Hyunseung Choo
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Hyunseung Choo has a total of 189 co-authors. Here are the top co-authors by number of shared publications:
- Duc-Tai Le
Sungkyunkwan University
37 shared publications - Moonseong Kim
Department of IT Convergence Software, Seoul Theological University, Bucheon, South Korea
19 shared publications - Syed M. Raza
James Watt School of Engineering, University of Glasgow, Glasgow, U.K.
18 shared publications - Junghyun Bum
Sungkyunkwan University
14 shared publications Duc-Tai Le
Sungkyunkwan University
13 shared publications- Byungseok Kang
Department of Computer Science and Engineering, Sungkyunkwan University, Suwon, Korea
9 shared publications
Latest Publications
- 16 November 2026
- Published by Elsevier in Ad Hoc Networks
- Vol. 196
Receiver-seed guided hybrid Q-learning for low-latency broadcast scheduling in multi-hop wireless networks
- 16 November 2026
- Published by Elsevier in Ad Hoc Networks
- Vol. 196
Receiver-seed guided hybrid Q-learning for low-latency broadcast scheduling in multi-hop wireless networks
- 25 September 2026
- Published by Springer Nature
- p. 118-127
Collaborative Multiscale Representation Learning for AMD Detection
- 25 September 2026
- Published by Springer Nature
- p. 118-127
Collaborative Multiscale Representation Learning for AMD Detection
- 24 September 2026
- Published by Springer Nature
- p. 523-533
Response-Aware Multimodal Learning for Post-treatment Visual Acuity Forecasting
- 24 September 2026
- Published by Springer Nature
- p. 523-533
Response-Aware Multimodal Learning for Post-treatment Visual Acuity Forecasting
- 4 August 2026
- Published by Springer Nature
- p. 447-462
Representation Learning with Semantic-Aware Instance and Sparse Token Alignments
- 4 August 2026
- Published by Springer Nature
- p. 447-462
Representation Learning with Semantic-Aware Instance and Sparse Token Alignments
- 3 August 2026
- Published by Springer Nature
- p. 355-370
Task-Aware Feature Modulation in Heterogeneous Multitask Learning for Fundus Landmark Extraction
- 3 August 2026
- Published by Springer Nature
- p. 355-370
Task-Aware Feature Modulation in Heterogeneous Multitask Learning for Fundus Landmark Extraction
- 25 July 2026
- Published by Springer Nature
- p. 689-703
Fundus to Cardiovascular Risk Factors with Anthropometric Guidance
- 25 July 2026
- Published by Springer Nature
- p. 689-703
Fundus to Cardiovascular Risk Factors with Anthropometric Guidance
- Open Access
- 9 July 2026
- Published by Elsevier in SSRN Electronic Journal
Receiver-SeedGuidedHybridQ-LearningforLow-LatencyBroadcastSchedulinginMulti-HopWirelessNetworks
Efficient broadcast scheduling is a fundamental operation in Wireless Sensor Networks ( WSNs ) for disseminating control messages, software updates, and routing information. However, minimizing network-wide broadcast latency while effectively managing wireless interference and collisions is a known NP-hard problem. Existing heuristic and deterministic approaches often converge to local optima, resulting in sub-optimal transmission delays. In this paper, we propose a novel receiver-seed guided hybrid reinforcement learning ( RL ) -based broadcast scheduling scheme that leverages Q-learning and heuristic sender-receiver matching method to minimize broadcast latency. We formulate the scheduling process as a Markov decision process, where the state represents the progressively expanding set of covered nodes. At each time slot, the RL agent selects a seed action, which is expanded by a collision-aware greedy mechanism into a maximal feasible set of concurrent transmissions. To mitigate the Q-table explosion, we propose an action aggregation method that measures the equivalence of seed actions given a network state, which helps in both reducing redundant Q-table entries and policy convergence. We further accelerate reward propagation along scheduling sequence using the eligibility traces mechanism to immediately reinforce best found scheduling sequence reward. Extensive simulations across diverse network sizes and topologies show that the proposed Q-learning framework achieves lower broadcast delay and higher convergence reliability than benchmarking schemes.
- 9 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessReceiver-SeedGuidedHybridQ-LearningforLow-LatencyBroadcastSchedulinginMulti-HopWirelessNetworks
Efficient broadcast scheduling is a fundamental operation in Wireless Sensor Networks ( WSNs ) for disseminating control messages, software updates, and routing information. However, minimizing network-wide broadcast latency while effectively managing wireless interference and collisions is a known NP-hard problem. Existing heuristic and deterministic approaches often converge to local optima, resulting in sub-optimal transmission delays. In this paper, we propose a novel receiver-seed guided hybrid reinforcement learning ( RL ) -based broadcast scheduling scheme that leverages Q-learning and heuristic sender-receiver matching method to minimize broadcast latency. We formulate the scheduling process as a Markov decision process, where the state represents the progressively expanding set of covered nodes. At each time slot, the RL agent selects a seed action, which is expanded by a collision-aware greedy mechanism into a maximal feasible set of concurrent transmissions. To mitigate the Q-table explosion, we propose an action aggregation method that measures the equivalence of seed actions given a network state, which helps in both reducing redundant Q-table entries and policy convergence. We further accelerate reward propagation along scheduling sequence using the eligibility traces mechanism to immediately reinforce best found scheduling sequence reward. Extensive simulations across diverse network sizes and topologies show that the proposed Q-learning framework achieves lower broadcast delay and higher convergence reliability than benchmarking schemes.
- 22 May 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 329
TIAMO: Text-image-audio multimodal model for respiratory sound classification
- 22 May 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 329
TIAMO: Text-image-audio multimodal model for respiratory sound classification
- Open Access
- 22 May 2026
- Published by Springer Nature in Medical & Biological Engineering & Computing
- Vol. 64 (6) , 2405-2419
Dual sagittal-guided coarse-to-fine rib segmentation and numbering in chest CT
Rib identification in chest CT is time-consuming and labor-intensive, as clinicians repetitively examine hundreds of CT slices to pinpoint a certain rib on a random slice. Computer-aided rib numbering tools exist, however, their performance hinges on precise rib segmentation which is challenged by missing ribs or oversegmentation of adjacent bone structures. We propose Dual sagittal-guided Coarse-to-fine Rib Segmentation (DCRS), a lightweight and practical 3D rib segmentation approach that coarsely captures the entire rib structure within 2 sagittal slices and finely constructs 3D ribs. DCRS operates in three stages: (1) Region of Non-Interest (RONI) removal, a preprocessing stage which excludes the soft tissues, spine, sternum, and transverse processes to isolate ribs; (2) a lightweight U-Net applied to only two representative sagittal slices for obtaining coarse sagittal ribs; and (3) Six-Neighborhood Outward Flood-Filling (SNOFF), a fine 3D rib construction that expands sagittal rib predictions into full 3D ribs. This coarse-to-fine approach ensures a complete 12-pair rib mask while decreasing the complexity and computational cost compared to coarse slice, patch, or point-cloud-wise rib segmentation methods. The training and testing of the proposed DCRS utilize a public rib segmentation dataset. DCRS achieves a Dice score of 92.83% and IoU of 88.21%, exceeding prior state-of-the-art by 3.1% and 4.1%, respectively. DCRS enables fast and reliable rib numbering, increasing throughput by 44% compared to the prior state-of-the-art. Rib numbering takes only 0.5% additional time relative to standard CT, highlighting its clinical practicality.
- 22 May 2026
- Published by Springer Nature in Medical & Biological Engineering & Computing
- Vol. 64 (6) , 2405-2419
Open AccessDual sagittal-guided coarse-to-fine rib segmentation and numbering in chest CT
Rib identification in chest CT is time-consuming and labor-intensive, as clinicians repetitively examine hundreds of CT slices to pinpoint a certain rib on a random slice. Computer-aided rib numbering tools exist, however, their performance hinges on precise rib segmentation which is challenged by missing ribs or oversegmentation of adjacent bone structures. We propose Dual sagittal-guided Coarse-to-fine Rib Segmentation (DCRS), a lightweight and practical 3D rib segmentation approach that coarsely captures the entire rib structure within 2 sagittal slices and finely constructs 3D ribs. DCRS operates in three stages: (1) Region of Non-Interest (RONI) removal, a preprocessing stage which excludes the soft tissues, spine, sternum, and transverse processes to isolate ribs; (2) a lightweight U-Net applied to only two representative sagittal slices for obtaining coarse sagittal ribs; and (3) Six-Neighborhood Outward Flood-Filling (SNOFF), a fine 3D rib construction that expands sagittal rib predictions into full 3D ribs. This coarse-to-fine approach ensures a complete 12-pair rib mask while decreasing the complexity and computational cost compared to coarse slice, patch, or point-cloud-wise rib segmentation methods. The training and testing of the proposed DCRS utilize a public rib segmentation dataset. DCRS achieves a Dice score of 92.83% and IoU of 88.21%, exceeding prior state-of-the-art by 3.1% and 4.1%, respectively. DCRS enables fast and reliable rib numbering, increasing throughput by 44% compared to the prior state-of-the-art. Rib numbering takes only 0.5% additional time relative to standard CT, highlighting its clinical practicality.
- 24 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Cognitive Communications and Networking
- Vol. 12, 7001-7016
Curated Collaborative AI Edge With Network Data Analytics for B5G/6G Radio Access Networks
Despite advancements, Radio Access Networks (RAN) still account for over 50% of the total power consumption in 5G networks. Existing RAN split options do not fully harness data potential, presenting an opportunity to reduce operational expenditures. This paper aims to reduce energy consumption and operational costs in 5G RANs by effectively leveraging network data through two key contributions. First, a Curated Collaborative Learning (CCL) framework that ensures selective collaboration among correlated data sources for accurate traffic and user count forecasting. CCL optimally determines whom, when, and what to collaborate with, significantly outperforming state-of-the-art approaches, including global, federated, personalized federated, and cyclic institutional incremental learnings by 43.9%, 39.1%, 40.8%, and 31.35%, respectively. Second, the Distributed Unit Pooling Scheme (DUPS) is proposed, leveraging deep reinforcement learning and prediction inferences from CCL to reduce the number of active DU servers efficiently. DUPS dynamically redirects traffic from underutilized DU servers to optimize resource utilization. This improves energy efficiency by up to 89% over conventional strategies, where all DU servers remain turned on regardless of load, thereby translating into substantial monetary benefits for operators. By integrating CCL-driven predictions with DUPS, this paper demonstrates a transformative approach for minimizing energy consumption and operational costs in next-generation mobile networks, significantly enhancing efficiency and cost-effectiveness.
- 24 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Cognitive Communications and Networking
- Vol. 12, 7001-7016
Curated Collaborative AI Edge With Network Data Analytics for B5G/6G Radio Access Networks
Despite advancements, Radio Access Networks (RAN) still account for over 50% of the total power consumption in 5G networks. Existing RAN split options do not fully harness data potential, presenting an opportunity to reduce operational expenditures. This paper aims to reduce energy consumption and operational costs in 5G RANs by effectively leveraging network data through two key contributions. First, a Curated Collaborative Learning (CCL) framework that ensures selective collaboration among correlated data sources for accurate traffic and user count forecasting. CCL optimally determines whom, when, and what to collaborate with, significantly outperforming state-of-the-art approaches, including global, federated, personalized federated, and cyclic institutional incremental learnings by 43.9%, 39.1%, 40.8%, and 31.35%, respectively. Second, the Distributed Unit Pooling Scheme (DUPS) is proposed, leveraging deep reinforcement learning and prediction inferences from CCL to reduce the number of active DU servers efficiently. DUPS dynamically redirects traffic from underutilized DU servers to optimize resource utilization. This improves energy efficiency by up to 89% over conventional strategies, where all DU servers remain turned on regardless of load, thereby translating into substantial monetary benefits for operators. By integrating CCL-driven predictions with DUPS, this paper demonstrates a transformative approach for minimizing energy consumption and operational costs in next-generation mobile networks, significantly enhancing efficiency and cost-effectiveness.