Shu-Ping Yeh
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Shu-Ping Yeh has a total of 69 co-authors. Here are the top co-authors by number of shared publications:
- Hosein Nikopour
Intel Corporation, Santa Clara, CA, USA
6 shared publications - Sergey Andreev
Tampere University
5 shared publications - Dmitri Moltchanov
Tampere University
3 shared publications - Dusit Niyato
Nanyang technological university, Singapore, Singapore
3 shared publications - Jiacheng Wang
Nanyang Technological University, Singapore
3 shared publications - Shilpa Talwar
Intel Corporation, Santa Clara, CA, USA
3 shared publications
Latest Publications
- Open Access
- 26 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-7
Think Before You Prune: Self-Reflective Structured Pruning for Reasoning Language Models
- 26 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-7
Open AccessThink Before You Prune: Self-Reflective Structured Pruning for Reasoning Language Models
- 2 July 2026
- Published by Springer Nature
- p. 265-296
Artificial Intelligence: Sustainability Challenges, Opportunities, and the Case of 6G
- 2 July 2026
- Published by Springer Nature
- p. 265-296
Artificial Intelligence: Sustainability Challenges, Opportunities, and the Case of 6G
- 27 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Magazine
- Vol. 9 (3) , 14-16
Intelligent and Secure Internet of Things for Low-Altitude Economy Networking (Part 2)
- 27 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Magazine
- Vol. 9 (3) , 14-16
Intelligent and Secure Internet of Things for Low-Altitude Economy Networking (Part 2)
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Cognitive Communications and Networking
- Vol. 12, 12720-12736
Feudal HRL for Joint Spatio-Temporal Energy Optimization in O-RAN Systems
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Cognitive Communications and Networking
- Vol. 12, 12720-12736
Feudal HRL for Joint Spatio-Temporal Energy Optimization in O-RAN Systems
- Open Access
Think Before You Prune: Self-Reflective Structured Pruning for Reasoning Language Models
Reasoning LLMs (RLMs) such as OpenAI o1, DeepSeek-R1, and Qwen3 deliver strong multi-step reasoning through chain-of-thought generation, but their large model sizes and lengthy decode-time outputs make them costly to deploy and unsuitable for resource-constrained settings. To reduce computing and memory cost, pruning offers a promising solution by removing unimportant parameters. However, despite their success on standard LLMs, existing pruning methods severely damage RLMs, as even moderate sparsity (e.g., 20%) can collapse accuracy and completely disrupt the model's reasoning coherence. We begin by analyzing why existing pruning pipelines fail on reasoning LLMs and find that their brittleness largely stems from a mismatch between the calibration data, the pruning objective, and the model's decode-time reasoning behavior. Our study further shows that the most reliable calibration signal comes not from human-written labels but from the model's own self-generated reasoning traces, which more accurately reflect its inference distribution. Guided by these insights, we introduce RESP, a self-reflective structured pruning framework that aligns pruning decisions with the model's reasoning dynamics through self-generated calibration, decode-only gradient-based importance estimation, and progressive regeneration that maintains calibration fidelity as sparsity increases. Experiments on Qwen3-8B demonstrate that RESP markedly outperforms existing structured pruning methods on both GSM8K and MathQA, preserving near-dense accuracy at 20-30% sparsity and substantially mitigating performance collapse at higher sparsity levels. At 40% sparsity, RESP attains 81.3% accuracy on GSM8K and 59.6% on MathQA, surpassing the strongest baselines by 66.87% and 47%, respectively.
Open AccessThink Before You Prune: Self-Reflective Structured Pruning for Reasoning Language Models
Reasoning LLMs (RLMs) such as OpenAI o1, DeepSeek-R1, and Qwen3 deliver strong multi-step reasoning through chain-of-thought generation, but their large model sizes and lengthy decode-time outputs make them costly to deploy and unsuitable for resource-constrained settings. To reduce computing and memory cost, pruning offers a promising solution by removing unimportant parameters. However, despite their success on standard LLMs, existing pruning methods severely damage RLMs, as even moderate sparsity (e.g., 20%) can collapse accuracy and completely disrupt the model's reasoning coherence. We begin by analyzing why existing pruning pipelines fail on reasoning LLMs and find that their brittleness largely stems from a mismatch between the calibration data, the pruning objective, and the model's decode-time reasoning behavior. Our study further shows that the most reliable calibration signal comes not from human-written labels but from the model's own self-generated reasoning traces, which more accurately reflect its inference distribution. Guided by these insights, we introduce RESP, a self-reflective structured pruning framework that aligns pruning decisions with the model's reasoning dynamics through self-generated calibration, decode-only gradient-based importance estimation, and progressive regeneration that maintains calibration fidelity as sparsity increases. Experiments on Qwen3-8B demonstrate that RESP markedly outperforms existing structured pruning methods on both GSM8K and MathQA, preserving near-dense accuracy at 20-30% sparsity and substantially mitigating performance collapse at higher sparsity levels. At 40% sparsity, RESP attains 81.3% accuracy on GSM8K and 59.6% on MathQA, surpassing the strongest baselines by 66.87% and 47%, respectively.
- 10 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Magazine
- Vol. 8 (6) , 14-16
Guest Editorial: Applications of Large Language Models in Internet of Things
- 10 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Magazine
- Vol. 8 (6) , 14-16
Guest Editorial: Applications of Large Language Models in Internet of Things
- 9 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Magazine
- Vol. 8 (5) , 61-63
Special Issue Editorial: Applications of Large Language Models in Internet of Things
- 9 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Magazine
- Vol. 8 (5) , 61-63
Special Issue Editorial: Applications of Large Language Models in Internet of Things
- Open Access
Deep Reinforcement Learning-based Cell DTX/DRX Configuration for Network Energy Saving
3GPP Release 18 cell discontinuous transmission and reception (cell DTX/DRX) is an important new network energy saving feature for 5G. As a time-domain technique, it periodically aggregates the user data transmissions in a given duration of time when the traffic load is not heavy, so that the remaining time can be kept silent and advanced sleep modes (ASM) can be enabled to shut down more radio components and save more energy for the cell. However, inevitably the packet delay is increased, as during the silent period no transmission is allowed. In this paper we study how to configure cell DTX/DRX to optimally balance energy saving and packet delay, so that for delay-sensitive traffic maximum energy saving can be achieved while the degradation of quality of service (QoS) is minimized. As the optimal configuration can be different for different network and traffic conditions, the problem is complex and we resort to deep reinforcement learning (DRL) framework to train an AI agent to solve it. Through careful design of 1) the learning algorithm, which implements a deep Q-network (DQN) on a contextual bandit (CB) model, and 2) the reward function, which utilizes a smooth approximation of a theoretically optimal but discontinuous reward function, we are able to train an AI agent that always tries to select the best possible Cell DTX/DRX configuration under any network and traffic conditions. Simulation results show that compared to the case when cell DTX/DRX is not used, our agent can achieve up to ~45% energy saving depending on the traffic load scenario, while always maintaining no more than ~1% QoS degradation.
Open AccessDeep Reinforcement Learning-based Cell DTX/DRX Configuration for Network Energy Saving
3GPP Release 18 cell discontinuous transmission and reception (cell DTX/DRX) is an important new network energy saving feature for 5G. As a time-domain technique, it periodically aggregates the user data transmissions in a given duration of time when the traffic load is not heavy, so that the remaining time can be kept silent and advanced sleep modes (ASM) can be enabled to shut down more radio components and save more energy for the cell. However, inevitably the packet delay is increased, as during the silent period no transmission is allowed. In this paper we study how to configure cell DTX/DRX to optimally balance energy saving and packet delay, so that for delay-sensitive traffic maximum energy saving can be achieved while the degradation of quality of service (QoS) is minimized. As the optimal configuration can be different for different network and traffic conditions, the problem is complex and we resort to deep reinforcement learning (DRL) framework to train an AI agent to solve it. Through careful design of 1) the learning algorithm, which implements a deep Q-network (DQN) on a contextual bandit (CB) model, and 2) the reward function, which utilizes a smooth approximation of a theoretically optimal but discontinuous reward function, we are able to train an AI agent that always tries to select the best possible Cell DTX/DRX configuration under any network and traffic conditions. Simulation results show that compared to the case when cell DTX/DRX is not used, our agent can achieve up to ~45% energy saving depending on the traffic load scenario, while always maintaining no more than ~1% QoS degradation.
- Open Access
- 23 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of the Communications Society
- Vol. 5, 5392-5407
Analysis of Duplexing Patterns in Multi-Hop mmWave Integrated Access and Backhaul Systems
Integrated Access and Backhaul (IAB) technology promises to facilitate cost-effective deployments of 5G New Radio (NR) systems operating in both sub-6 GHz and millimeter-wave (mmWave) bands. As full-duplex wireless systems are in their infancy, initial deployments of IAB networks may need to rely on half-duplex operation to coordinate transmissions between access and backhaul links. However, the use of half-duplex operation not only makes the scheduling of links in the IAB networks interdependent, but also the number of their feasible combinations grows exponentially with the network size, thereby posing challenges to the optimal design of such systems. In this paper, by accounting for mmWave radio characteristics, we propose a joint resource allocation and link scheduling framework to enhance the user equipment (UE) throughput in multi-hop in-band IAB systems. We keep the problem in the form of linear programming type for the feasibility of the practical applications. We show that the increased number of uplink and downlink transmission time interval (TTI) configurations does not result in improved UE throughput as compared to two configurations. Further, we demonstrate that in-band IAB systems tend to be backhaul-limited, and the utilization of multi-beam functionality at the IAB-donor alleviates this limitation by doubling the average UE throughput. Finally, we show that the use of proportional-fair allocations allows the average UE throughput to be improved by around 10% as compared to the max-min allocations.
- 23 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of the Communications Society
- Vol. 5, 5392-5407
Open AccessAnalysis of Duplexing Patterns in Multi-Hop mmWave Integrated Access and Backhaul Systems
Integrated Access and Backhaul (IAB) technology promises to facilitate cost-effective deployments of 5G New Radio (NR) systems operating in both sub-6 GHz and millimeter-wave (mmWave) bands. As full-duplex wireless systems are in their infancy, initial deployments of IAB networks may need to rely on half-duplex operation to coordinate transmissions between access and backhaul links. However, the use of half-duplex operation not only makes the scheduling of links in the IAB networks interdependent, but also the number of their feasible combinations grows exponentially with the network size, thereby posing challenges to the optimal design of such systems. In this paper, by accounting for mmWave radio characteristics, we propose a joint resource allocation and link scheduling framework to enhance the user equipment (UE) throughput in multi-hop in-band IAB systems. We keep the problem in the form of linear programming type for the feasibility of the practical applications. We show that the increased number of uplink and downlink transmission time interval (TTI) configurations does not result in improved UE throughput as compared to two configurations. Further, we demonstrate that in-band IAB systems tend to be backhaul-limited, and the utilization of multi-beam functionality at the IAB-donor alleviates this limitation by doubling the average UE throughput. Finally, we show that the use of proportional-fair allocations allows the average UE throughput to be improved by around 10% as compared to the max-min allocations.
- Open Access
- 21 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. 73 (10) , 15125-15139
Delay-Aware Link Scheduling in IAB Networks With Dynamic User Demands
Integrated Access and Backhaul (IAB) is a costefficient network densification technology for improving the coverage and capacity of the millimeter-wave (mmWave) cellular networks. In IAB systems, user traffic is forwarded to/from the wired base station by one or more relay stations, known as IAB nodes. Due to the multi-hop relaying, these systems may be subject to large packet delays and poor performance when the load is unevenly distributed among nodes. Addressing this limitation via delay-aware access and backhaul link scheduling in IAB networks is challenging due to potentially large network scale, complex topology, half-duplex, and interference constraints. In this paper, the topical link scheduling problem is formulated as a Markov decision problem (MDP) for a single-donor IAB system with a general topology that allows for users with different delay requirements and traffic dynamics. The proposed link scheduling strategy jointly optimizes (i) user traffic routing and (ii) multiplexing of access and backhaul links under half-duplex constraints and non-negligible interference that may arise in dense IAB systems even with high beam directionality. To address the complexity of our formulated MDP, we consider several approximation methods, namely, Q-learning, Monte Carlo Tree Search (MCTS), and genetic algorithms (GAs). Then, we propose a customized version of the GA, which provides the preferred optimality-complexity trade-off and offers a 15% packet delay reduction as compared to the state-of-the-art backpressure algorithm.
- 21 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. 73 (10) , 15125-15139
Open AccessDelay-Aware Link Scheduling in IAB Networks With Dynamic User Demands
Integrated Access and Backhaul (IAB) is a costefficient network densification technology for improving the coverage and capacity of the millimeter-wave (mmWave) cellular networks. In IAB systems, user traffic is forwarded to/from the wired base station by one or more relay stations, known as IAB nodes. Due to the multi-hop relaying, these systems may be subject to large packet delays and poor performance when the load is unevenly distributed among nodes. Addressing this limitation via delay-aware access and backhaul link scheduling in IAB networks is challenging due to potentially large network scale, complex topology, half-duplex, and interference constraints. In this paper, the topical link scheduling problem is formulated as a Markov decision problem (MDP) for a single-donor IAB system with a general topology that allows for users with different delay requirements and traffic dynamics. The proposed link scheduling strategy jointly optimizes (i) user traffic routing and (ii) multiplexing of access and backhaul links under half-duplex constraints and non-negligible interference that may arise in dense IAB systems even with high beam directionality. To address the complexity of our formulated MDP, we consider several approximation methods, namely, Q-learning, Monte Carlo Tree Search (MCTS), and genetic algorithms (GAs). Then, we propose a customized version of the GA, which provides the preferred optimality-complexity trade-off and offers a 15% packet delay reduction as compared to the state-of-the-art backpressure algorithm.