/3746027.3754922},
isbn = {9798400720352},
year = {2025},
date = {2025-10-28},
urldate = {2025-09-30},
booktitle = {Proceedings of the 33rd ACM International Conference on Multimedia},
pages = {12006–12015},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
series = {MM '25},
abstract = {With the rapid proliferation of the Internet of Things, video analytics has become a cornerstone application in wireless multimedia sensor networks. To support such applications under bandwidth constraints, learning-based adaptive quantization for video compression have demonstrated strong potential in reducing bitrate while maintaining analytical accuracy. However, existing frameworks often fail to fully exploit the fine-grained quality control enabled by modern blockbased video codecs, leaving significant compression efficiency untapped.
In this paper, we present How2Compress, a simple yet effective framework designed to enhance video compression efficiency through precise, fine-grained quality control at the macroblock level. How2Compress is a plug-and-play module and can be seamlessly integrated into any existing edge video analytics pipelines. We implement How2Compress on the H.264 codec and evaluate its performance across diverse real-world scenarios. Experimental results show that How2Compress achieves up to 50.4% bitrate savings and outperforms baselines by up to 3.01x without compromising accuracy, demonstrating its practical effectiveness and efficiency.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
With the rapid proliferation of the Internet of Things, video analytics has become a cornerstone application in wireless multimedia sensor networks. To support such applications under bandwidth constraints, learning-based adaptive quantization for video compression have demonstrated strong potential in reducing bitrate while maintaining analytical accuracy. However, existing frameworks often fail to fully exploit the fine-grained quality control enabled by modern blockbased video codecs, leaving significant compression efficiency untapped.
In this paper, we present How2Compress, a simple yet effective framework designed to enhance video compression efficiency through precise, fine-grained quality control at the macroblock level. How2Compress is a plug-and-play module and can be seamlessly integrated into any existing edge video analytics pipelines. We implement How2Compress on the H.264 codec and evaluate its performance across diverse real-world scenarios. Experimental results show that How2Compress achieves up to 50.4% bitrate savings and outperforms baselines by up to 3.01x without compromising accuracy, demonstrating its practical effectiveness and efficiency.
In this paper, we present How2Compress, a simple yet effective framework designed to enhance video compression efficiency through precise, fine-grained quality control at the macroblock level. How2Compress is a plug-and-play module and can be seamlessly integrated into any existing edge video analytics pipelines. We implement How2Compress on the H.264 codec and evaluate its performance across diverse real-world scenarios. Experimental results show that How2Compress achieves up to 50.4% bitrate savings and outperforms baselines by up to 3.01x without compromising accuracy, demonstrating its practical effectiveness and efficiency.
Lee, S.; Park, S.; Yang, J.; Lee, Dongman; Kim, J.; Cha, M.
Slum Detection Beyond Labels: A Decade of Poverty Mapping Using Satellite Imagery and Minimal Ground Truth Data Proceedings Article
In: Proceedings of the 2025 International Conference on Information Technology for Social Good, pp. 41–46, Association for Computing Machinery, New York, NY, USA, 2025, ISBN: 9798400720895.
@inproceedings{nokey,
title = {Slum Detection Beyond Labels: A Decade of Poverty Mapping Using Satellite Imagery and Minimal Ground Truth Data},
author = {S. Lee and S. Park and J. Yang and Dongman Lee and J. Kim and M. Cha},
url = {https://dl.acm.org/doi/10.1145/3748699.3749772},
doi = {10.1145/3748699.3749772},
isbn = {9798400720895},
year = {2025},
date = {2025-09-03},
urldate = {2025-09-03},
booktitle = {Proceedings of the 2025 International Conference on Information Technology for Social Good},
pages = {41–46},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
series = {GoodIT '25},
abstract = {Deep learning applied to very high resolution (VHR) satellite imagery has become a powerful method for analyzing urban poverty at scale. Traditionally, these models have required extensive labeled datasets for training. In this study, we propose a semi-supervised learning method that leverages vast amounts of unlabeled satellite images to detect slum settlements across diverse geographical regions and multiple time periods. Our approach can significantly reduce the dependency on manual data annotations. We showcase results based on four cities in sub-Saharan Africa and map slum settlements near a decade period between 2015 and 2024. The findings reveal localized settlement changes in slum areas at a resolution of 60 cm per pixel, which enables a more precise understanding of intra-city disparities. To support policymaking and humanitarian efforts, we make our data and code publicly available.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Deep learning applied to very high resolution (VHR) satellite imagery has become a powerful method for analyzing urban poverty at scale. Traditionally, these models have required extensive labeled datasets for training. In this study, we propose a semi-supervised learning method that leverages vast amounts of unlabeled satellite images to detect slum settlements across diverse geographical regions and multiple time periods. Our approach can significantly reduce the dependency on manual data annotations. We showcase results based on four cities in sub-Saharan Africa and map slum settlements near a decade period between 2015 and 2024. The findings reveal localized settlement changes in slum areas at a resolution of 60 cm per pixel, which enables a more precise understanding of intra-city disparities. To support policymaking and humanitarian efforts, we make our data and code publicly available.
Lee, S.; Kim, H.; Ahn, D.; Lee, D. G.; Kim, J.; Yun, H.; Jang, Chaemin; Jung, W.; Lee, Dongman
Tracking Economic Disparities in North Korea: Satellite Imagery Analysis Using AI with Policy Insights Proceedings Article
In: Proceedings of the 2025 International Conference on Information Technology for Social Good, pp. 155–163, Association for Computing Machinery, New York, NY, USA, 2025, ISBN: 9798400720895.
@inproceedings{Lee2025,
title = {Tracking Economic Disparities in North Korea: Satellite Imagery Analysis Using AI with Policy Insights},
author = {S. Lee and H. Kim and D. Ahn and D.G. Lee and J. Kim and H. Yun and Chaemin Jang and W. Jung and Dongman Lee},
url = {https://dl.acm.org/doi/10.1145/3748699.3749786},
doi = {10.1145/3748699.3749786},
isbn = {9798400720895},
year = {2025},
date = {2025-09-03},
urldate = {2025-09-03},
booktitle = {Proceedings of the 2025 International Conference on Information Technology for Social Good},
journal = {Proceedings of the 2025 International Conference on Information Technology for Social Good},
pages = {155–163},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
series = {GoodIT '25},
abstract = {here reliable economic data are scarce—as is often the case in authoritarian or otherwise data-limited settings—it becomes difficult to assess economic conditions or evaluate policy outcomes. To address this challenge, we test a human-AI framework that couples unsupervised clustering of satellite imagery with expert-guided ranking to create siScore, a grid-level indicator of economic development that requires no ground-truth data. Moving beyond earlier single-year studies to enable continuous monitoring, we assemble an annual panel for North Korea spanning 2018–2023 and demonstrate the framework’s capacity for sustained economic tracking. The series reveals (i) persistent urban-rural disparities: Pyongyang remains the most developed, whereas regions farther from the borders experienced deeper COVID-19 setbacks and slower recovery, and (ii) baseline conditions for assessing the regime’s new “20 x10 Regional Development Policy” Engagement with policymakers highlights a critical deployment barrier—temporal consistency—underscoring the need for robust preprocessing and multi-modal validation before the metric can inform practice. Overall, our work provides policymakers with a viable alternative for tracking economic dynamics where official statistics are unreliable.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
here reliable economic data are scarce—as is often the case in authoritarian or otherwise data-limited settings—it becomes difficult to assess economic conditions or evaluate policy outcomes. To address this challenge, we test a human-AI framework that couples unsupervised clustering of satellite imagery with expert-guided ranking to create siScore, a grid-level indicator of economic development that requires no ground-truth data. Moving beyond earlier single-year studies to enable continuous monitoring, we assemble an annual panel for North Korea spanning 2018–2023 and demonstrate the framework’s capacity for sustained economic tracking. The series reveals (i) persistent urban-rural disparities: Pyongyang remains the most developed, whereas regions farther from the borders experienced deeper COVID-19 setbacks and slower recovery, and (ii) baseline conditions for assessing the regime’s new “20 x10 Regional Development Policy” Engagement with policymakers highlights a critical deployment barrier—temporal consistency—underscoring the need for robust preprocessing and multi-modal validation before the metric can inform practice. Overall, our work provides policymakers with a viable alternative for tracking economic dynamics where official statistics are unreliable.
Lee, Hongju; Park, Youngjun; An, Jisun; Lee, Dongman
Enhancing Regional Airbnb Trend Forecasting Using LLM-Based Embeddings of Accessibility and Human Mobility Proceedings Article
In: International Conference on Advances in Social Networks Analysis and Mining (ASONAM 2025), pp. 376–390, Springer Nature Switzerland, 2025, ISBN: 978-3-032-13513-1.
@inproceedings{Lee2025b,
title = {Enhancing Regional Airbnb Trend Forecasting Using LLM-Based Embeddings of Accessibility and Human Mobility},
author = {Hongju Lee and Youngjun Park and Jisun An and Dongman Lee},
url = {https://link.springer.com/chapter/10.1007/978-3-032-13513-1_30#citeas},
doi = {https://doi.org/10.1007/978-3-032-13513-1_30},
isbn = {978-3-032-13513-1},
year = {2025},
date = {2025-08-27},
urldate = {2025-08-27},
booktitle = {International Conference on Advances in Social Networks Analysis and Mining (ASONAM 2025)},
volume = {16322},
pages = {376--390},
publisher = {Springer Nature Switzerland},
abstract = {The expansion of short-term rental platforms, such as Airbnb, has significantly disrupted local housing markets, often leading to increased rental prices and housing affordability issues. Accurately forecasting regional Airbnb market trends can thus offer critical insights for policymakers and urban planners aiming to mitigate these impacts. This study proposes a novel time-series forecasting framework to predict three key Airbnb indicators—Revenue, Reservation Days, and Number of Reservations—at the regional level. Using a sliding-window approach, the model forecasts trends 1 to 3 months ahead. Unlike prior studies that focus on individual listings at fixed time points, our approach constructs regional representations by integrating listing features with external contextual factors such as urban accessibility and human mobility. We convert structured tabular data into prompt-based inputs for a Large Language Model (LLM), producing comprehensive regional embeddings. These embeddings are then fed into advanced time-series models (RNN, LSTM, Transformer) to better capture complex spatio-temporal dynamics. Experiments on Seoul’s Airbnb dataset show that our method reduces both average RMSE and MAE by approximately 48% compared to conventional baselines, including traditional statistical and machine learning models. Our framework not only improves forecasting accuracy but also offers practical insights for detecting oversupplied regions and supporting data-driven urban policy decisions.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The expansion of short-term rental platforms, such as Airbnb, has significantly disrupted local housing markets, often leading to increased rental prices and housing affordability issues. Accurately forecasting regional Airbnb market trends can thus offer critical insights for policymakers and urban planners aiming to mitigate these impacts. This study proposes a novel time-series forecasting framework to predict three key Airbnb indicators—Revenue, Reservation Days, and Number of Reservations—at the regional level. Using a sliding-window approach, the model forecasts trends 1 to 3 months ahead. Unlike prior studies that focus on individual listings at fixed time points, our approach constructs regional representations by integrating listing features with external contextual factors such as urban accessibility and human mobility. We convert structured tabular data into prompt-based inputs for a Large Language Model (LLM), producing comprehensive regional embeddings. These embeddings are then fed into advanced time-series models (RNN, LSTM, Transformer) to better capture complex spatio-temporal dynamics. Experiments on Seoul’s Airbnb dataset show that our method reduces both average RMSE and MAE by approximately 48% compared to conventional baselines, including traditional statistical and machine learning models. Our framework not only improves forecasting accuracy but also offers practical insights for detecting oversupplied regions and supporting data-driven urban policy decisions.
Park, Sae Hyong; Shin, Seungjae; Ko, Namseok; Kim, Taeyeon; Lee, Dongman
GymProxy: A lightweight Python wrapper for seamless integration of standalone simulators into Gymnasium environments Journal Article
In: Computers and Electrical Engineering, vol. 127, pp. 110634, 2025, ISSN: 0045-7906.
@article{PARK2025110634,
title = {GymProxy: A lightweight Python wrapper for seamless integration of standalone simulators into Gymnasium environments},
author = {Sae Hyong Park and Seungjae Shin and Namseok Ko and Taeyeon Kim and Dongman Lee},
url = {https://www.sciencedirect.com/science/article/pii/S0045790625005774},
doi = {https://doi.org/10.1016/j.compeleceng.2025.110634},
issn = {0045-7906},
year = {2025},
date = {2025-08-23},
urldate = {2025-01-01},
journal = {Computers and Electrical Engineering},
volume = {127},
pages = {110634},
abstract = {Reinforcement learning (RL) often relies on extensive simulation-based training to refine agent behavior, as real-world errors can have significant repercussions. Integrating agents into existing standalone simulators typically requires code modifications and careful implementation of the synchronization mechanisms that ensure the agent and simulator execute in a turn-taking manner with proper data exchanges, which is often delicate and error-prone. In this paper, we introduce GymProxy, a lightweight and user-friendly open-source wrapper library designed to simplify the integration of standalone simulators into Gymnasium, the leading framework for developing RL environments. GymProxy provides simple and abstracted application programming interfaces (APIs) for seamlessly integrating an external simulator with the Gymnasium environment while handling synchronization internally. With GymProxy, developers can quickly build a custom Gymnasium environment that wraps existing standalone simulators with minimal implementation effort, eliminating the need to manually program complex synchronization logic.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Reinforcement learning (RL) often relies on extensive simulation-based training to refine agent behavior, as real-world errors can have significant repercussions. Integrating agents into existing standalone simulators typically requires code modifications and careful implementation of the synchronization mechanisms that ensure the agent and simulator execute in a turn-taking manner with proper data exchanges, which is often delicate and error-prone. In this paper, we introduce GymProxy, a lightweight and user-friendly open-source wrapper library designed to simplify the integration of standalone simulators into Gymnasium, the leading framework for developing RL environments. GymProxy provides simple and abstracted application programming interfaces (APIs) for seamlessly integrating an external simulator with the Gymnasium environment while handling synchronization internally. With GymProxy, developers can quickly build a custom Gymnasium environment that wraps existing standalone simulators with minimal implementation effort, eliminating the need to manually program complex synchronization logic.
Hong, Dasom; Han, Sumin; Park, Youngjun; An, Jisun; Lee, Dongman
Measuring capital with spatial media: How online popularity on Instagram shapes land value patterns in Seoul Journal Article
In: Area, pp. e70037, 2025, ISSN: 0004-0894.
@article{hong2025measuring,
title = {Measuring capital with spatial media: How online popularity on Instagram shapes land value patterns in Seoul},
author = {Dasom Hong and Sumin Han and Youngjun Park and Jisun An and Dongman Lee},
url = {https://rgs-ibg.onlinelibrary.wiley.com/doi/10.1111/area.70037},
issn = {0004-0894},
year = {2025},
date = {2025-08-06},
journal = {Area},
pages = {e70037},
abstract = {The recent proliferation of locative mobile devices and location-based services has sparked geographers' interest in how such ‘spatial media’ reshape individual's mobility, spatial experiences, and perceptions. Notably, the urban and digital geography literature has highlighted the relationship between spatial media, gentrification, and urban redevelopment processes, suggesting the potential of spatial media to generate new capitalist opportunities and increase the profitability of urban spaces. In this context, this study examines how spatial media reconfigure urban land value patterns, which have been primarily explained through urban forms. To this end, we employ the spatial capital model, a regression model that predicts property value patterns using variables associated with urban forms. We demonstrate that models incorporating both offline spatial configurations and online popularity outperform the spatial capital model relying solely on physical urban layouts, thereby underscoring the role of spatial media in reshaping the spatial arrangements of capital in urban centres. Using a machine learning algorithm, we construct and compare spatial capital models and regression models with variables reflecting online popularity on Instagram to estimate land prices in three neighbourhoods in Seoul: Yeonnam-dong, Seongsu-dong, and Gyeongridan-gil. These regions have experienced a surge in property values following heightened visibility on Instagram since the mid-2010s. Our findings indicate that the performance of models incorporating Instagram data exhibit superior predictive performance relative to traditional spatial capital models. Instagram-related variables also demonstrates greater explanatory power than conventional variables related to urban forms in predicting land price change rates. Additionally, the influence of online popularity on Instagram varies across time and space, closely aligned with the phase of development in online popularity within each area. In conclusion, spatial media have increasingly shaped urban land value patterns, while the dynamics between spatial media, urban forms, and property values can vary alongside the development of online popularity in specific regions.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The recent proliferation of locative mobile devices and location-based services has sparked geographers' interest in how such ‘spatial media’ reshape individual's mobility, spatial experiences, and perceptions. Notably, the urban and digital geography literature has highlighted the relationship between spatial media, gentrification, and urban redevelopment processes, suggesting the potential of spatial media to generate new capitalist opportunities and increase the profitability of urban spaces. In this context, this study examines how spatial media reconfigure urban land value patterns, which have been primarily explained through urban forms. To this end, we employ the spatial capital model, a regression model that predicts property value patterns using variables associated with urban forms. We demonstrate that models incorporating both offline spatial configurations and online popularity outperform the spatial capital model relying solely on physical urban layouts, thereby underscoring the role of spatial media in reshaping the spatial arrangements of capital in urban centres. Using a machine learning algorithm, we construct and compare spatial capital models and regression models with variables reflecting online popularity on Instagram to estimate land prices in three neighbourhoods in Seoul: Yeonnam-dong, Seongsu-dong, and Gyeongridan-gil. These regions have experienced a surge in property values following heightened visibility on Instagram since the mid-2010s. Our findings indicate that the performance of models incorporating Instagram data exhibit superior predictive performance relative to traditional spatial capital models. Instagram-related variables also demonstrates greater explanatory power than conventional variables related to urban forms in predicting land price change rates. Additionally, the influence of online popularity on Instagram varies across time and space, closely aligned with the phase of development in online popularity within each area. In conclusion, spatial media have increasingly shaped urban land value patterns, while the dynamics between spatial media, urban forms, and property values can vary alongside the development of online popularity in specific regions.
Park, Youngjun; Bae, Hyungchan; An, Jisun; Lee, Dongman
A Model for Monthly, Local-Level Airbnb Changes Using Public Dataset Proceedings Article
In: Proceedings of the 8th International Conference on Tourism Research, pp. 214-223, Academic Conferences International Ltd., 2025, ISSN: 2516-3604.
@inproceedings{park2025airbnb,
title = {A Model for Monthly, Local-Level Airbnb Changes Using Public Dataset},
author = {Youngjun Park and Hyungchan Bae and Jisun An and Dongman Lee},
url = {https://papers.academic-conferences.org/index.php/ictr},
issn = {2516-3604},
year = {2025},
date = {2025-04-28},
urldate = {2025-04-28},
booktitle = {Proceedings of the 8th International Conference on Tourism Research},
volume = {8},
issue = {1},
pages = {214-223},
publisher = {Academic Conferences International Ltd.},
abstract = {This study presents a methodological framework for estimating monthly variations in Airbnb listings across Seoul's administrative districts from 2017 to 2019, leveraging publicly available datasets. The proposed model integrates key factors such as housing supply, lodging unit density, proximity to tourist attractions, and retail sector activity, using negative binomial regression to quantify their impact on Airbnb dynamics. Our analysis demonstrates significant variations in these effects across different districts, with the model achieving superior statistical accuracy (average Pseudo $R^2$ of 0.7922) compared to baseline approaches. These findings underscore the utility of using accessible, timely data for capturing short-term rental market trends and highlight the model's potential applicability in diverse urban tourism settings. The research contributes valuable insights for tourism management, helping to develop policies that align with sustainable tourism growth and local economic development.
The results indicate that proximity to major tourist destinations and the number of foreign visitors are among the most influential factors in determining Airbnb activity, highlighting the strong connection between short-term rentals and tourism demand. Lodging unit density and retail composition also play significant roles in shaping the distribution of Airbnb listings, suggesting that tourism infrastructure and commercial activity are key determinants of short-term rental market expansion. The study finds that certain districts experience high Airbnb concentrations due to their tourism appeal, whereas others exhibit minimal short-term rental activity, reflecting the heterogeneous nature of urban tourism accommodation demand. Our monthly regression analysis captures dynamic fluctuations in Airbnb listings, addressing the limitations of traditional models that rely on annual data.
This research has critical implications for tourism stakeholders, including policymakers, destination managers, and hospitality businesses. By incorporating real-time, publicly available datasets, our approach facilitates responsive and data-driven decision-making to manage the short-term rental market effectively while fostering sustainable tourism growth. Future studies can extend this model to other urban contexts to explore how different socio-economic and regulatory environments influence short-term rental patterns. The adaptability of this methodology makes it a valuable tool for ongoing research in tourism analytics and policies.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This study presents a methodological framework for estimating monthly variations in Airbnb listings across Seoul's administrative districts from 2017 to 2019, leveraging publicly available datasets. The proposed model integrates key factors such as housing supply, lodging unit density, proximity to tourist attractions, and retail sector activity, using negative binomial regression to quantify their impact on Airbnb dynamics. Our analysis demonstrates significant variations in these effects across different districts, with the model achieving superior statistical accuracy (average Pseudo $R^2$ of 0.7922) compared to baseline approaches. These findings underscore the utility of using accessible, timely data for capturing short-term rental market trends and highlight the model's potential applicability in diverse urban tourism settings. The research contributes valuable insights for tourism management, helping to develop policies that align with sustainable tourism growth and local economic development.
The results indicate that proximity to major tourist destinations and the number of foreign visitors are among the most influential factors in determining Airbnb activity, highlighting the strong connection between short-term rentals and tourism demand. Lodging unit density and retail composition also play significant roles in shaping the distribution of Airbnb listings, suggesting that tourism infrastructure and commercial activity are key determinants of short-term rental market expansion. The study finds that certain districts experience high Airbnb concentrations due to their tourism appeal, whereas others exhibit minimal short-term rental activity, reflecting the heterogeneous nature of urban tourism accommodation demand. Our monthly regression analysis captures dynamic fluctuations in Airbnb listings, addressing the limitations of traditional models that rely on annual data.
This research has critical implications for tourism stakeholders, including policymakers, destination managers, and hospitality businesses. By incorporating real-time, publicly available datasets, our approach facilitates responsive and data-driven decision-making to manage the short-term rental market effectively while fostering sustainable tourism growth. Future studies can extend this model to other urban contexts to explore how different socio-economic and regulatory environments influence short-term rental patterns. The adaptability of this methodology makes it a valuable tool for ongoing research in tourism analytics and policies.
The results indicate that proximity to major tourist destinations and the number of foreign visitors are among the most influential factors in determining Airbnb activity, highlighting the strong connection between short-term rentals and tourism demand. Lodging unit density and retail composition also play significant roles in shaping the distribution of Airbnb listings, suggesting that tourism infrastructure and commercial activity are key determinants of short-term rental market expansion. The study finds that certain districts experience high Airbnb concentrations due to their tourism appeal, whereas others exhibit minimal short-term rental activity, reflecting the heterogeneous nature of urban tourism accommodation demand. Our monthly regression analysis captures dynamic fluctuations in Airbnb listings, addressing the limitations of traditional models that rely on annual data.
This research has critical implications for tourism stakeholders, including policymakers, destination managers, and hospitality businesses. By incorporating real-time, publicly available datasets, our approach facilitates responsive and data-driven decision-making to manage the short-term rental market effectively while fostering sustainable tourism growth. Future studies can extend this model to other urban contexts to explore how different socio-economic and regulatory environments influence short-term rental patterns. The adaptability of this methodology makes it a valuable tool for ongoing research in tourism analytics and policies.
Nguyen, Thanh-Tung; Liebe, Lucas; Tau, Nhat-Quang; Wu, Yuheng; Cheng, Jinghan; Lee, Dongman
OCTOPINF: Workload-Aware Inference Serving for Edge Video Analytics Proceedings Article
In: The 23rd International Conference on Pervasive Computing and Communications (PerCom 2025), Acceptance Rate 16%, IEEE 2025.
@inproceedings{nguyen2025octopinf,
title = {OCTOPINF: Workload-Aware Inference Serving for Edge Video Analytics},
author = {Thanh-Tung Nguyen and Lucas Liebe and Nhat-Quang Tau and Yuheng Wu and Jinghan Cheng and Dongman Lee},
url = {https://ieeexplore.ieee.org/document/11018743
https://github.com/tungngreen/PipelineScheduler},
doi = {10.1109/PerCom64205.2025.00032},
year = {2025},
date = {2025-03-17},
urldate = {2025-03-17},
booktitle = {The 23rd International Conference on Pervasive Computing and Communications (PerCom 2025), Acceptance Rate 16%},
organization = {IEEE},
abstract = {Edge Video Analytics (EVA) has become a major application of pervasive computing, enabling real-time visual processing. EVA pipelines, composed of deep neural networks (DNNs), typically demand efficient inference serving under stringent latency requirements, which is challenging due to the dynamic Edge environments (e.g., workload variability and network instability). Moreover, EVA pipelines face significant resource contention due to resource (e.g., GPU) constraints at the Edge. In this paper, we introduce OCTOPINF, a novel resource-efficient and workload-aware inference serving system designed for real-time EVA. OCTOPINF tackles the unique challenges of dynamic edge environments through fine-grained resource allocation, adaptive batching, and workload balancing between edge devices and servers. Furthermore, we propose a spatiotemporal scheduling algorithm that optimizes the co-location of inference tasks on GPUs, improving performance and ensuring service-level objectives (SLOs) compliance. Extensive evaluations on a real-world testbed demonstrate the effectiveness of our approach. It achieves an effective throughput increase of up to 10× compared to the baselines and shows better robustness in challenging scenarios. OCTOPINF can be used for any DNN-based EVA inference task with minimal adaptation and is available at https://github.com/tungngreen/PipelineScheduler.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Edge Video Analytics (EVA) has become a major application of pervasive computing, enabling real-time visual processing. EVA pipelines, composed of deep neural networks (DNNs), typically demand efficient inference serving under stringent latency requirements, which is challenging due to the dynamic Edge environments (e.g., workload variability and network instability). Moreover, EVA pipelines face significant resource contention due to resource (e.g., GPU) constraints at the Edge. In this paper, we introduce OCTOPINF, a novel resource-efficient and workload-aware inference serving system designed for real-time EVA. OCTOPINF tackles the unique challenges of dynamic edge environments through fine-grained resource allocation, adaptive batching, and workload balancing between edge devices and servers. Furthermore, we propose a spatiotemporal scheduling algorithm that optimizes the co-location of inference tasks on GPUs, improving performance and ensuring service-level objectives (SLOs) compliance. Extensive evaluations on a real-world testbed demonstrate the effectiveness of our approach. It achieves an effective throughput increase of up to 10× compared to the baselines and shows better robustness in challenging scenarios. OCTOPINF can be used for any DNN-based EVA inference task with minimal adaptation and is available at https://github.com/tungngreen/PipelineScheduler.
2024
Kim, Youngrae; Cho, Younggeol; Nguyen, Thanh-Tung; Hong, Seunghoon; Lee, Dongman
MetaWeather: Few-Shot Weather-Degraded Image Restoration Proceedings Article
In: European Conference on Computer Vision, pp. 206–222, Springer 2024, ISBN: 978-3-031-73464-9.
@inproceedings{kim2024metaweather,
title = {MetaWeather: Few-Shot Weather-Degraded Image Restoration},
author = {Youngrae Kim and Younggeol Cho and Thanh-Tung Nguyen and Seunghoon Hong and Dongman Lee},
url = {https://link.springer.com/chapter/10.1007/978-3-031-73464-9_13},
doi = {10.1007/978-3-031-73464-9_13},
isbn = {978-3-031-73464-9},
year = {2024},
date = {2024-12-04},
urldate = {2024-01-01},
booktitle = {European Conference on Computer Vision},
pages = {206–222},
organization = {Springer},
abstract = {Real-world weather conditions are intricate and often occur concurrently. However, most existing restoration approaches are limited in their applicability to specific weather conditions in training data and struggle to generalize to unseen weather types, including real-world weather conditions. To address this issue, we introduce MetaWeather, a universal approach that can handle diverse and novel weather conditions with a single unified model. Extending a powerful meta-learning framework, MetaWeather formulates the task of weather-degraded image restoration as a few-shot adaptation problem that predicts the degradation pattern of a query image, and learns to adapt to unseen weather conditions through a novel spatial-channel matching algorithm. Experimental results on the BID Task II.A, SPA-Data, and RealSnow datasets demonstrate that the proposed method can adapt to unseen weather conditions, significantly outperforming the state-of-the-art multi-weather image restoration methods. Code is available at https://github.com/RangeWING/MetaWeather.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Real-world weather conditions are intricate and often occur concurrently. However, most existing restoration approaches are limited in their applicability to specific weather conditions in training data and struggle to generalize to unseen weather types, including real-world weather conditions. To address this issue, we introduce MetaWeather, a universal approach that can handle diverse and novel weather conditions with a single unified model. Extending a powerful meta-learning framework, MetaWeather formulates the task of weather-degraded image restoration as a few-shot adaptation problem that predicts the degradation pattern of a query image, and learns to adapt to unseen weather conditions through a novel spatial-channel matching algorithm. Experimental results on the BID Task II.A, SPA-Data, and RealSnow datasets demonstrate that the proposed method can adapt to unseen weather conditions, significantly outperforming the state-of-the-art multi-weather image restoration methods. Code is available at https://github.com/RangeWING/MetaWeather.
Cho, Younggeol; Kim, Youngrae; Yoon, Junho; Hong, Seunghoon; Lee, Dongman
Feature Augmentation based Test-Time Adaptation Journal Article
In: arXiv preprint arXiv:2410.14178, 2024.
@article{cho2024feature,
title = {Feature Augmentation based Test-Time Adaptation},
author = {Younggeol Cho and Youngrae Kim and Junho Yoon and Seunghoon Hong and Dongman Lee},
url = {https://arxiv.org/abs/2410.14178},
doi = {10.48550/arXiv.2410.14178},
year = {2024},
date = {2024-10-18},
urldate = {2024-10-18},
journal = {arXiv preprint arXiv:2410.14178},
abstract = {Test-time adaptation (TTA) allows a model to be adapted to an unseen domain without accessing the source data. Due to the nature of practical environments, TTA has a limited amount of data for adaptation. Recent TTA methods further restrict this by filtering input data for reliability, making the effective data size even smaller and limiting adaptation potential. To address this issue, We propose Feature Augmentation based Test-time Adaptation (FATA), a simple method that fully utilizes the limited amount of input data through feature augmentation. FATA employs Normalization Perturbation to augment features and adapts the model using the FATA loss, which makes the outputs of the augmented and original features similar. FATA is model-agnostic and can be seamlessly integrated into existing models without altering the model architecture. We demonstrate the effectiveness of FATA on various models and scenarios on ImageNet-C and Office-Home, validating its superiority in diverse real-world conditions.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Test-time adaptation (TTA) allows a model to be adapted to an unseen domain without accessing the source data. Due to the nature of practical environments, TTA has a limited amount of data for adaptation. Recent TTA methods further restrict this by filtering input data for reliability, making the effective data size even smaller and limiting adaptation potential. To address this issue, We propose Feature Augmentation based Test-time Adaptation (FATA), a simple method that fully utilizes the limited amount of input data through feature augmentation. FATA employs Normalization Perturbation to augment features and adapts the model using the FATA loss, which makes the outputs of the augmented and original features similar. FATA is model-agnostic and can be seamlessly integrated into existing models without altering the model architecture. We demonstrate the effectiveness of FATA on various models and scenarios on ImageNet-C and Office-Home, validating its superiority in diverse real-world conditions.
Kim, Hyunju; Son, Heesuk; Lee, Dongman
Causality-Aware Pattern Mining Scheme for Group Activity Recognition Proceedings Article
In: 2024 IEEE 48th Annual Computers, Software, and Applications Conference (COMPSAC), pp. 908–917, IEEE 2024, ISSN: 2836-3795.
@inproceedings{kim2024causality,
title = {Causality-Aware Pattern Mining Scheme for Group Activity Recognition},
author = {Hyunju Kim and Heesuk Son and Dongman Lee},
url = {https://ieeexplore.ieee.org/abstract/document/10633339},
doi = {10.1109/COMPSAC61105.2024.00126},
issn = {2836-3795},
year = {2024},
date = {2024-07-02},
urldate = {2024-01-01},
booktitle = {2024 IEEE 48th Annual Computers, Software, and Applications Conference (COMPSAC)},
pages = {908–917},
organization = {IEEE},
abstract = {Human activity recognition is a key challenge in pervasive computing and its solutions have been presented based on various disciplines. Specifically, for human activity recognition in smart spaces without privacy and accessibility issues, data streams generated by deployed ambient sensors are leveraged. In this paper, we focus on group activities by which a group of users perform a collaborative task without user identification and propose an efficient group activity recognition scheme that extracts causality patterns from ambient sensor event sequences, to support as good recognition accuracy as the state-of-the-art models with missing or false data tolerance. To filter out irrelevant noise events from a given data stream, a set of rules is leveraged to highlight causally related events. Then, a pattern-tree algorithm extracts frequent causal patterns by means of a growing tree structure. Based on the extracted patterns, a weighted sum-based pattern-matching algorithm computes the likelihood of stored group activities to the given test event sequence using matched event pattern counts for group activity recognition. We evaluate the proposed scheme using the data collected from real-world testbed and open datasets where users perform their tasks on a daily basis. Experiment results show that the proposed scheme performs higher recognition accuracy and is tolerant to missing or false data with a smaller amount of runtime overhead than the existing schemes.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Human activity recognition is a key challenge in pervasive computing and its solutions have been presented based on various disciplines. Specifically, for human activity recognition in smart spaces without privacy and accessibility issues, data streams generated by deployed ambient sensors are leveraged. In this paper, we focus on group activities by which a group of users perform a collaborative task without user identification and propose an efficient group activity recognition scheme that extracts causality patterns from ambient sensor event sequences, to support as good recognition accuracy as the state-of-the-art models with missing or false data tolerance. To filter out irrelevant noise events from a given data stream, a set of rules is leveraged to highlight causally related events. Then, a pattern-tree algorithm extracts frequent causal patterns by means of a growing tree structure. Based on the extracted patterns, a weighted sum-based pattern-matching algorithm computes the likelihood of stored group activities to the given test event sequence using matched event pattern counts for group activity recognition. We evaluate the proposed scheme using the data collected from real-world testbed and open datasets where users perform their tasks on a daily basis. Experiment results show that the proposed scheme performs higher recognition accuracy and is tolerant to missing or false data with a smaller amount of runtime overhead than the existing schemes.
Park, Youngjun; Han, Sumin; Park, Doyoon; Bae, Hyeongchan; Thanh-Tung, Nguyen; Lee, Dongman
Integrative Geospatial Visualization for Urban Mobility Analysis Journal Article
In: 한국정보과학회 학술발표논문집, pp. 402–404, 2024.
@article{park2024integrative,
title = {Integrative Geospatial Visualization for Urban Mobility Analysis},
author = {Youngjun Park and Sumin Han and Doyoon Park and Hyeongchan Bae and Nguyen Thanh-Tung and Dongman Lee},
url = {https://www.dbpia.co.kr/pdf/pdfView.do?nodeId=NODE11861859},
year = {2024},
date = {2024-06-26},
urldate = {2024-06-26},
journal = {한국정보과학회 학술발표논문집},
pages = {402–404},
abstract = {This study introduces a visualization tool that integrates dynamic urban mobility data with static geographic and demographic datasets to enhance urban analysis. Employing GeoHash-based techniques, the tool facilitates the exploration and understanding of complex urban patterns in a multifaceted way. It leverages real-time and historical data from sources such as the New York Taxi Origin-Destination dataset and US Census information, enabling users to visualize and interpret urban mobility with unprecedented granularity and precision. By providing a dynamic, interactive platform, the tool not only supports urban planners in making informed decisions but also contributes to the broader field of smart city initiatives, fostering more sustainable and efficient urban environments. This paper details the tool’s development, showcases its practical applications through use cases, and discusses its potential for future integration with advanced analytical models. },
keywords = {},
pubstate = {published},
tppubtype = {article}
}
This study introduces a visualization tool that integrates dynamic urban mobility data with static geographic and demographic datasets to enhance urban analysis. Employing GeoHash-based techniques, the tool facilitates the exploration and understanding of complex urban patterns in a multifaceted way. It leverages real-time and historical data from sources such as the New York Taxi Origin-Destination dataset and US Census information, enabling users to visualize and interpret urban mobility with unprecedented granularity and precision. By providing a dynamic, interactive platform, the tool not only supports urban planners in making informed decisions but also contributes to the broader field of smart city initiatives, fostering more sustainable and efficient urban environments. This paper details the tool’s development, showcases its practical applications through use cases, and discusses its potential for future integration with advanced analytical models.
Elmoursi, Ahmad; Han, Sumin; Bae, Hyungchan; Park, Youngjun; Thanh-Tung, Nguyen; Lee, Dongman
Enhancing Geospatial Recommendations through Natural Language Processing with Large Language Models Journal Article
In: 한국정보과학회 학술발표논문집, pp. 2215–2217, 2024.
@article{elmoursi2024enhancing,
title = {Enhancing Geospatial Recommendations through Natural Language Processing with Large Language Models},
author = {Ahmad Elmoursi and Sumin Han and Hyungchan Bae and Youngjun Park and Nguyen Thanh-Tung and Dongman Lee},
url = {https://www.dbpia.co.kr/pdf/pdfView.do?nodeId=NODE11862488},
year = {2024},
date = {2024-06-26},
urldate = {2024-06-26},
journal = {한국정보과학회 학술발표논문집},
pages = {2215–2217},
abstract = {With the success of large language models (LLMs) such as OpenAI’s ChatGPT, Meta’s LLama, and Google’s Gemmini,there is a growing trend to leverage natural language queries for database processing. For example, Chroma DB storesdocuments in an embedded representation, allowing it to find data that closely matches the embeddings of user queries.However, the application of this technology to geospatial data, such as places of interest on a map, remains underexplored.This paper presents an exemplary application of place recommendation by utilizing the user’s location to crawl and filter afew recommended results, thereby finding the most relevant place based on the user’s natural-language-based query. Thisapproach offers new insights for enhancing location-based services, providing a framework for more contextually relevantand personalized user experiences.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
With the success of large language models (LLMs) such as OpenAI’s ChatGPT, Meta’s LLama, and Google’s Gemmini,there is a growing trend to leverage natural language queries for database processing. For example, Chroma DB storesdocuments in an embedded representation, allowing it to find data that closely matches the embeddings of user queries.However, the application of this technology to geospatial data, such as places of interest on a map, remains underexplored.This paper presents an exemplary application of place recommendation by utilizing the user’s location to crawl and filter afew recommended results, thereby finding the most relevant place based on the user’s natural-language-based query. Thisapproach offers new insights for enhancing location-based services, providing a framework for more contextually relevantand personalized user experiences.
Kim, Hyunju; Kim, Geon; Jain, Akanksha; Lee, Dongman
Mobile Robot based Personalized Thermal Comfort Scheme Proceedings Article
In: 2024 21st International Conference on Ubiquitous Robots (UR), pp. 84–91, IEEE 2024, ISBN: 979-8-3503-6107-0.
@inproceedings{kim2024mobile,
title = {Mobile Robot based Personalized Thermal Comfort Scheme},
author = {Hyunju Kim and Geon Kim and Akanksha Jain and Dongman Lee},
url = {https://ieeexplore.ieee.org/abstract/document/10597472},
doi = {10.1109/UR61395.2024.10597472},
isbn = {979-8-3503-6107-0},
year = {2024},
date = {2024-06-24},
urldate = {2024-06-24},
booktitle = {2024 21st International Conference on Ubiquitous Robots (UR)},
pages = {84–91},
organization = {IEEE},
abstract = {Indoor thermal control is a crucial technique for ensuring user comfort and efficient energy usage, and typically relies on conventional methods that standardize the indoor thermal environment, neglecting individual personalized preferences. Mobile robots have emerged as a potential solution for personalizing temperature comfort. However, the existing research often falls short in considering factors like robot movement control, user activities, human states, and potential disturbance to users, leading to inaccurate estimations of a user's temperature adjustment needs. This paper introduces a mobile robot-based personalized thermal control system, designed to enhance the accuracy in recognizing human states relevant to thermal comfort in real indoor environments. This system achieves accurate thermal comfort estimation using vision-based recognition while reducing robot movement to decrease user inconvenience. The robot dynamically navigates to optimal positions, guided by the confidence level in vision-based human state recognition. We train the robot's movement control policy using a Deep Reinforcement Learning-based model. Real-world evaluation shows the system's success in accurately recognizing human states with minimal movement trajectories and reduced user discomfort. The proposed robot-based approach offers a significant advancement in personalized thermal control, allowing for more accurate thermal comfort estimation.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Indoor thermal control is a crucial technique for ensuring user comfort and efficient energy usage, and typically relies on conventional methods that standardize the indoor thermal environment, neglecting individual personalized preferences. Mobile robots have emerged as a potential solution for personalizing temperature comfort. However, the existing research often falls short in considering factors like robot movement control, user activities, human states, and potential disturbance to users, leading to inaccurate estimations of a user's temperature adjustment needs. This paper introduces a mobile robot-based personalized thermal control system, designed to enhance the accuracy in recognizing human states relevant to thermal comfort in real indoor environments. This system achieves accurate thermal comfort estimation using vision-based recognition while reducing robot movement to decrease user inconvenience. The robot dynamically navigates to optimal positions, guided by the confidence level in vision-based human state recognition. We train the robot's movement control policy using a Deep Reinforcement Learning-based model. Real-world evaluation shows the system's success in accurately recognizing human states with minimal movement trajectories and reduced user discomfort. The proposed robot-based approach offers a significant advancement in personalized thermal control, allowing for more accurate thermal comfort estimation.
Kim, Hyunju; Kim, Geon; Lee, Taehoon; Kim, Kisoo; Lee, Dongman
A dataset of ambient sensors in a meeting room for activity recognition Journal Article
In: Scientific Data, vol. 11, no. 1, pp. 516, 2024, ISSN: 2052-4463.
@article{kim2024dataset,
title = {A dataset of ambient sensors in a meeting room for activity recognition},
author = {Hyunju Kim and Geon Kim and Taehoon Lee and Kisoo Kim and Dongman Lee},
url = {https://www.nature.com/articles/s41597-024-03344-7},
doi = {10.1038/s41597-024-03344-7},
issn = {2052-4463},
year = {2024},
date = {2024-05-21},
urldate = {2024-01-01},
journal = {Scientific Data},
volume = {11},
number = {1},
pages = {516},
publisher = {Nature Publishing Group UK London},
abstract = {As IoT technology advances, using machine learning to detect user activities emerges as a promising strategy for delivering a variety of smart services. It is essential to have access to high-quality data that also respects privacy concerns and data streams from ambient sensors in the surrounding environment meet this requirement. However, despite growing interest in research, there is a noticeable lack of datasets from ambient sensors designed for public spaces, as opposed to those for private settings. To bridge this gap, we design the DOO-RE dataset within an actual meeting room environment, equipped with three types of ambient sensors: those triggered by actuators, users, and the environment itself. This dataset is compiled from the activities of over twenty students throughout a period of four months. DOO-RE provides reliable and purpose-oriented activity data in a public setting, with activity labels verified by multiple annotators through a process of cross-validation to guarantee data integrity. DOO-RE categorizes nine different types of activities and facilitates the study of both single and group activities. We are optimistic that DOO-RE will play a significant role in advancing human activity recognition technologies, enhancing smart automation systems, and enabling the rapid setup of smart spaces through ambient sensors.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
As IoT technology advances, using machine learning to detect user activities emerges as a promising strategy for delivering a variety of smart services. It is essential to have access to high-quality data that also respects privacy concerns and data streams from ambient sensors in the surrounding environment meet this requirement. However, despite growing interest in research, there is a noticeable lack of datasets from ambient sensors designed for public spaces, as opposed to those for private settings. To bridge this gap, we design the DOO-RE dataset within an actual meeting room environment, equipped with three types of ambient sensors: those triggered by actuators, users, and the environment itself. This dataset is compiled from the activities of over twenty students throughout a period of four months. DOO-RE provides reliable and purpose-oriented activity data in a public setting, with activity labels verified by multiple annotators through a process of cross-validation to guarantee data integrity. DOO-RE categorizes nine different types of activities and facilitates the study of both single and group activities. We are optimistic that DOO-RE will play a significant role in advancing human activity recognition technologies, enhancing smart automation systems, and enabling the rapid setup of smart spaces through ambient sensors.
2023
Cho, Younggeol; Kim, Youngrae; Lee, Dongman
Beyond Entropy: Style Transfer Guided Single Image Continual Test-Time Adaptation Journal Article
In: arXiv preprint arXiv:2311.18270, 2023.
@article{cho2023beyond,
title = {Beyond Entropy: Style Transfer Guided Single Image Continual Test-Time Adaptation},
author = {Younggeol Cho and Youngrae Kim and Dongman Lee},
url = {https://arxiv.org/abs/2311.18270},
doi = {10.48550/arXiv.2311.18270},
year = {2023},
date = {2023-11-30},
urldate = {2023-01-01},
journal = {arXiv preprint arXiv:2311.18270},
abstract = {Continual test-time adaptation (cTTA) methods are designed to facilitate the continual adaptation of models to dynamically changing real-world environments where computational resources are limited. Due to this inherent limitation, existing approaches fail to simultaneously achieve accuracy and efficiency. In detail, when using a single image, the instability caused by batch normalization layers and entropy loss significantly destabilizes many existing methods in real-world cTTA scenarios. To overcome these challenges, we present BESTTA, a novel single image continual test-time adaptation method guided by style transfer, which enables stable and efficient adaptation to the target environment by transferring the style of the input image to the source style. To implement the proposed method, we devise BeIN, a simple yet powerful normalization method, along with the style-guided losses. We demonstrate that BESTTA effectively adapts to the continually changing target environment, leveraging only a single image on both semantic segmentation and image classification tasks. Remarkably, despite training only two parameters in a BeIN layer consuming the least memory, BESTTA outperforms existing state-of-the-art methods in terms of performance.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Continual test-time adaptation (cTTA) methods are designed to facilitate the continual adaptation of models to dynamically changing real-world environments where computational resources are limited. Due to this inherent limitation, existing approaches fail to simultaneously achieve accuracy and efficiency. In detail, when using a single image, the instability caused by batch normalization layers and entropy loss significantly destabilizes many existing methods in real-world cTTA scenarios. To overcome these challenges, we present BESTTA, a novel single image continual test-time adaptation method guided by style transfer, which enables stable and efficient adaptation to the target environment by transferring the style of the input image to the source style. To implement the proposed method, we devise BeIN, a simple yet powerful normalization method, along with the style-guided losses. We demonstrate that BESTTA effectively adapts to the continually changing target environment, leveraging only a single image on both semantic segmentation and image classification tasks. Remarkably, despite training only two parameters in a BeIN layer consuming the least memory, BESTTA outperforms existing state-of-the-art methods in terms of performance.
Han, Sumin; Park, Youngjun; Lee, Minji; An, Jisun; Lee, Dongman
Enhancing Spatio-temporal Traffic Prediction through Urban Human Activity Analysis Proceedings Article
In: Proceedings of the 32nd ACM International Conference on Information and Knowledge Management, pp. 689–698, Association for Computing Machinery, New York, NY, USA, 2023, ISBN: 9798400701245.
@inproceedings{han2023enhancing,
title = {Enhancing Spatio-temporal Traffic Prediction through Urban Human Activity Analysis},
author = {Sumin Han and Youngjun Park and Minji Lee and Jisun An and Dongman Lee},
url = {https://dl.acm.org/doi/abs/10.1145/3583780.3614867},
doi = {10.1145/3583780.3614867},
isbn = {9798400701245},
year = {2023},
date = {2023-10-21},
urldate = {2023-01-01},
booktitle = {Proceedings of the 32nd ACM International Conference on Information and Knowledge Management},
pages = {689–698},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
series = {CIKM '23},
abstract = {Traffic prediction is one of the key elements to ensure the safety and convenience of citizens. Existing traffic prediction models primarily focus on deep learning architectures to capture spatial and temporal correlation. They often overlook the underlying nature of traffic. Specifically, the sensor networks in most traffic datasets do not accurately represent the actual road network exploited by vehicles, failing to provide insights into the traffic patterns in urban activities. To overcome these limitations, we propose an improved traffic prediction method based on graph convolution deep learning algorithms. We leverage human activity frequency data from National Household Travel Survey to enhance the inference capability of a causal relationship between activity and traffic patterns. Despite making minimal modifications to the conventional graph convolutional recurrent networks and graph convolutional transformer architectures, our approach achieves state-of-the-art performance without introducing excessive computational overhead.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Traffic prediction is one of the key elements to ensure the safety and convenience of citizens. Existing traffic prediction models primarily focus on deep learning architectures to capture spatial and temporal correlation. They often overlook the underlying nature of traffic. Specifically, the sensor networks in most traffic datasets do not accurately represent the actual road network exploited by vehicles, failing to provide insights into the traffic patterns in urban activities. To overcome these limitations, we propose an improved traffic prediction method based on graph convolution deep learning algorithms. We leverage human activity frequency data from National Household Travel Survey to enhance the inference capability of a causal relationship between activity and traffic patterns. Despite making minimal modifications to the conventional graph convolutional recurrent networks and graph convolutional transformer architectures, our approach achieves state-of-the-art performance without introducing excessive computational overhead.
Kim, Geon; Kim, Hyunju; Kim, Youngjae; Lee, Dongman
Exploring Technical Challenges for Thermal Sensation Estimation in Mobile Robot Driven Personalized Thermal Control Proceedings Article
In: 2023 20th International Conference on Ubiquitous Robots (UR), pp. 997–1002, IEEE 2023, ISBN: 979-8-3503-3517-0.
@inproceedings{kim2023exploring,
title = {Exploring Technical Challenges for Thermal Sensation Estimation in Mobile Robot Driven Personalized Thermal Control},
author = {Geon Kim and Hyunju Kim and Youngjae Kim and Dongman Lee},
url = {https://ieeexplore.ieee.org/abstract/document/10202444},
doi = {10.1109/UR57808.2023.10202444},
isbn = {979-8-3503-3517-0},
year = {2023},
date = {2023-06-25},
urldate = {2023-01-01},
booktitle = {2023 20th International Conference on Ubiquitous Robots (UR)},
pages = {997–1002},
organization = {IEEE},
abstract = {The personalized thermal control studies estimate users' thermal sensations based on wearable or vision-based approaches to control the temperature, humidity, and ventilation in response to individuals' different thermal preferences. However, there is a limitation that both conventional wearable-based and vision-based methods are practical for real-world deployment due to additional equipment, and fixed field of view. This paper discusses the mobile robot-driven human state estimation for personalized thermal control as one of the solutions to overcome these limitations. We investigate the technical issue and its possible solution in designing mobile robot-driven estimation and performed a simulation-based preliminary evaluation to evaluate the effectiveness of a mobile robot on human state estimation. Evaluation results show that the mobile robot can help in the accurate estimation of the user's thermal sensation with only vision data. Finally, we present future directions to further increase the accuracy to achieve wearable-level.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The personalized thermal control studies estimate users' thermal sensations based on wearable or vision-based approaches to control the temperature, humidity, and ventilation in response to individuals' different thermal preferences. However, there is a limitation that both conventional wearable-based and vision-based methods are practical for real-world deployment due to additional equipment, and fixed field of view. This paper discusses the mobile robot-driven human state estimation for personalized thermal control as one of the solutions to overcome these limitations. We investigate the technical issue and its possible solution in designing mobile robot-driven estimation and performed a simulation-based preliminary evaluation to evaluate the effectiveness of a mobile robot on human state estimation. Evaluation results show that the mobile robot can help in the accurate estimation of the user's thermal sensation with only vision data. Finally, we present future directions to further increase the accuracy to achieve wearable-level.
Lee, Tae-Hoon; Kim, Hyunju; Lee, Dongman
Transformer Based Early Classification for Real-Time Human Activity Recognition in Smart Homes Proceedings Article
In: Proceedings of the 38th ACM/SIGAPP Symposium on Applied Computing, pp. 410–417, Association for Computing Machinery, New York, NY, USA, 2023, ISBN: 9781450395175.
@inproceedings{lee2023transformer,
title = {Transformer Based Early Classification for Real-Time Human Activity Recognition in Smart Homes},
author = {Tae-Hoon Lee and Hyunju Kim and Dongman Lee},
url = {https://dl.acm.org/doi/abs/10.1145/3555776.3577693},
doi = {10.1145/3555776.3577693},
isbn = {9781450395175},
year = {2023},
date = {2023-06-07},
urldate = {2023-01-01},
booktitle = {Proceedings of the 38th ACM/SIGAPP Symposium on Applied Computing},
pages = {410–417},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
series = {SAC '23},
abstract = {Human activity recognition (HAR) plays a key role in intelligent systems. Ambient sensors are utilized to avoid privacy concerns and to collect data streams in a less intrusive manner in smart homes. In scenarios requiring immediate intervention, the systems must perform HAR in real-time. Since it is difficult to segment the exact transition point in real-time, data unrelated to the target activity can appear at the beginning of the time series, which we call unrefined data. It leads us to a new challenge that the HAR model recognizes a user's activity as early as possible with unrefined data. To explore the impact of unrefined data on real-time HAR, we design an experimental system that consists of a Transformer-based filtering network and an LSTM-based early classifier. We evaluate the experimental system with 3 public datasets collected on testbeds with ambient sensors installed. Our results reveal that unrefined data degrade HAR performance in terms of accuracy and earliness, and the use of the filtering network that filters out unrefined data improves recognition performance.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Human activity recognition (HAR) plays a key role in intelligent systems. Ambient sensors are utilized to avoid privacy concerns and to collect data streams in a less intrusive manner in smart homes. In scenarios requiring immediate intervention, the systems must perform HAR in real-time. Since it is difficult to segment the exact transition point in real-time, data unrelated to the target activity can appear at the beginning of the time series, which we call unrefined data. It leads us to a new challenge that the HAR model recognizes a user's activity as early as possible with unrefined data. To explore the impact of unrefined data on real-time HAR, we design an experimental system that consists of a Transformer-based filtering network and an LSTM-based early classifier. We evaluate the experimental system with 3 public datasets collected on testbeds with ambient sensors installed. Our results reveal that unrefined data degrade HAR performance in terms of accuracy and earliness, and the use of the filtering network that filters out unrefined data improves recognition performance.
Kim, Geon; Kim, Hyunju; Lee, Dongman
Towards deployment of mobile robot driven preference learning for user-state-specific thermal control in a real-world smart space Proceedings Article
In: Proceedings of the 38th ACM/SIGAPP Symposium on Applied Computing, pp. 724–731, Association for Computing Machinery, New York, NY, USA, 2023, ISBN: 9781450395175.
@inproceedings{kim2023towards,
title = {Towards deployment of mobile robot driven preference learning for user-state-specific thermal control in a real-world smart space},
author = {Geon Kim and Hyunju Kim and Dongman Lee},
url = {https://dl.acm.org/doi/abs/10.1145/3555776.3577760},
doi = {10.1145/3555776.3577760},
isbn = {9781450395175},
year = {2023},
date = {2023-06-07},
urldate = {2023-01-01},
booktitle = {Proceedings of the 38th ACM/SIGAPP Symposium on Applied Computing},
pages = {724–731},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
series = {SAC '23},
abstract = {Indoor Environment Quality (IEQ) is one of the most important goals for smart spaces. Thermal comfort is typically considered the most emphasized factor in IEQ that depends on personalized thermal preference. In this paper, we explore technical challenges to deploying a robot-driven personalized thermal control system that uses a mobile robot for learning user-state-specific preference efficiently. We conduct a few experiments that give a clue to overcome such challenges (i.e. low image recognition) when the system is deployed in a real world. We present future directions to improve robot-driven preference learning from the exploration.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Indoor Environment Quality (IEQ) is one of the most important goals for smart spaces. Thermal comfort is typically considered the most emphasized factor in IEQ that depends on personalized thermal preference. In this paper, we explore technical challenges to deploying a robot-driven personalized thermal control system that uses a mobile robot for learning user-state-specific preference efficiently. We conduct a few experiments that give a clue to overcome such challenges (i.e. low image recognition) when the system is deployed in a real world. We present future directions to improve robot-driven preference learning from the exploration.
Nguyen, Thanh-Tung; Jang, Si Young; Kostadinov, Boyan; Lee, Dongman
PreActo: Efficient Cross-Camera Object Tracking System in Video Analytics Edge Computing Proceedings Article
In: 2023 IEEE International Conference on Pervasive Computing and Communications (PerCom), Acceptance rate 17%, pp. 101–110, IEEE 2023, ISSN: 2474-249X.
@inproceedings{nguyen2023preacto,
title = {PreActo: Efficient Cross-Camera Object Tracking System in Video Analytics Edge Computing},
author = {Thanh-Tung Nguyen and Si Young Jang and Boyan Kostadinov and Dongman Lee},
url = {https://ieeexplore.ieee.org/document/10099298},
doi = {10.1109/PERCOM56429.2023.10099298},
issn = {2474-249X},
year = {2023},
date = {2023-03-13},
urldate = {2023-03-13},
booktitle = {2023 IEEE International Conference on Pervasive Computing and Communications (PerCom), Acceptance rate 17%},
pages = {101--110},
organization = {IEEE},
abstract = {Cross-camera real-time object tracking is one of the important, yet challenging applications of video analytics in edge computing environments. To provide accurate and efficient real-time tracking, a tracking target's future movements need to be predicted. Particularly, the destination camera and travel time of the target object are to be identified so that tracking duties can be handover-ed seamlessly. In this paper, we propose a collaborative cross-camera tracking system, called PreActo, with two key features: (1) ResNet-based trajectory learning to exploit the rich spatio-temporal information embedded within objects' moving patterns, which has not been utilized by the existing literature, and (2) collaboration between the edge server and the edge device for real-time trajectory prediction and tracking handover. To prove the validity of our proposed system, we evaluate PreActo on a video dataset leveraging real-world trajectories. Evaluation results show that the proposed system reduces up to 7× the number of processed frames for handover, with 2× lower latency while providing 1.5× tracking precision improvement compared to the state-of-the-art.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Cross-camera real-time object tracking is one of the important, yet challenging applications of video analytics in edge computing environments. To provide accurate and efficient real-time tracking, a tracking target's future movements need to be predicted. Particularly, the destination camera and travel time of the target object are to be identified so that tracking duties can be handover-ed seamlessly. In this paper, we propose a collaborative cross-camera tracking system, called PreActo, with two key features: (1) ResNet-based trajectory learning to exploit the rich spatio-temporal information embedded within objects' moving patterns, which has not been utilized by the existing literature, and (2) collaboration between the edge server and the edge device for real-time trajectory prediction and tracking handover. To prove the validity of our proposed system, we evaluate PreActo on a video dataset leveraging real-world trajectories. Evaluation results show that the proposed system reduces up to 7× the number of processed frames for handover, with 2× lower latency while providing 1.5× tracking precision improvement compared to the state-of-the-art.
Kim, Youngrae; Lim, Jinsu; Cho, Hoonhee; Lee, Minji; Lee, Dongman; Yoon, Kuk-Jin; Choi, Ho-Jin
Efficient reference-based video super-resolution (ervsr): Single reference image is all you need Proceedings Article
In: 2023 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pp. 1828–1837, IEEE 2023, ISSN: 2642-9381.
@inproceedings{kim2023efficient,
title = {Efficient reference-based video super-resolution (ervsr): Single reference image is all you need},
author = {Youngrae Kim and Jinsu Lim and Hoonhee Cho and Minji Lee and Dongman Lee and Kuk-Jin Yoon and Ho-Jin Choi},
url = {https://ieeexplore.ieee.org/document/10030311},
doi = {10.1109/WACV56688.2023.00187},
issn = {2642-9381},
year = {2023},
date = {2023-01-02},
urldate = {2023-01-01},
booktitle = {2023 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
pages = {1828–1837},
organization = {IEEE},
abstract = {Reference-based video super-resolution (RefVSR) is a promising domain of super-resolution that recovers high-frequency textures of a video using reference video. The multiple cameras with different focal lengths in mobile devices aid recent works in RefVSR, which aim to super-resolve a low-resolution ultra-wide video by utilizing wide-angle videos. Previous works in RefVSR used all reference frames of a Ref video at each time step for the super-resolution of low-resolution videos. However, computation on higher-resolution images increases the runtime and memory consumption, hence hinders the practical application of RefVSR. To solve this problem, we propose an Efficient Reference-based Video Super-Resolution (ERVSR) that exploits a single reference frame to super-resolve whole low-resolution video frames. We introduce an attention-based feature align module and an aggregation upsampling module that attends LR features using the correlation between the reference and LR frames. The proposed ERVSR achieves 12xfaster speed, 1/4 memory consumption than previous state-of-the-art RefVSR networks, and competitive performance on the RealMCVSR dataset while using a single reference image.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Reference-based video super-resolution (RefVSR) is a promising domain of super-resolution that recovers high-frequency textures of a video using reference video. The multiple cameras with different focal lengths in mobile devices aid recent works in RefVSR, which aim to super-resolve a low-resolution ultra-wide video by utilizing wide-angle videos. Previous works in RefVSR used all reference frames of a Ref video at each time step for the super-resolution of low-resolution videos. However, computation on higher-resolution images increases the runtime and memory consumption, hence hinders the practical application of RefVSR. To solve this problem, we propose an Efficient Reference-based Video Super-Resolution (ERVSR) that exploits a single reference frame to super-resolve whole low-resolution video frames. We introduce an attention-based feature align module and an aggregation upsampling module that attends LR features using the correlation between the reference and LR frames. The proposed ERVSR achieves 12xfaster speed, 1/4 memory consumption than previous state-of-the-art RefVSR networks, and competitive performance on the RealMCVSR dataset while using a single reference image.
2022
Jang, Si Young; Park, Sung Kyu; Cho, Jin Hee; Lee, Dongman
CARES: Context-aware trust estimation for realtime crowdsensing services in vehicular edge networks Journal Article
In: ACM Transactions on Internet Technology, vol. 22, no. 4, pp. 1–24, 2022, ISSN: 1533-5399.
@article{jang2022cares,
title = {CARES: Context-aware trust estimation for realtime crowdsensing services in vehicular edge networks},
author = {Si Young Jang and Sung Kyu Park and Jin Hee Cho and Dongman Lee},
url = {https://dl.acm.org/doi/10.1145/3514243},
doi = {10.1145/3514243},
issn = {1533-5399},
year = {2022},
date = {2022-11-14},
urldate = {2022-11-14},
journal = {ACM Transactions on Internet Technology},
volume = {22},
number = {4},
pages = {1–24},
publisher = {ACM New York, NY},
abstract = {The growing number of smart vehicles makes it possible to envision a crowdsensing service where vehicles can share video data of their surroundings for seeking out traffic conditions and car accidents ahead. However, the service may need to deal with situations like malicious vehicles propagating false information to divert other vehicles to arrive at destinations earlier or lead them to dangerous locations. This article proposes a context-aware trust estimation scheme that can allow roadside units in a vehicular edge network to provide real-time crowdsensing services in a reliable manner by selectively using information from trustworthy sources. Our proposed scheme is novel in that its trust estimation does not require any prior knowledge of vehicles on roads but quickly obtains the accurate trust value of each vehicle by leveraging transfer learning. and its Q-learning-based dynamic adjustment scheme autonomously estimates trust levels of oncoming vehicles with the aim of detecting malicious vehicles and accordingly filtering out untrustworthy input from them. Based on an extensive simulation study, we prove that the proposed scheme outperforms existing ones in terms of malicious vehicle detection accuracy.
},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The growing number of smart vehicles makes it possible to envision a crowdsensing service where vehicles can share video data of their surroundings for seeking out traffic conditions and car accidents ahead. However, the service may need to deal with situations like malicious vehicles propagating false information to divert other vehicles to arrive at destinations earlier or lead them to dangerous locations. This article proposes a context-aware trust estimation scheme that can allow roadside units in a vehicular edge network to provide real-time crowdsensing services in a reliable manner by selectively using information from trustworthy sources. Our proposed scheme is novel in that its trust estimation does not require any prior knowledge of vehicles on roads but quickly obtains the accurate trust value of each vehicle by leveraging transfer learning. and its Q-learning-based dynamic adjustment scheme autonomously estimates trust levels of oncoming vehicles with the aim of detecting malicious vehicles and accordingly filtering out untrustworthy input from them. Based on an extensive simulation study, we prove that the proposed scheme outperforms existing ones in terms of malicious vehicle detection accuracy.
Kim, Kisoo; Kim, Hyunju; Lee, Dongman
A Correlation-based Real-time Segmentation Scheme for Multi-user Collaborative Activities Proceedings Article
In: 2022 IEEE 46th Annual Computers, Software, and Applications Conference (COMPSAC), pp. 972–981, IEEE 2022, ISBN: 978-1-6654-8810-5.
@inproceedings{kim2022correlationb,
title = {A Correlation-based Real-time Segmentation Scheme for Multi-user Collaborative Activities},
author = {Kisoo Kim and Hyunju Kim and Dongman Lee},
url = {https://ieeexplore.ieee.org/document/9842485},
doi = {10.1109/COMPSAC54236.2022.00150},
isbn = {978-1-6654-8810-5},
year = {2022},
date = {2022-08-10},
urldate = {2022-08-10},
booktitle = {2022 IEEE 46th Annual Computers, Software, and Applications Conference (COMPSAC)},
pages = {972–981},
organization = {IEEE},
abstract = {Activity Segmentation, dividing a continuous sensor stream into a set of activity segments, is a crucial pre-process in Human Activity Recognition (HAR) and it is required to be done in real-time for real-world smart services. Existing single-user activity segmentation schemes fail to correctly detect transition points due to concurrent and overlapping events from multiple users in case of Multi-user Collaborative Activity Recognition (MCAR). In this paper, we propose a novel scheme for activity segmentation for MCAR that expresses complex events and the correlations between them. For this, the proposed scheme first creates an event stream from a sensor stream and defines event sets in terms of time windows. For each time window, two types of correlations for every event pair are calculated: duration correlation and history correlation. After calculating event correlation, the change score of a time window is measured by comparing the calculated correlation values with those of the preceding windows. Then, the proposed scheme elects as an activity transition point a time window whose change score exceeds the transition threshold. We evaluate the proposed method on two multi-user collaborative activity datasets and experiment results show that the proposed scheme achieves better segmentation performance than existing approaches.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Activity Segmentation, dividing a continuous sensor stream into a set of activity segments, is a crucial pre-process in Human Activity Recognition (HAR) and it is required to be done in real-time for real-world smart services. Existing single-user activity segmentation schemes fail to correctly detect transition points due to concurrent and overlapping events from multiple users in case of Multi-user Collaborative Activity Recognition (MCAR). In this paper, we propose a novel scheme for activity segmentation for MCAR that expresses complex events and the correlations between them. For this, the proposed scheme first creates an event stream from a sensor stream and defines event sets in terms of time windows. For each time window, two types of correlations for every event pair are calculated: duration correlation and history correlation. After calculating event correlation, the change score of a time window is measured by comparing the calculated correlation values with those of the preceding windows. Then, the proposed scheme elects as an activity transition point a time window whose change score exceeds the transition threshold. We evaluate the proposed method on two multi-user collaborative activity datasets and experiment results show that the proposed scheme achieves better segmentation performance than existing approaches.
Kim, Kisoo; Kim, Hyunju; Lee, Dongman
A Correlation-based Real-time Segmentation Scheme for Multi-user Collaborative Activities Proceedings Article
In: 2022 IEEE 46th Annual Computers, Software, and Applications Conference (COMPSAC), pp. 972–981, IEEE 2022, ISBN: 978-1-6654-8810-5.
@inproceedings{kim2022correlation,
title = {A Correlation-based Real-time Segmentation Scheme for Multi-user Collaborative Activities},
author = {Kisoo Kim and Hyunju Kim and Dongman Lee},
url = {https://ieeexplore.ieee.org/document/9842485},
doi = {10.1109/COMPSAC54236.2022.00150},
isbn = {978-1-6654-8810-5},
year = {2022},
date = {2022-01-01},
urldate = {2022-01-01},
booktitle = {2022 IEEE 46th Annual Computers, Software, and Applications Conference (COMPSAC)},
pages = {972–981},
organization = {IEEE},
abstract = {Activity Segmentation, dividing a continuous sensor stream into a set of activity segments, is a crucial pre-process in Human Activity Recognition (HAR) and it is required to be done in real-time for real-world smart services. Existing single-user activity segmentation schemes fail to correctly detect transition points due to concurrent and overlapping events from multiple users in case of Multi-user Collaborative Activity Recognition (MCAR). In this paper, we propose a novel scheme for activity segmentation for MCAR that expresses complex events and the correlations between them. For this, the proposed scheme first creates an event stream from a sensor stream and defines event sets in terms of time windows. For each time window, two types of correlations for every event pair are calculated: duration correlation and history correlation. After calculating event correlation, the change score of a time window is measured by comparing the calculated correlation values with those of the preceding windows. Then, the proposed scheme elects as an activity transition point a time window whose change score exceeds the transition threshold. We evaluate the proposed method on two multi-user collaborative activity datasets and experiment results show that the proposed scheme achieves better segmentation performance than existing approaches.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Activity Segmentation, dividing a continuous sensor stream into a set of activity segments, is a crucial pre-process in Human Activity Recognition (HAR) and it is required to be done in real-time for real-world smart services. Existing single-user activity segmentation schemes fail to correctly detect transition points due to concurrent and overlapping events from multiple users in case of Multi-user Collaborative Activity Recognition (MCAR). In this paper, we propose a novel scheme for activity segmentation for MCAR that expresses complex events and the correlations between them. For this, the proposed scheme first creates an event stream from a sensor stream and defines event sets in terms of time windows. For each time window, two types of correlations for every event pair are calculated: duration correlation and history correlation. After calculating event correlation, the change score of a time window is measured by comparing the calculated correlation values with those of the preceding windows. Then, the proposed scheme elects as an activity transition point a time window whose change score exceeds the transition threshold. We evaluate the proposed method on two multi-user collaborative activity datasets and experiment results show that the proposed scheme achieves better segmentation performance than existing approaches.
2021
Cho, Eunho; Yoon, Juyeon; Baek, Daehyeon; Lee, Dongman; Bae, Doo-Hwan
Dnn model deployment on distributed edges Proceedings Article
In: International Conference on Web Engineering, pp. 15–26, Springer 2021, ISBN: 978-3-030-92230-6.
@inproceedings{cho2021dnn,
title = {Dnn model deployment on distributed edges},
author = {Eunho Cho and Juyeon Yoon and Daehyeon Baek and Dongman Lee and Doo-Hwan Bae},
url = {https://link.springer.com/chapter/10.1007/978-3-030-92231-3_2},
doi = {10.1007/978-3-030-92231-3_2},
isbn = {978-3-030-92230-6},
year = {2021},
date = {2021-12-08},
urldate = {2021-12-08},
booktitle = {International Conference on Web Engineering},
pages = {15–26},
organization = {Springer},
abstract = {Deep learning-based visual analytic applications have drawn attention by suggesting fruitful combinations with Deep Neural Network (DNN) models and visual data sensors. Because of the high cost of DNN inference, most systems adopt offloading techniques utilizing a high-end cloud. However, tasks that require real-time streaming often suffer from the problem of an imbalanced pipeline due to the limited bandwidth and latency between camera sensors and the cloud. Several DNN slicing approaches show that effectively utilizing the edge computing paradigm effectively lowers the frame drop rate and overall latency, but recent research has primarily focused on building a general framework that only considers a few fixed settings. However, we observed that the optimal split strategy for DNN models can vary significantly based on application requirements. Hence, we focus on the characteristics and explainability of split points derived from various application goals. First, we propose a new simulation framework for flexible software-level configuration, including latency and bandwidth, using dockercompose, and we experiment on a 14-layered Convolutional Neural Network (CNN) model with diverse layer types. We report the results of the total process time and frame drop rate of 50 frames with three different configurations and further discuss recommendations for providing proper decision guidelines on split points, considering the target goals and properties of the CNN layers.
},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Deep learning-based visual analytic applications have drawn attention by suggesting fruitful combinations with Deep Neural Network (DNN) models and visual data sensors. Because of the high cost of DNN inference, most systems adopt offloading techniques utilizing a high-end cloud. However, tasks that require real-time streaming often suffer from the problem of an imbalanced pipeline due to the limited bandwidth and latency between camera sensors and the cloud. Several DNN slicing approaches show that effectively utilizing the edge computing paradigm effectively lowers the frame drop rate and overall latency, but recent research has primarily focused on building a general framework that only considers a few fixed settings. However, we observed that the optimal split strategy for DNN models can vary significantly based on application requirements. Hence, we focus on the characteristics and explainability of split points derived from various application goals. First, we propose a new simulation framework for flexible software-level configuration, including latency and bandwidth, using dockercompose, and we experiment on a 14-layered Convolutional Neural Network (CNN) model with diverse layer types. We report the results of the total process time and frame drop rate of 50 frames with three different configurations and further discuss recommendations for providing proper decision guidelines on split points, considering the target goals and properties of the CNN layers.
Son, Heesuk; Lee, Dongman
An Efficient Interaction Protocol Inference Scheme for Incompatible Updates in IoT Environments Journal Article
In: ACM Transactions on Internet Technology (TOIT), vol. 22, no. 2, pp. 1–25, 2021, ISSN: 1533-5399.
@article{son2021efficient,
title = {An Efficient Interaction Protocol Inference Scheme for Incompatible Updates in IoT Environments},
author = {Heesuk Son and Dongman Lee},
url = {https://dl.acm.org/doi/10.1145/3430501},
doi = {10.1145/3430501},
issn = {1533-5399},
year = {2021},
date = {2021-10-22},
urldate = {2021-10-22},
journal = {ACM Transactions on Internet Technology (TOIT)},
volume = {22},
number = {2},
pages = {1–25},
publisher = {ACM New York, NY},
abstract = {Incompatible updates of IoT systems and protocols give rise to interoperability problems. Even though various protocol adaptation and unknown protocol inference schemes have been proposed, they either do not work where the updated protocol specifications are not given or suffer from inefficiency issues. In this work, we present an efficient protocol inference scheme for incompatible updates in IoT environments. The scheme refines an active automata learning algorithm, L*, by incorporating a knowledge base of the legacy protocol behavior into its membership query selection procedure for updated protocol behavior inference. It also infers protocol syntax based on our previous work that computes the most probable message field updates and adapts the legacy protocol message accordingly. We evaluate the proposed scheme with two case studies with the most popular IoT protocols and prove that it infers updated protocols efficiently while improving the L* algorithm’s performance for resolving the incompatibility.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Incompatible updates of IoT systems and protocols give rise to interoperability problems. Even though various protocol adaptation and unknown protocol inference schemes have been proposed, they either do not work where the updated protocol specifications are not given or suffer from inefficiency issues. In this work, we present an efficient protocol inference scheme for incompatible updates in IoT environments. The scheme refines an active automata learning algorithm, L*, by incorporating a knowledge base of the legacy protocol behavior into its membership query selection procedure for updated protocol behavior inference. It also infers protocol syntax based on our previous work that computes the most probable message field updates and adapts the legacy protocol message accordingly. We evaluate the proposed scheme with two case studies with the most popular IoT protocols and prove that it infers updated protocols efficiently while improving the L* algorithm’s performance for resolving the incompatibility.
Kim, Hyunju; Lee, Dongman
ADELA: attention based deep ensemble learning for activity recognition in smart collaborative environments Proceedings Article
In: Proceedings of the 36th Annual ACM Symposium on Applied Computing, pp. 436–444, 2021, ISBN: 9781450381048.
@inproceedings{kim2021adela,
title = {ADELA: attention based deep ensemble learning for activity recognition in smart collaborative environments},
author = {Hyunju Kim and Dongman Lee},
url = {https://dl.acm.org/doi/10.1145/3412841.344192},
doi = {10.1109/COMPSAC54236.2022.00150},
isbn = {9781450381048},
year = {2021},
date = {2021-04-22},
urldate = {2021-04-22},
booktitle = {Proceedings of the 36th Annual ACM Symposium on Applied Computing},
pages = {436–444},
abstract = {Human activity recognition (HAR) is a key challenge in pervasive computing and its solutions have been presented based on various disciplines. For HAR in a smart environment without privacy and accessibility issues, data streams generated by environmental sensors are leveraged. In this paper, we propose ADELA, a novel activity recognition scheme in a smart collaborative environment where a group of users performs an activity without user identification. ADELA calculates the importance of events from sensor data streams depending on their impact on recognizing activities and finds the best activity recognition base models using attention-based ensemble learning. After the training phase, each base model obtains its weight through weighted majority calculation, and ADELA stores the information to Trained Model Storage to reuse it for inferring. We evaluate ADELA using the data collected from our testbed and CASAS dataset where users perform their tasks daily and validate the effectiveness of ADELA in a real environment. Experiment results show that the proposed scheme performs higher recognition performance than existing approaches.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Human activity recognition (HAR) is a key challenge in pervasive computing and its solutions have been presented based on various disciplines. For HAR in a smart environment without privacy and accessibility issues, data streams generated by environmental sensors are leveraged. In this paper, we propose ADELA, a novel activity recognition scheme in a smart collaborative environment where a group of users performs an activity without user identification. ADELA calculates the importance of events from sensor data streams depending on their impact on recognizing activities and finds the best activity recognition base models using attention-based ensemble learning. After the training phase, each base model obtains its weight through weighted majority calculation, and ADELA stores the information to Trained Model Storage to reuse it for inferring. We evaluate ADELA using the data collected from our testbed and CASAS dataset where users perform their tasks daily and validate the effectiveness of ADELA in a real environment. Experiment results show that the proposed scheme performs higher recognition performance than existing approaches.
Lee, Seungho; Jang, Si Young; Hyun, Soon J; Lee, Dongman
DQN-based coverage maximization for mobile video camera networks Proceedings Article
In: 2021 IEEE 18th Annual Consumer Communications & Networking Conference (CCNC), pp. 01–02, IEEE 2021, ISBN: 978-1-7281-9794-4.
@inproceedings{lee2021dqn,
title = {DQN-based coverage maximization for mobile video camera networks},
author = {Seungho Lee and Si Young Jang and Soon J Hyun and Dongman Lee},
url = {https://ieeexplore.ieee.org/document/9369618},
doi = {10.1109/CCNC49032.2021.9369618},
isbn = {978-1-7281-9794-4},
year = {2021},
date = {2021-03-11},
urldate = {2021-01-01},
booktitle = {2021 IEEE 18th Annual Consumer Communications & Networking Conference (CCNC)},
pages = {01–02},
organization = {IEEE},
abstract = {Coverage maximization is an important issue of Mobile wireless sensor networks (M-WSN). Especially for visual sensors like video camera which have a specific sensing direction and range, obstacles and the position of the sensors should also be considered. In this paper, we propose an efficient coverage maximization method for mobile video camera networks by leveraging DQN, a deep reinforcement learning algorithm. Evaluation results show that the proposed method can cover up to 4.93% better than existing ones.
},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Coverage maximization is an important issue of Mobile wireless sensor networks (M-WSN). Especially for visual sensors like video camera which have a specific sensing direction and range, obstacles and the position of the sensors should also be considered. In this paper, we propose an efficient coverage maximization method for mobile video camera networks by leveraging DQN, a deep reinforcement learning algorithm. Evaluation results show that the proposed method can cover up to 4.93% better than existing ones.
Jang, Si Young; Kostadinov, Boyan; Lee, Dongman
Microservice-based edge device architecture for video analytics Proceedings Article
In: 2021 IEEE/ACM Symposium on Edge Computing (SEC), pp. 165–177, IEEE Computer Society 2021, ISBN: 978-1-4503-8390-5.
@inproceedings{jang2021microservice,
title = {Microservice-based edge device architecture for video analytics},
author = {Si Young Jang and Boyan Kostadinov and Dongman Lee},
url = {https://ieeexplore.ieee.org/document/9709018},
doi = {10.1145/3453142.3491283},
isbn = {978-1-4503-8390-5},
year = {2021},
date = {2021-01-01},
urldate = {2021-01-01},
booktitle = {2021 IEEE/ACM Symposium on Edge Computing (SEC)},
pages = {165–177},
organization = {IEEE Computer Society},
abstract = {With today's ubiquitous deployment of video cameras and other edge devices, progress in edge computing is happening at an incredible speed. Yet, one aspect of real-time video analytics at the edge that is still underdeveloped is the support for processing multitenant, multi-application scenarios with a limited set of resources. Existing systems either fail to provide the necessary performance, or rely too heavily on edge or cloud servers to handle the workload. This work proposes a new approach, inspired by both Function-as-a-Service and microservices architecture in order to efficiently place and execute video analytics pipelines on edge devices. The main contributions of this work are the ability to dynamically add and run new applications on already deployed systems, and the capability to horizontally distribute pipelines across other neigh-bouring edge devices. We prototype an implementation that we evaluate using multiple concurrent applications per device. Results show that our system provides more flexibility for on-the-fly re-configuration than existing works do, with 20 % improvement in latency and 3.9 X increase in throughput.
},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
With today's ubiquitous deployment of video cameras and other edge devices, progress in edge computing is happening at an incredible speed. Yet, one aspect of real-time video analytics at the edge that is still underdeveloped is the support for processing multitenant, multi-application scenarios with a limited set of resources. Existing systems either fail to provide the necessary performance, or rely too heavily on edge or cloud servers to handle the workload. This work proposes a new approach, inspired by both Function-as-a-Service and microservices architecture in order to efficiently place and execute video analytics pipelines on edge devices. The main contributions of this work are the ability to dynamically add and run new applications on already deployed systems, and the capability to horizontally distribute pipelines across other neigh-bouring edge devices. We prototype an implementation that we evaluate using multiple concurrent applications per device. Results show that our system provides more flexibility for on-the-fly re-configuration than existing works do, with 20 % improvement in latency and 3.9 X increase in throughput.
Kim, Hyunju; Dongman,
AR-T: Temporal relation embedded transformer for the real world activity recognition Proceedings Article
In: International Conference on Mobile and Ubiquitous Systems: Computing, Networking, and Services, pp. 617–633, Springer 2021, ISBN: 978-3-030-94821-4.
@inproceedings{kim2021ar,
title = {AR-T: Temporal relation embedded transformer for the real world activity recognition},
author = {Hyunju Kim and Dongman},
url = {https://link.springer.com/chapter/10.1007/978-3-030-94822-1_40},
doi = {10.1007/978-3-030-94822-1_40},
isbn = {978-3-030-94821-4},
year = {2021},
date = {2021-01-01},
urldate = {2021-01-01},
booktitle = {International Conference on Mobile and Ubiquitous Systems: Computing, Networking, and Services},
pages = {617–633},
organization = {Springer},
abstract = {Activity recognition is a fundamental way to support context-aware services for users in smart spaces. Data sources such as video or wearable devices are used in many recognition approaches, but there are challenges in utilizing them in the real world. Recent approaches propose deep learning-based methods on IoT sensor data streams to overcome the issues. Since they only describe single user-based spaces, they are vulnerable to complex sequences of events triggered by multiple users. When multiple users exist in a space, various overlapping events occur with longer correlations than a single user situation. Additionally, ambient sensor-based events appear far more than actuator-based events, making it difficult to extract actuator-based events as important features. We propose a transformer-based approach to derive long-term event correlations and important events as elements of activity patterns. We also develop a duration incorporated embedding method to differentiate between the same type but different duration events and add a sequential manner to the transformer approach. In the experiments section, we prove that our approach outperforms the existing approaches based on real datasets.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Activity recognition is a fundamental way to support context-aware services for users in smart spaces. Data sources such as video or wearable devices are used in many recognition approaches, but there are challenges in utilizing them in the real world. Recent approaches propose deep learning-based methods on IoT sensor data streams to overcome the issues. Since they only describe single user-based spaces, they are vulnerable to complex sequences of events triggered by multiple users. When multiple users exist in a space, various overlapping events occur with longer correlations than a single user situation. Additionally, ambient sensor-based events appear far more than actuator-based events, making it difficult to extract actuator-based events as important features. We propose a transformer-based approach to derive long-term event correlations and important events as elements of activity patterns. We also develop a duration incorporated embedding method to differentiate between the same type but different duration events and add a sequential manner to the transformer approach. In the experiments section, we prove that our approach outperforms the existing approaches based on real datasets.
Kim, Hyunju; Lee, Dongman
TAP: A Transformer based Activity Prediction Exploiting Temporal Relations in Collaborative Tasks Proceedings Article
In: 2021 IEEE International Conference on Pervasive Computing and Communications Workshops and other Affiliated Events (PerCom Workshops), IEEE 2021, ISBN: 978-1-6654-0424-2.
@inproceedings{kim2021tap,
title = {TAP: A Transformer based Activity Prediction Exploiting Temporal Relations in Collaborative Tasks},
author = {Hyunju Kim and Dongman Lee},
url = {https://ieeexplore.ieee.org/document/9431008},
doi = {10.1109/PerComWorkshops51409.2021.9431008},
isbn = {978-1-6654-0424-2},
year = {2021},
date = {2021-01-01},
urldate = {2021-01-01},
booktitle = {2021 IEEE International Conference on Pervasive Computing and Communications Workshops and other Affiliated Events (PerCom Workshops)},
organization = {IEEE},
abstract = {Activity prediction is an important challenge to provide cognitive supports in a smart space. For user activity prediction without privacy and inconsistent data collection issues, recent studies leverage data stream generated by ambient sensors. They use a deep learning model for single-user tasks which only contain simple sequential relations between activities. In this paper, we propose TAP, a Transformer-based activity prediction approach for inferring the next activity in a collaborative smart environment where collaborative tasks are conducted. To represent and analyze complex relations between activities of users, Allen's temporal relations are employed for representing temporal relations between activities and we leverage the Transformer to predict not only the next activity but also temporal relation with its preceding activity. TAP yields higher accuracy for the next activity prediction and temporal relations with the current activity in a collaborative task than existing approaches.
},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Activity prediction is an important challenge to provide cognitive supports in a smart space. For user activity prediction without privacy and inconsistent data collection issues, recent studies leverage data stream generated by ambient sensors. They use a deep learning model for single-user tasks which only contain simple sequential relations between activities. In this paper, we propose TAP, a Transformer-based activity prediction approach for inferring the next activity in a collaborative smart environment where collaborative tasks are conducted. To represent and analyze complex relations between activities of users, Allen's temporal relations are employed for representing temporal relations between activities and we leverage the Transformer to predict not only the next activity but also temporal relation with its preceding activity. TAP yields higher accuracy for the next activity prediction and temporal relations with the current activity in a collaborative task than existing approaches.
2020
Kim, Dongmin; Han, Sumin; Son, Heesuk; Lee, Dongman
Human Activity Recognition Using Semi-supervised Multi-modal DEC for Instagram Data Proceedings Article
In: Advances in Knowledge Discovery and Data Mining: 24th Pacific-Asia Conference, PAKDD 2020, Singapore, May 11–14, 2020, Proceedings, Part I 24, pp. 869–880, Springer 2020, ISBN: 978-3-030-47425-6.
@inproceedings{kim2020human,
title = {Human Activity Recognition Using Semi-supervised Multi-modal DEC for Instagram Data},
author = {Dongmin Kim and Sumin Han and Heesuk Son and Dongman Lee},
url = {https://dl.acm.org/doi/10.1007/978-3-030-47426-3_67},
doi = {10.1007/978-3-030-47426-3_67},
isbn = {978-3-030-47425-6},
year = {2020},
date = {2020-05-11},
urldate = {2020-05-11},
booktitle = {Advances in Knowledge Discovery and Data Mining: 24th Pacific-Asia Conference, PAKDD 2020, Singapore, May 11–14, 2020, Proceedings, Part I 24},
pages = {869–880},
organization = {Springer},
abstract = {Human Activity Recognition (HAR) using social media provides a solid basis for a variety of context-aware applications. Existing HAR approaches have adopted supervised machine learning algorithms using texts and their meta-data such as time, venue, and keywords. However, their recognition accuracy may decrease when applied to image-sharing social media where users mostly describe their daily activities and thoughts using both texts and images. In this paper, we propose a semi-supervised multi-modal deep embedding clustering method to recognize human activities on Instagram. Our proposed method learns multi-modal feature representations by alternating a supervised learning phase and an unsupervised learning phase. By utilizing a large number of unlabeled data, it learns a more generalized feature distribution for each HAR class and avoids overfitting to limited labeled data. Evaluation results show that leveraging multi-modality and unlabeled data is effective for HAR and our method outperforms existing approaches.
},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Human Activity Recognition (HAR) using social media provides a solid basis for a variety of context-aware applications. Existing HAR approaches have adopted supervised machine learning algorithms using texts and their meta-data such as time, venue, and keywords. However, their recognition accuracy may decrease when applied to image-sharing social media where users mostly describe their daily activities and thoughts using both texts and images. In this paper, we propose a semi-supervised multi-modal deep embedding clustering method to recognize human activities on Instagram. Our proposed method learns multi-modal feature representations by alternating a supervised learning phase and an unsupervised learning phase. By utilizing a large number of unlabeled data, it learns a more generalized feature distribution for each HAR class and avoids overfitting to limited labeled data. Evaluation results show that leveraging multi-modality and unlabeled data is effective for HAR and our method outperforms existing approaches.
Yang, Hongmin; Cho, Jin-Hee; Son, Heesuk; Lee, Dongman
Context-aware trust estimation for realtime crowdsensing services in vehicular edge networks Proceedings Article
In: 2020 IEEE 17th Annual Consumer Communications & Networking Conference (CCNC), pp. 1–6, IEEE 2020, ISBN: 978-1-7281-3893-0.
@inproceedings{yang2020context,
title = {Context-aware trust estimation for realtime crowdsensing services in vehicular edge networks},
author = {Hongmin Yang and Jin-Hee Cho and Heesuk Son and Dongman Lee},
url = {https://ieeexplore.ieee.org/abstract/document/9045221},
doi = {10.1109/CCNC46108.2020.9045221},
isbn = {978-1-7281-3893-0},
year = {2020},
date = {2020-03-26},
urldate = {2020-01-01},
booktitle = {2020 IEEE 17th Annual Consumer Communications & Networking Conference (CCNC)},
pages = {1–6},
organization = {IEEE},
abstract = {This work proposes a context-aware trust estimation scheme that can allow roadside units in a vehicular edge network to provide real-time crowdsensing services in a reliable manner by selectively using information from trustworthy sources. Our proposed scheme is novel in that its trust estimation does not require any prior knowledge towards vehicles on roads but quickly obtains an accurate trust value of each vehicle. To that end, we particularly leverage the concept of I-sharing which removes a cold-start problem during the system bootstrapping period. Based on an extensive simulation study, we prove that the proposed scheme outperforms its competitive counterpart and baseline models in terms of trust bias and malicious vehicle detection accuracy.
},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This work proposes a context-aware trust estimation scheme that can allow roadside units in a vehicular edge network to provide real-time crowdsensing services in a reliable manner by selectively using information from trustworthy sources. Our proposed scheme is novel in that its trust estimation does not require any prior knowledge towards vehicles on roads but quickly obtains an accurate trust value of each vehicle. To that end, we particularly leverage the concept of I-sharing which removes a cold-start problem during the system bootstrapping period. Based on an extensive simulation study, we prove that the proposed scheme outperforms its competitive counterpart and baseline models in terms of trust bias and malicious vehicle detection accuracy.
Han, Sumin; Hong, Dasom; Lee, Dongman
Exploring commercial gentrification using Instagram data Proceedings Article
In: 2020 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM), pp. 557–564, IEEE 2020, ISBN: 978-1-7281-1056-1.
@inproceedings{han2020exploring,
title = {Exploring commercial gentrification using Instagram data},
author = {Sumin Han and Dasom Hong and Dongman Lee},
url = {https://ieeexplore.ieee.org/abstract/document/9381374},
doi = {10.1109/ASONAM49781.2020.9381374},
isbn = {978-1-7281-1056-1},
year = {2020},
date = {2020-03-24},
urldate = {2020-03-24},
booktitle = {2020 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM)},
pages = {557–564},
organization = {IEEE},
abstract = {Commercial gentrification refers to the replacement of low-value businesses like small local stores into high-value businesses like boutiques and chain stores. A handful of research efforts have been made to identify gentrification and their change by leveraging social media. However, their approaches lack in inferring how much commercial gentrification is developed in a target area and how long it has taken for the area to get to that phase. In this paper, we propose a novel scheme to estimate the commercial gentrification status of a target area and its development in terms of time and geographic dispersion using Instagram data. For this, we define our commercial gentrification phase criteria based on the conceptual model from the urban study. Then, we extract social features from both images and texts of Instagram posts, and leverage regression models to infer the commercial gentrification phase of a target area at the monthly timestamp. We also measure how geographical dispersion of geo-tagged Instagram posts matches the boutiques, which is the physical variable that has the strongest correlation with the commercial gentrification. Evaluation results show that our method yields a good quality of estimation compared to the ground truth. This assures that our method could be a meaningful tool for urban planners and policymakers to investigate and manage commercial gentrification.
},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Commercial gentrification refers to the replacement of low-value businesses like small local stores into high-value businesses like boutiques and chain stores. A handful of research efforts have been made to identify gentrification and their change by leveraging social media. However, their approaches lack in inferring how much commercial gentrification is developed in a target area and how long it has taken for the area to get to that phase. In this paper, we propose a novel scheme to estimate the commercial gentrification status of a target area and its development in terms of time and geographic dispersion using Instagram data. For this, we define our commercial gentrification phase criteria based on the conceptual model from the urban study. Then, we extract social features from both images and texts of Instagram posts, and leverage regression models to infer the commercial gentrification phase of a target area at the monthly timestamp. We also measure how geographical dispersion of geo-tagged Instagram posts matches the boutiques, which is the physical variable that has the strongest correlation with the commercial gentrification. Evaluation results show that our method yields a good quality of estimation compared to the ground truth. This assures that our method could be a meaningful tool for urban planners and policymakers to investigate and manage commercial gentrification.
Kim, Dongjae; Lee, Dongman; Kim, Myungchul; Hyun, Soon J
Enrichment of ontology instances using linked data and supplemental string data Conference
2018 International Conference on Computational Science and Computational Intelligence (CSCI), IEEE 2020, ISBN: 978-1-7281-1360-9.
@conference{kim2018enrichment,
title = {Enrichment of ontology instances using linked data and supplemental string data},
author = {Dongjae Kim and Dongman Lee and Myungchul Kim and Soon J Hyun},
url = {https://ieeexplore.ieee.org/abstract/document/8947684},
doi = {10.1109/CSCI46756.2018.00188},
isbn = {978-1-7281-1360-9},
year = {2020},
date = {2020-01-02},
urldate = {2020-01-02},
booktitle = {2018 International Conference on Computational Science and Computational Intelligence (CSCI)},
pages = {966–971},
organization = {IEEE},
abstract = {Various emerging applications using IoT require frequent ontology re-construction in ways to enrich the knowledge map. We propose a hybrid approach for a new semi-automated ontology enrichment system which expands the present ontology by using both linked data extracted from other ontology of the same (or similar) application domain and string data crawled from the web. Our system enriches an ontology with new instances of concepts and relations in two phases. First, it extracts new instances and relations from a reference ontology of the same or similar domain. Second, it validates the possible relations between the original instances and the new ones using crawled data from the web search. Our system computes confidence value to check validity of those relations before adding them to the present ontology. Our experiment demonstrates that the proposed two-phase hybrid approach achieves improved efficiency and accuracy for enriching ontology instances.},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Various emerging applications using IoT require frequent ontology re-construction in ways to enrich the knowledge map. We propose a hybrid approach for a new semi-automated ontology enrichment system which expands the present ontology by using both linked data extracted from other ontology of the same (or similar) application domain and string data crawled from the web. Our system enriches an ontology with new instances of concepts and relations in two phases. First, it extracts new instances and relations from a reference ontology of the same or similar domain. Second, it validates the possible relations between the original instances and the new ones using crawled data from the web search. Our system computes confidence value to check validity of those relations before adding them to the present ontology. Our experiment demonstrates that the proposed two-phase hybrid approach achieves improved efficiency and accuracy for enriching ontology instances.
Han, Sumin; Park, Kinam; Lee, Dongman
Discovering Daily POI Exploitation Using LTE Cell Tower Access Traces in Urban Environment Proceedings Article
In: International Conference on Social Informatics, pp. 81–94, Springer 2020, ISBN: 978-3-030-60974-0.
@inproceedings{han2020discovering,
title = {Discovering Daily POI Exploitation Using LTE Cell Tower Access Traces in Urban Environment},
author = {Sumin Han and Kinam Park and Dongman Lee},
url = {https://link.springer.com/chapter/10.1007/978-3-030-60975-7_7},
doi = {10.1007/978-3-030-60975-7_7},
isbn = {978-3-030-60974-0},
year = {2020},
date = {2020-01-01},
urldate = {2020-01-01},
booktitle = {International Conference on Social Informatics},
pages = {81–94},
organization = {Springer},
abstract = {Point of interest (POI) in an urban space represents the perception of city dwellers and visitors of a certain place. LTE cell tower access trace data is one of the promising data sources which has the potential to show real-time POI exploitation analysis. However, there is not much discussion on how it is correlated to diachronic POIs and their exploitation pattern. In this paper, we first show that the access trace pattern from the LTE cell tower can be used to discover which types of POIs exist in a certain area. Then, we propose a daily POI exploitation discovery scheme which can extract patterns of how POIs are daily used. Our analysis can provide a good insight into future urban space-based services such as urban planning and tourism.
},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Point of interest (POI) in an urban space represents the perception of city dwellers and visitors of a certain place. LTE cell tower access trace data is one of the promising data sources which has the potential to show real-time POI exploitation analysis. However, there is not much discussion on how it is correlated to diachronic POIs and their exploitation pattern. In this paper, we first show that the access trace pattern from the LTE cell tower can be used to discover which types of POIs exist in a certain area. Then, we propose a daily POI exploitation discovery scheme which can extract patterns of how POIs are daily used. Our analysis can provide a good insight into future urban space-based services such as urban planning and tourism.
2019
Son, Heesuk; Park, Jeongwook; Kim, Hyunju; Lee, Dongman
Distributed multi-agent preference learning for an IoT-enriched smart space Proceedings Article
In: 2019 IEEE 39th International Conference on Distributed Computing Systems (ICDCS), pp. 2090–2100, IEEE 2019, ISBN: 978-1-7281-2519-0.
@inproceedings{son2019distributed,
title = {Distributed multi-agent preference learning for an IoT-enriched smart space},
author = {Heesuk Son and Jeongwook Park and Hyunju Kim and Dongman Lee},
url = {https://ieeexplore.ieee.org/document/8884817},
doi = {10.1109/ICDCS.2019.00206},
isbn = {978-1-7281-2519-0},
year = {2019},
date = {2019-10-31},
urldate = {2019-10-31},
booktitle = {2019 IEEE 39th International Conference on Distributed Computing Systems (ICDCS)},
pages = {2090–2100},
organization = {IEEE},
abstract = {There have been several efforts on preference learning in a smart space by means of multi-agent collaborations. Each agent captures a user action or handles part of learning but decision makings are done in a centralized manner. This makes it difficult for a smart space to deal with learning complexity due to the increase and reconfiguration of smart devices. While the complexity is relieved by articulating the learning space, it is not flexible because the articulation procedure needs to be resumed whenever a smart space reconfiguration occurs. In this paper, we propose a distributed multi-agent preference learning architecture which allows a group of physically separate agents to collaborate with each other for learning a user's task preference efficiently in an IoT enriched smart space. For this, the proposed scheme provides four key features: ontology-based knowledge structure for task-driven agent collaboration, knowledge exchange protocol for task-aware causality among agents, Q-learners for observing and learning from user behaviors, and negotiation and acknowledgement protocol for preventing agents from performing disorganized actions. Evaluation results show that the proposed scheme allows smart device agents to learn user preferences in a fully distributed way and outperforms existing approaches in terms of learning speed and system overhead.
},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
There have been several efforts on preference learning in a smart space by means of multi-agent collaborations. Each agent captures a user action or handles part of learning but decision makings are done in a centralized manner. This makes it difficult for a smart space to deal with learning complexity due to the increase and reconfiguration of smart devices. While the complexity is relieved by articulating the learning space, it is not flexible because the articulation procedure needs to be resumed whenever a smart space reconfiguration occurs. In this paper, we propose a distributed multi-agent preference learning architecture which allows a group of physically separate agents to collaborate with each other for learning a user's task preference efficiently in an IoT enriched smart space. For this, the proposed scheme provides four key features: ontology-based knowledge structure for task-driven agent collaboration, knowledge exchange protocol for task-aware causality among agents, Q-learners for observing and learning from user behaviors, and negotiation and acknowledgement protocol for preventing agents from performing disorganized actions. Evaluation results show that the proposed scheme allows smart device agents to learn user preferences in a fully distributed way and outperforms existing approaches in terms of learning speed and system overhead.
Son, Heesuk; Lee, Dongman
Towards interactive networking: Runtime message inference approach for incompatible protocol updates in IoT environments Journal Article
In: Future Generation Computer Systems, vol. 96, pp. 563–578, 2019, ISBN: 0167-739X.
@article{son2019towards,
title = {Towards interactive networking: Runtime message inference approach for incompatible protocol updates in IoT environments},
author = {Heesuk Son and Dongman Lee},
url = {https://dl.acm.org/doi/10.1016/j.future.2019.02.007},
doi = {10.1016/j.future.2019.02.007},
isbn = {0167-739X},
year = {2019},
date = {2019-07-01},
urldate = {2019-01-01},
journal = {Future Generation Computer Systems},
volume = {96},
pages = {563–578},
publisher = {Elsevier},
abstract = {As IoT (Internet of Things) devices become pervasive in our surroundings, it is more important for them to dynamically discover and interact with nearby IoT devices. However, as interaction protocols are often updated without backward compatibility, interaction opportunities between smart objects may disappear. To overcome this, we can apply existing works which try to adapt one interaction protocol to another at runtime, assuming their specifications are given. However, they cannot generate a valid adapter if the specification of the updated protocol syntax is not available. Automatic protocol reverse engineering methods can extract protocol message syntax without prior knowledge, but they cannot be applied to a runtime scenario. In this paper, we propose SeM2Bit, an efficient protocol message inference scheme which adapts a legacy protocol’s message based on protocol domain knowledge. Iterative message adaptations make a legacy protocol agent interact with its updated but incompatible version at runtime and have a meaningful interaction without the corresponding specification. The experiment result shows that the use of knowledge is effective to make a meaningful interaction between the incompatible versions with a reasonably small number of message adaptations.
},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
As IoT (Internet of Things) devices become pervasive in our surroundings, it is more important for them to dynamically discover and interact with nearby IoT devices. However, as interaction protocols are often updated without backward compatibility, interaction opportunities between smart objects may disappear. To overcome this, we can apply existing works which try to adapt one interaction protocol to another at runtime, assuming their specifications are given. However, they cannot generate a valid adapter if the specification of the updated protocol syntax is not available. Automatic protocol reverse engineering methods can extract protocol message syntax without prior knowledge, but they cannot be applied to a runtime scenario. In this paper, we propose SeM2Bit, an efficient protocol message inference scheme which adapts a legacy protocol’s message based on protocol domain knowledge. Iterative message adaptations make a legacy protocol agent interact with its updated but incompatible version at runtime and have a meaningful interaction without the corresponding specification. The experiment result shows that the use of knowledge is effective to make a meaningful interaction between the incompatible versions with a reasonably small number of message adaptations.
Kim, Taehun; Lee, Dongman; Hyun, Soon J; Doh, Young Yim
UrbanSocialRadar: a place-aware social matching model for estimating serendipitous interaction willingness in Korean cultural context Journal Article
In: International Journal of Human-Computer Studies, vol. 125, pp. 81–103, 2019, ISBN: 1071-5819.
@article{kim2019urbansocialradar,
title = {UrbanSocialRadar: a place-aware social matching model for estimating serendipitous interaction willingness in Korean cultural context},
author = {Taehun Kim and Dongman Lee and Soon J Hyun and Young Yim Doh},
url = {https://dl.acm.org/doi/10.1016/j.ijhcs.2018.12.011},
doi = {10.1016/j.ijhcs.2018.12.011},
isbn = {1071-5819},
year = {2019},
date = {2019-05-01},
urldate = {2019-01-01},
journal = {International Journal of Human-Computer Studies},
volume = {125},
pages = {81–103},
publisher = {Elsevier},
abstract = {In the era of perpetual digital connectedness, information and communication technology has significantly altered the way people communicate and interact with each other. Nonetheless, the computer-mediated communication should only complement offline communication rather than substituting it, as the resultant online ties are not as strong as face-to-face ties. In an effort to understand the motives in making offline social interactions real and ultimately to predict willingness to engage in serendipitous interactions with people encountered in a public place, we propose a place-aware social matching model driven by interpersonal factors (i.e., similarity, complementarity, and intimacy) and socio-spatial factors (i.e., place sociability, information acquisition expectancy, and perceived personal space in a place). Through a web-based social matching survey experiment (N = 1139 matches from 99 participants in Korea) based on a bogus stranger paradigm, we examine the interrelationship between those factors and the interaction willingness using a series of multiple regression analyses and build a prediction model by devising predictive features based on several machine learning models. From this, we find that both factors have statistically significant influence on interaction willingness, yet interpersonal factors have a higher relative importance than the socio-spatial factors. The interesting point is that the predictive power of these factors varies according to the place characteristics and the level of interaction willingness. We also empirically test the predictability of the model built from the controlled lab experiment through real-world experiments. The results reveal that the proposed model predicts interaction willingness in a real world with under 21% error rate within the Korean cultural context. Findings have implications for the design of mobile social networking systems that endeavor to facilitate serendipitous interactions.
},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
In the era of perpetual digital connectedness, information and communication technology has significantly altered the way people communicate and interact with each other. Nonetheless, the computer-mediated communication should only complement offline communication rather than substituting it, as the resultant online ties are not as strong as face-to-face ties. In an effort to understand the motives in making offline social interactions real and ultimately to predict willingness to engage in serendipitous interactions with people encountered in a public place, we propose a place-aware social matching model driven by interpersonal factors (i.e., similarity, complementarity, and intimacy) and socio-spatial factors (i.e., place sociability, information acquisition expectancy, and perceived personal space in a place). Through a web-based social matching survey experiment (N = 1139 matches from 99 participants in Korea) based on a bogus stranger paradigm, we examine the interrelationship between those factors and the interaction willingness using a series of multiple regression analyses and build a prediction model by devising predictive features based on several machine learning models. From this, we find that both factors have statistically significant influence on interaction willingness, yet interpersonal factors have a higher relative importance than the socio-spatial factors. The interesting point is that the predictive power of these factors varies according to the place characteristics and the level of interaction willingness. We also empirically test the predictability of the model built from the controlled lab experiment through real-world experiments. The results reveal that the proposed model predicts interaction willingness in a real world with under 21% error rate within the Korean cultural context. Findings have implications for the design of mobile social networking systems that endeavor to facilitate serendipitous interactions.
Lee, Dongman; Jang, Si Young; Shin, Byoungheon; Lee, Yoonhyung
Towards dynamically reconfigurable IoT camera virtualization for video analytics edge cloud services Journal Article
In: IEEE Internet Computing, vol. 23, no. 4, pp. 10–17, 2019, ISBN: 1089-7801.
@article{lee2019towards,
title = {Towards dynamically reconfigurable IoT camera virtualization for video analytics edge cloud services},
author = {Dongman Lee and Si Young Jang and Byoungheon Shin and Yoonhyung Lee},
url = {https://ieeexplore.ieee.org/document/8618343},
doi = {10.1109/MIC.2019.2893872},
isbn = {1089-7801},
year = {2019},
date = {2019-01-18},
urldate = {2019-01-18},
journal = {IEEE Internet Computing},
volume = {23},
number = {4},
pages = {10–17},
publisher = {IEEE},
abstract = {Video analytics edge computing exploiting IoT cameras has gained high attention. Running such tasks on the network edge is very challenging since video and image processing are bandwidth-hungry and computationally intensive. IoT cameras are heavily dependent on environmental factors such as the brightness of the view. In this paper, we propose an edge IoT camera virtualization architecture that enables an IoT camera to accommodate multiple application operation semantics and dynamically adjust its configuration to preserve them in the presence of environment context changes. For this, we develop an ontology-based application description model, a virtualization architecture with the container technology, and a context-aware dynamic reconfiguration scheme.
},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Video analytics edge computing exploiting IoT cameras has gained high attention. Running such tasks on the network edge is very challenging since video and image processing are bandwidth-hungry and computationally intensive. IoT cameras are heavily dependent on environmental factors such as the brightness of the view. In this paper, we propose an edge IoT camera virtualization architecture that enables an IoT camera to accommodate multiple application operation semantics and dynamically adjust its configuration to preserve them in the presence of environment context changes. For this, we develop an ontology-based application description model, a virtualization architecture with the container technology, and a context-aware dynamic reconfiguration scheme.
2018
Jang, Si Young; Lee, Yoonhyung; Shin, Byoungheon; Lee, Dongman
Application-aware IoT camera virtualization for video analytics edge computing Conference
2018 IEEE/ACM Symposium on Edge Computing (SEC), IEEE 2018, ISBN: 978-1-5386-9445-9.
@conference{jang2018application,
title = {Application-aware IoT camera virtualization for video analytics edge computing},
author = {Si Young Jang and Yoonhyung Lee and Byoungheon Shin and Dongman Lee},
url = {https://ieeexplore.ieee.org/abstract/document/8567662},
doi = {10.1109/SEC.2018.00017},
isbn = {978-1-5386-9445-9},
year = {2018},
date = {2018-12-09},
urldate = {2018-12-09},
booktitle = {2018 IEEE/ACM Symposium on Edge Computing (SEC)},
pages = {132–144},
organization = {IEEE},
abstract = {Video analytics edge computing exploiting IoT cameras has gained high attention. Running such tasks on the network edge is very challenging since video and image processing are both bandwidth-hungry and computationally intensive. Unlike traditional computing systems, IoT cameras are heavily dependent on the environmental factors such as brightness of the view. In this paper, we propose an edge IoT camera virtualization architecture. For this, we leverage an ontology-based application description model and virtualize the IoT camera with container technology that decouples the physical camera and support multiple applications on board. We also develop an IoT camera reconfiguration scheme that allows IoT cameras to dynamically adjust their configuration to environmental context changes without degrading application QoS. Experimental results based on our prototype implementation show that the responsiveness of our system is 2.8x faster than existing approaches in reconfiguring to the environmental context changes.},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Video analytics edge computing exploiting IoT cameras has gained high attention. Running such tasks on the network edge is very challenging since video and image processing are both bandwidth-hungry and computationally intensive. Unlike traditional computing systems, IoT cameras are heavily dependent on the environmental factors such as brightness of the view. In this paper, we propose an edge IoT camera virtualization architecture. For this, we leverage an ontology-based application description model and virtualize the IoT camera with container technology that decouples the physical camera and support multiple applications on board. We also develop an IoT camera reconfiguration scheme that allows IoT cameras to dynamically adjust their configuration to environmental context changes without degrading application QoS. Experimental results based on our prototype implementation show that the responsiveness of our system is 2.8x faster than existing approaches in reconfiguring to the environmental context changes.
Shin, Byoungheon; Lee, Dongman
An Efficient Local Repair-Based Multi-Constrained Routing for Congestion Control in Wireless Mesh Networks Journal Article
In: Wireless Communications and Mobile Computing, vol. 2018, no. 1, pp. 2893494, 2018.
@article{shin2018efficient,
title = {An Efficient Local Repair-Based Multi-Constrained Routing for Congestion Control in Wireless Mesh Networks},
author = {Byoungheon Shin and Dongman Lee},
url = {https://onlinelibrary.wiley.com/doi/full/10.1155/2018/2893494},
doi = { https://doi.org/10.1155/2018/2893494},
year = {2018},
date = {2018-11-14},
urldate = {2018-11-14},
journal = {Wireless Communications and Mobile Computing},
volume = {2018},
number = {1},
pages = {2893494},
publisher = {Wiley Online Library},
abstract = {Multi-constrained routing is a key driver to support quality-of-service (QoS) for real-time multimedia applications in wireless mesh networks (WMNs). Due to the difficulty of applying strict admission control into a public WMN, it is inevitable to accommodate multiple application flows with different QoS requirements exceeding the capacity of a certain link shared by multiple flows. However, existing multi-constrained routing protocols under such an environment find the QoS degradation based on end-to-end path quality probing and trigger flooding-based route discovery from a scratch for resolving the QoS degradation, which incurs a longer recovery time and much routing overhead. In this paper, we propose a novel multi-constrained routing protocol for WMNs that finds problematic links that may affect QoS degradation to end-to-end paths and replaces them with a detour path using a local repair principle. We model congestion threshold estimation for finding problematic links and design algorithms for quickly finding detour paths and selecting an optimal path by minimizing the negative effect on existing flows nearby the detour path. Simulation results show that the proposed routing protocol achieves up to 19.6% more goodput of live video streaming applications with up to 33% reduced routing overhead compared with an existing work.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Multi-constrained routing is a key driver to support quality-of-service (QoS) for real-time multimedia applications in wireless mesh networks (WMNs). Due to the difficulty of applying strict admission control into a public WMN, it is inevitable to accommodate multiple application flows with different QoS requirements exceeding the capacity of a certain link shared by multiple flows. However, existing multi-constrained routing protocols under such an environment find the QoS degradation based on end-to-end path quality probing and trigger flooding-based route discovery from a scratch for resolving the QoS degradation, which incurs a longer recovery time and much routing overhead. In this paper, we propose a novel multi-constrained routing protocol for WMNs that finds problematic links that may affect QoS degradation to end-to-end paths and replaces them with a detour path using a local repair principle. We model congestion threshold estimation for finding problematic links and design algorithms for quickly finding detour paths and selecting an optimal path by minimizing the negative effect on existing flows nearby the detour path. Simulation results show that the proposed routing protocol achieves up to 19.6% more goodput of live video streaming applications with up to 33% reduced routing overhead compared with an existing work.
Kalra, Gaurav; Yu, Minsang; Lee, Dongman; Cha, Meeyoung; Kim, Daeyoung
Ballparking the urban placeness: A case study of analyzing starbucks posts on instagram Conference
International Conference on Social Informatics, Springer 2018, ISBN: 978-3-030-01129-1.
@conference{kalra2018ballparking,
title = {Ballparking the urban placeness: A case study of analyzing starbucks posts on instagram},
author = {Gaurav Kalra and Minsang Yu and Dongman Lee and Meeyoung Cha and Daeyoung Kim},
url = {https://link.springer.com/chapter/10.1007/978-3-030-01129-1_18},
doi = {https://doi.org/10.1007/978-3-030-01129-1_18},
isbn = {978-3-030-01129-1},
year = {2018},
date = {2018-09-20},
urldate = {2018-01-01},
booktitle = {International Conference on Social Informatics},
pages = {291–307},
organization = {Springer},
abstract = {Placeness or the “sense of a place” plays a vital role in urban design and planning. Research on placeness in the past has been conducted via conventional methods like surveys to reveal essential insights for urban planners and architects. For taking a glimpse into placeness by analyzing common factors across geographies, we choose Instagram posts from Starbucks as a case study, owing to its the-next-door coffee shop psychological construct. We conduct our research by first adopting a flexible ontological framework to organize the concepts governing placeness. Next, we curate a dataset of community generated Instagram posts from Starbucks in three major metropolitan cities of the world: New York, Seoul, and Tokyo. The curated dataset is then automatically enriched with contextual attributes such as activity, visitor demographics, and time via machine learning techniques. We finally analyze and validate the quantitative variations in contextual information with findings from well-accepted cross-cultural case studies. Our results show that placeness mined from Starbucks, a prominent urban third-place, can be reliably utilized to discover surrounding urban placeness.},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Placeness or the “sense of a place” plays a vital role in urban design and planning. Research on placeness in the past has been conducted via conventional methods like surveys to reveal essential insights for urban planners and architects. For taking a glimpse into placeness by analyzing common factors across geographies, we choose Instagram posts from Starbucks as a case study, owing to its the-next-door coffee shop psychological construct. We conduct our research by first adopting a flexible ontological framework to organize the concepts governing placeness. Next, we curate a dataset of community generated Instagram posts from Starbucks in three major metropolitan cities of the world: New York, Seoul, and Tokyo. The curated dataset is then automatically enriched with contextual attributes such as activity, visitor demographics, and time via machine learning techniques. We finally analyze and validate the quantitative variations in contextual information with findings from well-accepted cross-cultural case studies. Our results show that placeness mined from Starbucks, a prominent urban third-place, can be reliably utilized to discover surrounding urban placeness.
Gwak, Bumjin; Cho, Jin-Hee; Lee, Dongman; Son, Heesuk
Taras: Trust-aware role-based access control system in public internet-of-things Conference
2018 17th IEEE International Conference On Trust, Security And Privacy In Computing And Communications/12th IEEE International Conference On Big Data Science And Engineering (TrustCom/BigDataSE), IEEE 2018, ISSN: 2324-9013.
@conference{gwak2018taras,
title = {Taras: Trust-aware role-based access control system in public internet-of-things},
author = {Bumjin Gwak and Jin-Hee Cho and Dongman Lee and Heesuk Son},
url = {https://ieeexplore.ieee.org/abstract/document/8455890},
doi = {10.1109/TrustCom/BigDataSE.2018.00022},
issn = {2324-9013},
year = {2018},
date = {2018-09-06},
urldate = {2018-01-01},
booktitle = {2018 17th IEEE International Conference On Trust, Security And Privacy In Computing And Communications/12th IEEE International Conference On Big Data Science And Engineering (TrustCom/BigDataSE)},
pages = {74–85},
organization = {IEEE},
abstract = {Due to the proliferation of Internet-of-Things (IoT) environments, humans working with heterogeneous, smart objects in public IoT environments become more popular than ever before. This situation often requires to establish trust relationships between a user and a smart object for their secure interactions, but without the presence of prior interactions. In this work, we are interested in how a smart object can grant an access right to a human user in the absence of any prior knowledge in which some users may be malicious aiming to breach security goals of the IoT system. To solve this problem, we propose a trust-aware, role-based access control system, namely TARAS, which provides adaptive authorization to users based on dynamic trust estimation. In TARAS, for the initial trust establishment, we take a multidisciplinary approach by adopting the concept of I-sharing from psychology. The I-sharing follows the rationale that people with similar roles and traits are more likely to respond in a similar way. This theory provides a powerful tool to quickly establish trust between a smart object and a new user with no prior interactions. In addition, TARAS can adaptively filter malicious users out by revoking their access rights based on adaptive, dynamic trust estimation. Our experimental results show that the proposed TARAS mechanism can maximize system integrity in terms of correctly detecting malicious or benign users while maximizing service availability to users particularly when the system is fine-tuned based on the identified optimal setting in terms of an optimal trust threshold.},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Due to the proliferation of Internet-of-Things (IoT) environments, humans working with heterogeneous, smart objects in public IoT environments become more popular than ever before. This situation often requires to establish trust relationships between a user and a smart object for their secure interactions, but without the presence of prior interactions. In this work, we are interested in how a smart object can grant an access right to a human user in the absence of any prior knowledge in which some users may be malicious aiming to breach security goals of the IoT system. To solve this problem, we propose a trust-aware, role-based access control system, namely TARAS, which provides adaptive authorization to users based on dynamic trust estimation. In TARAS, for the initial trust establishment, we take a multidisciplinary approach by adopting the concept of I-sharing from psychology. The I-sharing follows the rationale that people with similar roles and traits are more likely to respond in a similar way. This theory provides a powerful tool to quickly establish trust between a smart object and a new user with no prior interactions. In addition, TARAS can adaptively filter malicious users out by revoking their access rights based on adaptive, dynamic trust estimation. Our experimental results show that the proposed TARAS mechanism can maximize system integrity in terms of correctly detecting malicious or benign users while maximizing service availability to users particularly when the system is fine-tuned based on the identified optimal setting in terms of an optimal trust threshold.