Hyunggon Park
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Hyunggon Park has a total of 26 co-authors. Here are the top co-authors by number of shared publications:
- Minhae Kwon
Department of Electrical and Computer Engineering, Sungkyunkwan University, Suwon, Republic of Korea
10 shared publications - Chaeyeon Cha
Department of Electronic and Electrical Engineering, Ewha Womans University, Seoul, South Korea
5 shared publications - Joohong Rheey
Department of Electronic and Electrical Engineering, Ewha Womans University, Seoul, South Korea
5 shared publications - Dayoung Choi
Smart Factory Multidisciplinary Program, Department of Electronic and Electrical Engineering College of Engineering, Ewha Womans University Seoul Republic of Korea
4 shared publications Jungmin Kwon
Department of Electrical and Electronics Engineering, Kangwon National University, Gangwon-do, Republic of Korea
3 shared publications- Eunhye Choi
Agency for Defense Development, Daejeon, South Korea
2 shared publications
Latest Publications
- 11 September 2026
- Published by Elsevier in Computer Networks
- Vol. 290
Minimal-Feature Machine Learning for Scalable AI-RAN under Communication Constraints
- 11 September 2026
- Published by Elsevier in Computer Networks
- Vol. 290
Minimal-Feature Machine Learning for Scalable AI-RAN under Communication Constraints
- 15 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 33 (10709908) , 2505-2509
Post-Hoc Wasserstein Certificates for Autoencoder Reconstructions
Assessment of autoencoders is frequently constrained to average empirical losses, which lack principled and distributional guarantees. Existing methods incorporating optimal transport are training-centric and fail to provide post-hoc certificates for a trained model. In this paper, we derive a high-probability, data-dependent upper bound on the Wasserstein-1 distance between the input and reconstruction distributions. The bound is computed directly from the empirical mean absolute error on a hold-out set, requiring neither retraining nor the use of an optimal transport solver. We provide an explicit finite-sample deviation radius scaling. Moreover, we introduce an analytic attribution framework based on a joint-Gaussian surrogate, which provides interpretable insight into the structure of the reconstruction error. This framework decomposes the expected reconstruction loss into bias, variance, and alignment components in closed form. This enables an auditable procedure for both certifying and diagnosing autoencoder reconstructions.
- 15 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 33 (10709908) , 2505-2509
Post-Hoc Wasserstein Certificates for Autoencoder Reconstructions
Assessment of autoencoders is frequently constrained to average empirical losses, which lack principled and distributional guarantees. Existing methods incorporating optimal transport are training-centric and fail to provide post-hoc certificates for a trained model. In this paper, we derive a high-probability, data-dependent upper bound on the Wasserstein-1 distance between the input and reconstruction distributions. The bound is computed directly from the empirical mean absolute error on a hold-out set, requiring neither retraining nor the use of an optimal transport solver. We provide an explicit finite-sample deviation radius scaling. Moreover, we introduce an analytic attribution framework based on a joint-Gaussian surrogate, which provides interpretable insight into the structure of the reconstruction error. This framework decomposes the expected reconstruction loss into bias, variance, and alignment components in closed form. This enables an auditable procedure for both certifying and diagnosing autoencoder reconstructions.
- Open Access
- 29 May 2026
- Published by Wiley in Journal of Advanced Nursing
- Vol. 82 (9) , 8682-8690
Shift‐Specific Patterns of Nursing Workloads in the Emergency Department: AI Powered Analysis
Aim: To identify and differentiate workload patterns across shifts and to provide evidence for optimizing nursing workforce allocation in emergency departments: Design: A cross‐sectional study. Methods: Real time data were collected from an emergency department in a general hospital in Seoul, South Korea, between October 30, 2023 to October 24, 2024. Smartphones, beacons, and smartwatches were used to capture nursing time, physical activity, work‐related characteristics, and location transitions across 238 shifts. A multiclass eXtreme Gradient Boosting model was developed and evaluated to classify working shifts (day, evening, night). Shapely Additive exPlanations were applied to identify key contributing features, and shift‐specific differences were examined using analysis of variance with post hoc tests. Results: The model demonstrated strong performance in distinguishing shifts. Key features included the number of admissions, discharges, assigned patients, and both direct and indirect nursing time, all of which varied across shifts. In contrast, location transition patterns were relatively consistent. Conclusion: Shift‐specific nursing workloads in emergency departments can be effectively identified using multidimensional, real‐world nursing activity data. Implications for the Profession and/or Patient Care: Findings support the development of staffing strategies that account for variation in workload across shifts, with potential to improve efficiency and maintain quality of care. Impact: This study addresses the lack of objective evidence for shift‐specific workload differences in emergency nursing. It demonstrates that multidimensional activity data can distinguish workload patterns across shifts. The findings may inform staffing decisions for emergency department nurses and support improvements in workforce management and patient care. Reporting Method: This study adhered to the Strengthening the Reporting of Observational Studies in Epidemiology guidelines. Patient or Public Contribution: No patients or members of the public were involved in the design, conduct, analysis, or reporting of this study.
- 29 May 2026
- Published by Wiley in Journal of Advanced Nursing
- Vol. 82 (9) , 8682-8690
Open AccessShift‐Specific Patterns of Nursing Workloads in the Emergency Department: AI Powered Analysis
Aim: To identify and differentiate workload patterns across shifts and to provide evidence for optimizing nursing workforce allocation in emergency departments: Design: A cross‐sectional study. Methods: Real time data were collected from an emergency department in a general hospital in Seoul, South Korea, between October 30, 2023 to October 24, 2024. Smartphones, beacons, and smartwatches were used to capture nursing time, physical activity, work‐related characteristics, and location transitions across 238 shifts. A multiclass eXtreme Gradient Boosting model was developed and evaluated to classify working shifts (day, evening, night). Shapely Additive exPlanations were applied to identify key contributing features, and shift‐specific differences were examined using analysis of variance with post hoc tests. Results: The model demonstrated strong performance in distinguishing shifts. Key features included the number of admissions, discharges, assigned patients, and both direct and indirect nursing time, all of which varied across shifts. In contrast, location transition patterns were relatively consistent. Conclusion: Shift‐specific nursing workloads in emergency departments can be effectively identified using multidimensional, real‐world nursing activity data. Implications for the Profession and/or Patient Care: Findings support the development of staffing strategies that account for variation in workload across shifts, with potential to improve efficiency and maintain quality of care. Impact: This study addresses the lack of objective evidence for shift‐specific workload differences in emergency nursing. It demonstrates that multidimensional activity data can distinguish workload patterns across shifts. The findings may inform staffing decisions for emergency department nurses and support improvements in workforce management and patient care. Reporting Method: This study adhered to the Strengthening the Reporting of Observational Studies in Epidemiology guidelines. Patient or Public Contribution: No patients or members of the public were involved in the design, conduct, analysis, or reporting of this study.
- 18 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. PP (1530437X) , 1
Lightweight Cooperative Anomaly Detection Framework for Multi-Sensor Edge Systems
Edge-based anomaly detection has become increasingly important in Industrial IoT (IIoT) systems due to stringent latency, privacy, and connectivity requirements. However, edge devices such as IoT sensors operate under severe computational and memory constraints, making lightweight models essential to preserve detection performance under limited resources. This paper presents an anomaly detection framework that achieves robust multi-sensor cooperation by integrating multiple lightweight sensor models using entropy-based adaptive weighting. The weighting coefficient for each sensor is analytically derived from the entropy of its reconstruction error distribution, providing a principled, information-theoretic measure of sensor contribution without requiring raw data sharing, label supervision, or auxiliary meta-models. Experiments on an industrial acoustic benchmark show that the proposed weighting scheme improves key detection metrics—particularly recall—while maintaining robust performance across spatially heterogeneous sensing conditions. Additional analysis reveals that cooperation is most beneficial when sensors exhibit complementary error behaviors. Finally, a full implementation on Raspberry Pi 4 sensor nodes demonstrates that the method operates efficiently on low-power edge hardware, confirming its suitability for practical IIoT deployments.
- 18 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. PP (1530437X) , 1
Lightweight Cooperative Anomaly Detection Framework for Multi-Sensor Edge Systems
Edge-based anomaly detection has become increasingly important in Industrial IoT (IIoT) systems due to stringent latency, privacy, and connectivity requirements. However, edge devices such as IoT sensors operate under severe computational and memory constraints, making lightweight models essential to preserve detection performance under limited resources. This paper presents an anomaly detection framework that achieves robust multi-sensor cooperation by integrating multiple lightweight sensor models using entropy-based adaptive weighting. The weighting coefficient for each sensor is analytically derived from the entropy of its reconstruction error distribution, providing a principled, information-theoretic measure of sensor contribution without requiring raw data sharing, label supervision, or auxiliary meta-models. Experiments on an industrial acoustic benchmark show that the proposed weighting scheme improves key detection metrics—particularly recall—while maintaining robust performance across spatially heterogeneous sensing conditions. Additional analysis reveals that cooperation is most beneficial when sensors exhibit complementary error behaviors. Finally, a full implementation on Raspberry Pi 4 sensor nodes demonstrates that the method operates efficiently on low-power edge hardware, confirming its suitability for practical IIoT deployments.
- 24 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 25 (5) , 6413-6428
Efficient and Resilient Packet Recovery for Federated Learning via Approximation
To alleviate network congestion resulting from packet retransmission in Federated Learning (FL) systems, the User Datagram Protocol (UDP) has been adopted. However, data loss under UDP inevitably degrades FL performance, as learning is sensitive to incomplete model parameters. This paper addresses such degradation by proposing a low-rank approximation–based parameter transmission method for FL. This approach decomposes model parameters at the transmitter to extract dominant singular values, which are essential for accurate approximation of the model parameters at the receiver. Although the selective delivery of singular values and vectors reduces communication overhead, their loss would cause severe performance degradation, making it essential to employ protection mechanisms. Therefore, to protect the extracted singular values and vectors, we use Systematic Network Coding (SysNC). The SysNC with low-rank approximation can recover original information under a packet loss environment, enhancing robustness against partial loss during the transmission of model parameters. Theoretical analysis and experiments confirm its effectiveness. As an example from our experiments, with a packet loss ratio of$p=0.1$ , the proposed method achieves over 96% of the accuracy observed in the packet loss-free case, while reducing the end-to-end delay for model parameter transmission by approximately 50% compared to a UDP baseline.- 24 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 25 (5) , 6413-6428
Efficient and Resilient Packet Recovery for Federated Learning via Approximation
To alleviate network congestion resulting from packet retransmission in Federated Learning (FL) systems, the User Datagram Protocol (UDP) has been adopted. However, data loss under UDP inevitably degrades FL performance, as learning is sensitive to incomplete model parameters. This paper addresses such degradation by proposing a low-rank approximation–based parameter transmission method for FL. This approach decomposes model parameters at the transmitter to extract dominant singular values, which are essential for accurate approximation of the model parameters at the receiver. Although the selective delivery of singular values and vectors reduces communication overhead, their loss would cause severe performance degradation, making it essential to employ protection mechanisms. Therefore, to protect the extracted singular values and vectors, we use Systematic Network Coding (SysNC). The SysNC with low-rank approximation can recover original information under a packet loss environment, enhancing robustness against partial loss during the transmission of model parameters. Theoretical analysis and experiments confirm its effectiveness. As an example from our experiments, with a packet loss ratio of$p=0.1$ , the proposed method achieves over 96% of the accuracy observed in the packet loss-free case, while reducing the end-to-end delay for model parameter transmission by approximately 50% compared to a UDP baseline.- 29 May 2025
- Published by Elsevier in Computers & Security
- Vol. 157
Attack-specific feature analysis framework for NetFlow IoT datasets
- 29 May 2025
- Published by Elsevier in Computers & Security
- Vol. 157
Attack-specific feature analysis framework for NetFlow IoT datasets
- Open Access
- 29 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 75666-75678
SV-SAE: Layer-Wise Pruning for Autoencoder Based on Link Contributions
Autoencoders are a type of deep neural network and are widely used for unsupervised learning, particularly in tasks that require feature extraction and dimensionality reduction. While most research focuses on compressing input data, less attention has been given to reducing the size and complexity of the autoencoder model itself, which is crucial for deployment on resource-constrained edge devices. This paper introduces a layer-wise pruning algorithm specifically for multilayer perceptron-based autoencoder. The resulting pruned model is referred to as a Shapley Value-based Sparse AutoEncoder (SV-SAE). Using cooperative game theory, the proposed algorithm models the autoencoder as a coalition of interconnected units and links, where the Shapley value quantifies their individual contributions to overall performance. This enables the selective removal of less important components, achieving an optimal balance between sparsity and accuracy. Experimental results confirm that the SV-SAE reaches an accuracy of 99.25%, utilizing only 10% of the original links. Notably, the SV-SAE remains robust under high sparsity levels with minimal performance degradation whereas other algorithms experience sharp declines as the pruning ratio increases. Designed for edge environments, the SV-SAE offers an interpretable framework for controlling layer-wise sparsity while preserving essential features in latent representations. The results highlight its potential for efficient deployment in resource-constrained scenarios, where model size and inference speed are critical factors.
- 29 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 75666-75678
Open AccessSV-SAE: Layer-Wise Pruning for Autoencoder Based on Link Contributions
Autoencoders are a type of deep neural network and are widely used for unsupervised learning, particularly in tasks that require feature extraction and dimensionality reduction. While most research focuses on compressing input data, less attention has been given to reducing the size and complexity of the autoencoder model itself, which is crucial for deployment on resource-constrained edge devices. This paper introduces a layer-wise pruning algorithm specifically for multilayer perceptron-based autoencoder. The resulting pruned model is referred to as a Shapley Value-based Sparse AutoEncoder (SV-SAE). Using cooperative game theory, the proposed algorithm models the autoencoder as a coalition of interconnected units and links, where the Shapley value quantifies their individual contributions to overall performance. This enables the selective removal of less important components, achieving an optimal balance between sparsity and accuracy. Experimental results confirm that the SV-SAE reaches an accuracy of 99.25%, utilizing only 10% of the original links. Notably, the SV-SAE remains robust under high sparsity levels with minimal performance degradation whereas other algorithms experience sharp declines as the pruning ratio increases. Designed for edge environments, the SV-SAE offers an interpretable framework for controlling layer-wise sparsity while preserving essential features in latent representations. The results highlight its potential for efficient deployment in resource-constrained scenarios, where model size and inference speed are critical factors.
- 4 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (10) , 18529-18542
Lightweight Machine Sound Anomaly Detector Based on Parallel Discrete Wavelet Transform
The development of Industrial Internet of Things (IIoT) technology and network infrastructures has enabled the acquisition of substantial data, enabling data-driven condition monitoring and analysis. Detecting anomalies in machinery equipment is crucial in IIoT environments for safety enhancement, productivity, and reliability. To provide effective anomaly detection at IIoT edge nodes without delay, it is necessary to efficiently collect and process vast amounts of data from various sensors. While this demands a significant amount of computing resources, edge nodes only have limited data storage and processing capabilities. Therefore, our focus is on developing a lightweight anomaly detection algorithm for acoustic signal processing, considering the computational resources of the IIoT edge node. In this paper, we propose the parallel discrete wavelet transform as an efficient method for compressing and processing acoustic signals received at edge nodes. This approach significantly alleviates memory consumption and reduces the computational time at the edge. In addition, by harnessing preprocessed features through parallel discrete wavelet transform, we can develop lightweight anomaly detection models suitable for deployment at the edge, making them highly practical for real-world implementation. The experimental results using real-world data collected from industrial machines confirm the effectiveness of the proposed solution.
- 4 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (10) , 18529-18542
Lightweight Machine Sound Anomaly Detector Based on Parallel Discrete Wavelet Transform
The development of Industrial Internet of Things (IIoT) technology and network infrastructures has enabled the acquisition of substantial data, enabling data-driven condition monitoring and analysis. Detecting anomalies in machinery equipment is crucial in IIoT environments for safety enhancement, productivity, and reliability. To provide effective anomaly detection at IIoT edge nodes without delay, it is necessary to efficiently collect and process vast amounts of data from various sensors. While this demands a significant amount of computing resources, edge nodes only have limited data storage and processing capabilities. Therefore, our focus is on developing a lightweight anomaly detection algorithm for acoustic signal processing, considering the computational resources of the IIoT edge node. In this paper, we propose the parallel discrete wavelet transform as an efficient method for compressing and processing acoustic signals received at edge nodes. This approach significantly alleviates memory consumption and reduces the computational time at the edge. In addition, by harnessing preprocessed features through parallel discrete wavelet transform, we can develop lightweight anomaly detection models suitable for deployment at the edge, making them highly practical for real-world implementation. The experimental results using real-world data collected from industrial machines confirm the effectiveness of the proposed solution.
- 27 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 32 (10709908) , 1500-1504
Alternating Offer-Based Payment Allocation for Privacy Non-Disclosure in Federated Learning
In federated learning (FL), it is essential to implement a payment allocation mechanism that compensates clients for the costs incurred from participating in FL tasks. In this letter, we formulate the payment allocation as a bargaining game between a global server and clients and adopt the Nash bargaining solution (NBS) to achieve optimal and fair payment assignments among clients. Unlike existing payment allocation mechanisms that require the disclosure of private information from the clients, the proposed approach ensures privacy non-disclosure for bargaining. The key idea is to decompose the one-to-many bargaining game into independent one-to-one bargaining games and use alternating-offers, which do not require the disclosure of private information from clients. We design an alternating-offers strategy and acceptance criteria to ensure fair agreements without the private information of clients. Simulation results show that the proposed payment allocation strategy can fairly allocate payments to clients while maintaining the accuracy of the global server in FL tasks.
- 27 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 32 (10709908) , 1500-1504
Alternating Offer-Based Payment Allocation for Privacy Non-Disclosure in Federated Learning
In federated learning (FL), it is essential to implement a payment allocation mechanism that compensates clients for the costs incurred from participating in FL tasks. In this letter, we formulate the payment allocation as a bargaining game between a global server and clients and adopt the Nash bargaining solution (NBS) to achieve optimal and fair payment assignments among clients. Unlike existing payment allocation mechanisms that require the disclosure of private information from the clients, the proposed approach ensures privacy non-disclosure for bargaining. The key idea is to decompose the one-to-many bargaining game into independent one-to-one bargaining games and use alternating-offers, which do not require the disclosure of private information from clients. We design an alternating-offers strategy and acceptance criteria to ensure fair agreements without the private information of clients. Simulation results show that the proposed payment allocation strategy can fairly allocate payments to clients while maintaining the accuracy of the global server in FL tasks.