Zhiwei Yu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Zhiwei Yu has a total of 18 co-authors. Here are the top co-authors by number of shared publications:
- Jialong Li
Shenzhen University of Advanced Technology, Shenzhen, China
9 shared publications - Chengze Du
Computer Science and Control Engineering, Shenzhen University of Advanced Technology, Shenzhen, China
7 shared publications - Heng Xu
Shenzhen University of Advanced Technology, Shenzhen, China
7 shared publications - Bo Liu
Shenzhen University of Advanced Technology
5 shared publications - Ying Zhou
Beijing Jiaotong University, Beijing, China
5 shared publications - Letian Li
Department of Information Engineering, The Chinese University of Hong Kong, Shatin, Hong Kong
3 shared publications
Latest Publications
- 19 August 2026
- Published by Elsevier in Journal of Network and Computer Applications
- Vol. 255
QoS fusion-driven optimization of routing cooperation mechanisms in IoT multi-AS communication networks
- 19 August 2026
- Published by Elsevier in Journal of Network and Computer Applications
- Vol. 255
QoS fusion-driven optimization of routing cooperation mechanisms in IoT multi-AS communication networks
- Open Access
- 15 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 1798-1802
FMD-AL: Cold-Start Active Learning based on Foundation Model
- 15 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 1798-1802
Open AccessFMD-AL: Cold-Start Active Learning based on Foundation Model
- 11 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Cloud Computing
- Vol. PP (3) , 1-14
Temporal-Aware GPU Resource Allocation for Distributed LLM Inference via Reinforcement Learning
The rapid growth of large language model (LLM) services imposes increasing demands on distributed GPU inference infrastructure. Most existing scheduling systems follow a reactive paradigm, relying solely on the current system state to make decisions, without considering how task demand and resource availability evolve over time. This lack of temporal awareness in reactive approaches leads to inefficient GPU utilization, high task migration overhead, and poor system responsiveness under dynamic workloads. In this work, we identify the fundamental limitations of these instantaneous-state-only scheduling approaches and propose Temporal Optimal Resource scheduling via Two-layer Architecture (TORTA). TORTA introduces a spatiotemporal scheduling framework that captures both long-term workload patterns and short-term execution constraints. It adopts a two-layer design: a macro-level scheduler leverages reinforcement learning and optimal transport to coordinate inter-region task distribution, while a micro-level allocator refines task-to-server assignments within each region to reduce latency and switching costs. Experimental results across multiple network topologies show that TORTA reduces average inference response time by up to 15%, improves load balance by approximately 4-5%, and cuts total operational cost by 10-20% compared to state-of-the-art baseline methods.
- 11 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Cloud Computing
- Vol. PP (3) , 1-14
Temporal-Aware GPU Resource Allocation for Distributed LLM Inference via Reinforcement Learning
The rapid growth of large language model (LLM) services imposes increasing demands on distributed GPU inference infrastructure. Most existing scheduling systems follow a reactive paradigm, relying solely on the current system state to make decisions, without considering how task demand and resource availability evolve over time. This lack of temporal awareness in reactive approaches leads to inefficient GPU utilization, high task migration overhead, and poor system responsiveness under dynamic workloads. In this work, we identify the fundamental limitations of these instantaneous-state-only scheduling approaches and propose Temporal Optimal Resource scheduling via Two-layer Architecture (TORTA). TORTA introduces a spatiotemporal scheduling framework that captures both long-term workload patterns and short-term execution constraints. It adopts a two-layer design: a macro-level scheduler leverages reinforcement learning and optimal transport to coordinate inter-region task distribution, while a micro-level allocator refines task-to-server assignments within each region to reduce latency and switching costs. Experimental results across multiple network topologies show that TORTA reduces average inference response time by up to 15%, improves load balance by approximately 4-5%, and cuts total operational cost by 10-20% compared to state-of-the-art baseline methods.
- 20 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network Science and Engineering
- Vol. 13, 9572-9590
Gsched: Coordinated Flow-Control and Priority Scheduling for DNN Training in AI Cluster
Distributed deep neural network (DNN) training in multi-tenant cloud environments generates long-lived, bandwidth-intensive, and interdependent flows that challenge traditional fairness-based network scheduling. We present Gsched, a framework that orchestrates centralized planning with hardware priority enforcement to realize communication interleaving, thereby eliminating contention and maximizing pipeline efficiency. Unlike methods relying on complex rate control or custom switch hardware, Gsched leverages standard commodity switch features like DSCP and SPQ to enforce strict priority scheduling. It employs a cold-start profiling mechanism to capture task characteristics without user intervention, and a centralized solver to compute the optimal interleaving order that maximizes Average Transmission Efficiency (ATE). Our analysis proves that this prioritization strategy deterministically drives the system to a contention-free state, while a local perturbation theorem guarantees robustness against distributed execution conflicts via network backpressure. Across workloads, Gsched achieves performance close to the state-of-the-art Crux under heterogeneous workloads and remains competitive in highly diverse cases. More importantly, Gsched provides a theoretical foundation that justifies scalable per-bottleneck scheduling in multi-bottleneck networks.
- 20 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network Science and Engineering
- Vol. 13, 9572-9590
Gsched: Coordinated Flow-Control and Priority Scheduling for DNN Training in AI Cluster
Distributed deep neural network (DNN) training in multi-tenant cloud environments generates long-lived, bandwidth-intensive, and interdependent flows that challenge traditional fairness-based network scheduling. We present Gsched, a framework that orchestrates centralized planning with hardware priority enforcement to realize communication interleaving, thereby eliminating contention and maximizing pipeline efficiency. Unlike methods relying on complex rate control or custom switch hardware, Gsched leverages standard commodity switch features like DSCP and SPQ to enforce strict priority scheduling. It employs a cold-start profiling mechanism to capture task characteristics without user intervention, and a centralized solver to compute the optimal interleaving order that maximizes Average Transmission Efficiency (ATE). Our analysis proves that this prioritization strategy deterministically drives the system to a contention-free state, while a local perturbation theorem guarantees robustness against distributed execution conflicts via network backpressure. Across workloads, Gsched achieves performance close to the state-of-the-art Crux under heterogeneous workloads and remains competitive in highly diverse cases. More importantly, Gsched provides a theoretical foundation that justifies scalable per-bottleneck scheduling in multi-bottleneck networks.
- Open Access
- 14 April 2026
- Published by Elsevier in SSRN Electronic Journal
QoS Fusion-Driven Optimization of Routing Cooperation Mechanisms in IoT Multi-AS Communication Networks
To meet the specific Quality of Service (QoS) requirements of Internet of Things (IoT) networks, routing optimization in inter-domain communication is crucial. Existing inter-AS routing protocols typically focus on route reachability based solely on neighboring ASes, neglecting global QoS metrics such as latency and packet loss. Moreover, intra-AS protocols are limited to single-AS optimization, preventing effective cross-AS coordination. This paper presents a novel QoS-Aware Federated Reinforcement Learning Routing algorithm (Q-FRLR), which constructs a transmission delay probability distribution using both historical and real-time AS link data. By mapping delay thresholds to packet loss probabilities, the multi-objective optimization problem is reformulated into a single-objective problem to minimize end-to-end delay during path selection. Q-FRLR incorporates homomorphic encryption to securely share desensitized five-tuple data across ASes, allowing each AS to train a local reinforcement learning model using historical QoS data, while fine-tuning it with cross-AS information. A Software Defined Networking (SDN) controller aggregates QoS scores, dynamically adjusts routing weights, and makes real-time routing decisions. Experimental results conducted in Mininet on three large-scale network topologies demonstrate that Q-FRLR reduces average end-to-end delay by 9.01% compared to the optimal benchmark, and decreases controller decision time by 11.34%, eliminating the need for global model training.
- 14 April 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessQoS Fusion-Driven Optimization of Routing Cooperation Mechanisms in IoT Multi-AS Communication Networks
To meet the specific Quality of Service (QoS) requirements of Internet of Things (IoT) networks, routing optimization in inter-domain communication is crucial. Existing inter-AS routing protocols typically focus on route reachability based solely on neighboring ASes, neglecting global QoS metrics such as latency and packet loss. Moreover, intra-AS protocols are limited to single-AS optimization, preventing effective cross-AS coordination. This paper presents a novel QoS-Aware Federated Reinforcement Learning Routing algorithm (Q-FRLR), which constructs a transmission delay probability distribution using both historical and real-time AS link data. By mapping delay thresholds to packet loss probabilities, the multi-objective optimization problem is reformulated into a single-objective problem to minimize end-to-end delay during path selection. Q-FRLR incorporates homomorphic encryption to securely share desensitized five-tuple data across ASes, allowing each AS to train a local reinforcement learning model using historical QoS data, while fine-tuning it with cross-AS information. A Software Defined Networking (SDN) controller aggregates QoS scores, dynamically adjusts routing weights, and makes real-time routing decisions. Experimental results conducted in Mininet on three large-scale network topologies demonstrate that Q-FRLR reduces average end-to-end delay by 9.01% compared to the optimal benchmark, and decreases controller decision time by 11.34%, eliminating the need for global model training.
- 23 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Networking
- Vol. 34, 4431-4448
RailS: Load Balancing for All-to-All Communication in Distributed Mixture-of-Experts Training
Training Mixture-of-Experts (MoE) models introduces sparse and highly imbalanced all-to-all communication that dominates iteration time. Conventional load-balancing methods fail to exploit the deterministic topology of Rail architectures, leaving multi-NIC bandwidth underutilized. We present RailS, a distributed load-balancing framework that minimizes all-to-all completion time in MoE training. RailS leverages the Rail topology’s symmetry to prove that uniform sending ensures uniform receiving, transforming global coordination into local scheduling. Each node independently executes a Longest Processing Time First (LPT) spraying scheduler to proactively balance traffic using local information. RailS activates N parallel rails for fine-grained, topology-aware multipath transmission. Across synthetic and real-world MoE workloads, RailS improves bus bandwidth by 20%–78% and reduces completion time by 17%–78%. For Mixtral workloads, it shortens iteration time by 18%–40% and achieves near-optimal load balance, fully exploiting architectural parallelism in distributed training.
- 23 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Networking
- Vol. 34, 4431-4448
RailS: Load Balancing for All-to-All Communication in Distributed Mixture-of-Experts Training
Training Mixture-of-Experts (MoE) models introduces sparse and highly imbalanced all-to-all communication that dominates iteration time. Conventional load-balancing methods fail to exploit the deterministic topology of Rail architectures, leaving multi-NIC bandwidth underutilized. We present RailS, a distributed load-balancing framework that minimizes all-to-all completion time in MoE training. RailS leverages the Rail topology’s symmetry to prove that uniform sending ensures uniform receiving, transforming global coordination into local scheduling. Each node independently executes a Longest Processing Time First (LPT) spraying scheduler to proactively balance traffic using local information. RailS activates N parallel rails for fine-grained, topology-aware multipath transmission. Across synthetic and real-world MoE workloads, RailS improves bus bandwidth by 20%–78% and reduces completion time by 17%–78%. For Mixtral workloads, it shortens iteration time by 18%–40% and achieves near-optimal load balance, fully exploiting architectural parallelism in distributed training.
- 10 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 23, 2528-2542
REACH: Reinforcement Learning for Efficient Allocation in Community and Heterogeneous Networks
Community GPU(Graphics Processing Unit) platforms are emerging as a cost-effective and democratized alternative to centralized GPU clusters for AI(Artificial Intelligence) workloads, aggregating idle consumer GPUs from globally distributed and heterogeneous environments. However, their extreme hardware/software diversity, volatile availability, and variable network conditions render traditional schedulers ineffective, leading to suboptimal task completion. In this work, we present REACH (Reinforcement Learning for Efficient Allocation in Community and Heterogeneous Networks), a Transformer-based reinforcement learning framework that redefines task scheduling as a sequence scoring problem to balance performance, reliability, cost, and network efficiency. By modeling both global GPU states and task requirements, REACH learns to adaptively co-locate computation with data, prioritize critical jobs, and mitigate the impact of unreliable resources. Extensive simulation results show that REACH improves task completion rates by up to 17%, more than doubles the success rate for high-priority tasks, and reduces bandwidth penalties by over 80% compared to state-of-the-art baselines. Stress tests further demonstrate its robustness to GPU churn and network congestion, while scalability experiments confirm its effectiveness in large-scale, high-contention scenarios.
- 10 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 23, 2528-2542
REACH: Reinforcement Learning for Efficient Allocation in Community and Heterogeneous Networks
Community GPU(Graphics Processing Unit) platforms are emerging as a cost-effective and democratized alternative to centralized GPU clusters for AI(Artificial Intelligence) workloads, aggregating idle consumer GPUs from globally distributed and heterogeneous environments. However, their extreme hardware/software diversity, volatile availability, and variable network conditions render traditional schedulers ineffective, leading to suboptimal task completion. In this work, we present REACH (Reinforcement Learning for Efficient Allocation in Community and Heterogeneous Networks), a Transformer-based reinforcement learning framework that redefines task scheduling as a sequence scoring problem to balance performance, reliability, cost, and network efficiency. By modeling both global GPU states and task requirements, REACH learns to adaptively co-locate computation with data, prioritize critical jobs, and mitigate the impact of unreliable resources. Extensive simulation results show that REACH improves task completion rates by up to 17%, more than doubles the success rate for high-priority tasks, and reduces bandwidth penalties by over 80% compared to state-of-the-art baselines. Stress tests further demonstrate its robustness to GPU churn and network congestion, while scalability experiments confirm its effectiveness in large-scale, high-contention scenarios.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Networking
- Vol. 34, 6724-6737
PlatoNT: Learning a Platonic Representation for Unified Network Tomography
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Networking
- Vol. 34, 6724-6737
PlatoNT: Learning a Platonic Representation for Unified Network Tomography
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 23, 5592-5605
Distributed Flow Control for Efficient DNN Training Scheduling
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 23, 5592-5605
Distributed Flow Control for Efficient DNN Training Scheduling