Sungwon Yi
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Sungwon Yi has a total of 42 co-authors. Here are the top co-authors by number of shared publications:
Xidong Deng
Department of Electrical and Computer Engineering, St. Cloud State University, Minnesota, U.S.A.
6 shared publicationsC.R. Das
Pennsylvania State University
5 shared publicationsG. Kesidis
Pennsylvania State University
5 shared publications- Byunghyun Yoo
Electronics and Telecommunications Research Institute (ETRI) Daejeon South Korea
2 shared publications - Chita R. Das
The Pennsylvania State University
2 shared publications George Kesidis
Pennsylvania State University
2 shared publications
Latest Publications
- 2 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 23, 2051-2063
Joint Multi-Agent Reinforcement Learning and Message-Passing for Resilient Multi-UAV Networks
This paper introduces a novel resilient algorithm designed for distributed unmanned aerial vehicles (UAVs) in dynamic and unreliable network environments. Initially, the UAVs should be trained via multi-agent reinforcement learning (MARL) for autonomous mission-critical operations and are fundamentally grounded by centralized training and decentralized execution (CTDE) using a centralized MARL server. In this situation, it is crucial to consider the case where several UAVs cannot receive CTDE-based MARL learning parameters for resilient operations in unreliable network conditions. To tackle this issue, a communication graph is used where its edges are established when two UAVs/nodes are communicable. Then, the edge-connected UAVs can share their training data if one of the UAVs cannot be connected to the CTDE-based MARL server under unreliable network conditions. Additionally, the edge cost considers power efficiency. Based on this given communication graph, message-passing is used for electing the UAVs that can provide their MARL learning parameters to their edge-connected peers. Lastly, performance evaluations demonstrate the superiority of our proposed algorithm in terms of power efficiency and resilient UAV task management, outperforming existing benchmark algorithms.
- 2 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 23, 2051-2063
Joint Multi-Agent Reinforcement Learning and Message-Passing for Resilient Multi-UAV Networks
This paper introduces a novel resilient algorithm designed for distributed unmanned aerial vehicles (UAVs) in dynamic and unreliable network environments. Initially, the UAVs should be trained via multi-agent reinforcement learning (MARL) for autonomous mission-critical operations and are fundamentally grounded by centralized training and decentralized execution (CTDE) using a centralized MARL server. In this situation, it is crucial to consider the case where several UAVs cannot receive CTDE-based MARL learning parameters for resilient operations in unreliable network conditions. To tackle this issue, a communication graph is used where its edges are established when two UAVs/nodes are communicable. Then, the edge-connected UAVs can share their training data if one of the UAVs cannot be connected to the CTDE-based MARL server under unreliable network conditions. Additionally, the edge cost considers power efficiency. Based on this given communication graph, message-passing is used for electing the UAVs that can provide their MARL learning parameters to their edge-connected peers. Lastly, performance evaluations demonstrate the superiority of our proposed algorithm in terms of power efficiency and resilient UAV task management, outperforming existing benchmark algorithms.
- Open Access
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- p. 1
GraDual: Resolving the Semantic–Frequency Trade-off via Gradient-Regulated Dual-Branch Multimodal Manipulation Detection
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- p. 1
Open AccessGraDual: Resolving the Semantic–Frequency Trade-off via Gradient-Regulated Dual-Branch Multimodal Manipulation Detection
- Open Access
- 24 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 35674-35687
SA-MARL: Novel Self-Attention-Based Multi-Agent Reinforcement Learning With Stochastic Gradient Descent
In the rapidly advancing Reinforcement Learning (RL) field, Multi-Agent Reinforcement Learning (MARL) has emerged as a key player in solving complex real-world challenges. A pivotal development in this realm is the introduction of the mixing network, representing a significant leap forward in the capabilities of multi-agent systems. Drawing inspiration from COMA and VDN methodologies, the mixing network overcomes limitations in extracting combined Q-values from joint state-action interactions. Previous approaches like COMA and VDN faced constraints in fully utilising the state-provided information during training, limiting their effectiveness. QMIX and QVMinMax addressed this issue by employing neural networks to convert centralized states into weights for a second neural network, akin to hyper-networks. However, these solutions presented challenges such as computational intensity and susceptibility to local minima. To overcome these hurdles, our proposed methodology introduces three key contributions. First, we introduce the state- fusion network, an innovative alternative to traditional mixing, with a self-attention mechanism . Second, to address the local optima problem in MARL algorithms, we leverage the Grey Wolf Optimizer for weight and bias selection, adding a stochastic element for improved optimization. Finally, we comprehensively compare with QMIX, evaluating performance under two optimization methods: Gradient Descent and Stochastic Optimizer. Using the StarCraft II Learning Environment (SC2LE) as our experimental platform, our results demonstrate the superiority of our methodology over QMIX, QVMinMax, and QSOD in absolute performance, particularly when operating under resource constraints. Our proposed methodology contributes to the ongoing evolution of MARL techniques, showcasing advancements in attention mechanisms and optimization strategies for enhanced multi-agent system capabilities.
- 24 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 35674-35687
Open AccessSA-MARL: Novel Self-Attention-Based Multi-Agent Reinforcement Learning With Stochastic Gradient Descent
In the rapidly advancing Reinforcement Learning (RL) field, Multi-Agent Reinforcement Learning (MARL) has emerged as a key player in solving complex real-world challenges. A pivotal development in this realm is the introduction of the mixing network, representing a significant leap forward in the capabilities of multi-agent systems. Drawing inspiration from COMA and VDN methodologies, the mixing network overcomes limitations in extracting combined Q-values from joint state-action interactions. Previous approaches like COMA and VDN faced constraints in fully utilising the state-provided information during training, limiting their effectiveness. QMIX and QVMinMax addressed this issue by employing neural networks to convert centralized states into weights for a second neural network, akin to hyper-networks. However, these solutions presented challenges such as computational intensity and susceptibility to local minima. To overcome these hurdles, our proposed methodology introduces three key contributions. First, we introduce the state- fusion network, an innovative alternative to traditional mixing, with a self-attention mechanism . Second, to address the local optima problem in MARL algorithms, we leverage the Grey Wolf Optimizer for weight and bias selection, adding a stochastic element for improved optimization. Finally, we comprehensively compare with QMIX, evaluating performance under two optimization methods: Gradient Descent and Stochastic Optimizer. Using the StarCraft II Learning Environment (SC2LE) as our experimental platform, our results demonstrate the superiority of our methodology over QMIX, QVMinMax, and QSOD in absolute performance, particularly when operating under resource constraints. Our proposed methodology contributes to the ongoing evolution of MARL techniques, showcasing advancements in attention mechanisms and optimization strategies for enhanced multi-agent system capabilities.
- Open Access
- 22 July 2024
- Published by Elsevier in Neural Networks
- Vol. 179, 106565
MuDE: Multi-agent decomposed reward-based exploration
- 22 July 2024
- Published by Elsevier in Neural Networks
- Vol. 179, 106565
Open AccessMuDE: Multi-agent decomposed reward-based exploration
- Open Access
- 16 August 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 10, 87254-87265
CuMARL: Curiosity-Based Learning in Multiagent Reinforcement Learning
In this paper, we propose a novel curiosity-based learning algorithm for Multi-agent Reinforcement Learning (MARL) to attain efficient and effective decision-making. We employ the centralized training with decentralized execution framework (CTDE) and consider that each agent has knowledge of the prior action distribution of others. To quantify the difference in agents’ knowledge, curiosity, we introduce conditional mutual information (CMI) regularization and use the amount of information for updating decision-making policy. Then, to deploy these learning frameworks in a large-scale MARL setting while acquiring high sample efficiency, we consider a Kullback-Leibler (KL) divergence-based prioritization of experiences. We evaluate the effectiveness of the proposed algorithm in three different levels of StarCraft Multi Agent Challenge (SMAC) scenarios using the PyMARL framework. The simulation-based performance analysis shows that the proposed technique significantly improves the test win rate compared to various state-of-the-art MARL benchmarks, such as the Optimistically Weighted Monotonic Value Function Factorization (OW_QMIX) and Learning Individual Intrinsic Reward (LIIR).
- 16 August 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 10, 87254-87265
Open AccessCuMARL: Curiosity-Based Learning in Multiagent Reinforcement Learning
In this paper, we propose a novel curiosity-based learning algorithm for Multi-agent Reinforcement Learning (MARL) to attain efficient and effective decision-making. We employ the centralized training with decentralized execution framework (CTDE) and consider that each agent has knowledge of the prior action distribution of others. To quantify the difference in agents’ knowledge, curiosity, we introduce conditional mutual information (CMI) regularization and use the amount of information for updating decision-making policy. Then, to deploy these learning frameworks in a large-scale MARL setting while acquiring high sample efficiency, we consider a Kullback-Leibler (KL) divergence-based prioritization of experiences. We evaluate the effectiveness of the proposed algorithm in three different levels of StarCraft Multi Agent Challenge (SMAC) scenarios using the PyMARL framework. The simulation-based performance analysis shows that the proposed technique significantly improves the test win rate compared to various state-of-the-art MARL benchmarks, such as the Optimistically Weighted Monotonic Value Function Factorization (OW_QMIX) and Learning Individual Intrinsic Reward (LIIR).
- 1 February 2013
- Published by Elsevier in Microprocessors and Microsystems
- Vol. 37 (1) , 24-32
PTL: PRAM translation layer
- 1 February 2013
- Published by Elsevier in Microprocessors and Microsystems
- Vol. 37 (1) , 24-32
PTL: PRAM translation layer
- 1 December 2010
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-5
A Modified Multi-Resolution Approach for Port Scan Detection
Although port scan detection techniques have been widely adopted by the modern network based security systems, the effectiveness of these techniques can significantly be limited since the detection performance heavily relies on the statically determined detection threshold. To tackle the problem, a multi-resolution approach called MRDS, maintaining multiple monitoring windows with the corresponding detection thresholds, has been proposed. However, deploying such technique in a high speed network is not easy due to the time and space complexity required for calculating the number of unique destination addresses contacted in the multiple monitoring windows. In this paper, we present a novel failed flow dispersion estimation technique, called Multi-Window State Map (MWSM), which requires a small amount of memory and a constant number of memory access for implementing the multi-resolution concept. We then extend the proposed MWSM into a complete port scan detector. Simulation results with real world traffic traces indicate that the proposed estimation technique manages the expected relative error and average standard error of less than 0.8% and 9% respectively and thus the MWSM based detection scheme reduces false positives by 60% compared to MRDS.
- 1 December 2010
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-5
A Modified Multi-Resolution Approach for Port Scan Detection
Although port scan detection techniques have been widely adopted by the modern network based security systems, the effectiveness of these techniques can significantly be limited since the detection performance heavily relies on the statically determined detection threshold. To tackle the problem, a multi-resolution approach called MRDS, maintaining multiple monitoring windows with the corresponding detection thresholds, has been proposed. However, deploying such technique in a high speed network is not easy due to the time and space complexity required for calculating the number of unique destination addresses contacted in the multiple monitoring windows. In this paper, we present a novel failed flow dispersion estimation technique, called Multi-Window State Map (MWSM), which requires a small amount of memory and a constant number of memory access for implementing the multi-resolution concept. We then extend the proposed MWSM into a complete port scan detector. Simulation results with real world traffic traces indicate that the proposed estimation technique manages the expected relative error and average standard error of less than 0.8% and 9% respectively and thus the MWSM based detection scheme reduces false positives by 60% compared to MRDS.
- 19 April 2008
- Published by Springer Nature in Telecommunication Systems
- Vol. 37 (4) , 169-189
A dynamic quarantine scheme for controlling unresponsive TCP sessions
- 19 April 2008
- Published by Springer Nature in Telecommunication Systems
- Vol. 37 (4) , 169-189
A dynamic quarantine scheme for controlling unresponsive TCP sessions
- Open Access
- 11 March 2007
- Published by Association for Computing Machinery (ACM)
- p. 264-269
Memory-efficient content filtering hardware for high-speed intrusion detection systems
Content filtering-based Intrusion Detection Systems have been widely deployed in enterprise networks, and have become a standard measure to protect networks and network users from cyber attacks. Although several solutions have been proposed recently, finding an efficient solution is considered as a difficult problem due to the limitations in resources such as a small memory size, as well as the growing link speed. In this paper, we present a novel content filtering technique called Table-driven Bottom-up Tree (TBT), which was designed i) to fully exploit hardware parallelism to achieve real-time packet inspection, ii) to require a small memory for storing signatures, iii) to be flexible in modifying the signature database, and iv) to support complex signature representation such as regular expressions. We configured TBT considering the hardware specifications and limitations, and implemented it using a FPGA. Simulation based performance evaluations showed that the proposed technique used only 350 Kilobytes of memory for storing the latest version of SNORT rule consisting of 2770 signatures. In addition, unlike many other hardware-based solutions, modification to signature database does not require hardware re-compilation in TBT.
- 11 March 2007
- Published by Association for Computing Machinery (ACM)
- p. 264-269
Open AccessMemory-efficient content filtering hardware for high-speed intrusion detection systems
Content filtering-based Intrusion Detection Systems have been widely deployed in enterprise networks, and have become a standard measure to protect networks and network users from cyber attacks. Although several solutions have been proposed recently, finding an efficient solution is considered as a difficult problem due to the limitations in resources such as a small memory size, as well as the growing link speed. In this paper, we present a novel content filtering technique called Table-driven Bottom-up Tree (TBT), which was designed i) to fully exploit hardware parallelism to achieve real-time packet inspection, ii) to require a small memory for storing signatures, iii) to be flexible in modifying the signature database, and iv) to support complex signature representation such as regular expressions. We configured TBT considering the hardware specifications and limitations, and implemented it using a FPGA. Simulation based performance evaluations showed that the proposed technique used only 350 Kilobytes of memory for storing the latest version of SNORT rule consisting of 2770 signatures. In addition, unlike many other hardware-based solutions, modification to signature database does not require hardware re-compilation in TBT.
- 12 September 2006
- Published by Wiley in Wireless Communications and Mobile Computing
- Vol. 8 (4) , 421-434
Proxy‐RED: an AQM scheme for wireless local area networks
Wireless access points (APs) act as bridges between wired and wireless networks. Since the actually available bandwidth in wireless networks is much smaller than the bandwidth in wired networks, there is a disparity in channel capacity which makes the access point a significant network congestion point in the downstream direction. A current architectural trend in wireless local area networks (WLAN) is to move functionality from APs to a centralized gateway in order to reduce cost and improve features. In this paper, we study the use of RED, a well known active queue management (AQM) scheme, and explicit congestion notification (ECN) to handle bandwidth disparity between the wired and the wireless interface of an access point. Then, we propose the Proxy‐RED scheme, as a solution for reducing the AQM overhead from the access point. Simulations‐based performance analysis indicates that the proposed Proxy‐RED scheme improves the overall performance of a network. In particular, the Proxy‐RED scheme significantly reduces packet loss rate and improves goodput for a small buffer, and minimizes delay for a large buffer size. Copyright © 2006 John Wiley & Sons, Ltd.
- 12 September 2006
- Published by Wiley in Wireless Communications and Mobile Computing
- Vol. 8 (4) , 421-434
Proxy‐RED: an AQM scheme for wireless local area networks
Wireless access points (APs) act as bridges between wired and wireless networks. Since the actually available bandwidth in wireless networks is much smaller than the bandwidth in wired networks, there is a disparity in channel capacity which makes the access point a significant network congestion point in the downstream direction. A current architectural trend in wireless local area networks (WLAN) is to move functionality from APs to a centralized gateway in order to reduce cost and improve features. In this paper, we study the use of RED, a well known active queue management (AQM) scheme, and explicit congestion notification (ECN) to handle bandwidth disparity between the wired and the wireless interface of an access point. Then, we propose the Proxy‐RED scheme, as a solution for reducing the AQM overhead from the access point. Simulations‐based performance analysis indicates that the proposed Proxy‐RED scheme improves the overall performance of a network. In particular, the Proxy‐RED scheme significantly reduces packet loss rate and improves goodput for a small buffer, and minimizes delay for a large buffer size. Copyright © 2006 John Wiley & Sons, Ltd.