Yi Wang
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
Yi Wang has a total of 190 co-authors. Here are the top co-authors by number of shared publications:
Bin Liu
Tsinghua National Laboratory for Information Science and Technology, Departement of Computer Science and Technology, Tsinghua University, China
17 shared publicationsHuichen Dai
Tsinghua National Laboratory for Information Science and Technology, Departement of Computer Science and Technology, Tsinghua University, China
9 shared publicationsJianyuan Lu
Tsinghua National Laboratory for Information Science and Technology, Departement of Computer Science and Technology, Tsinghua University, China
6 shared publicationsTing Zhang
Tsinghua National Laboratory for Information Science and Technology, Departement of Computer Science and Technology, Tsinghua University, China
6 shared publications- Guoming Tang
The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China
5 shared publications - Ruide Cao
Southern University of Science and Technology, China
5 shared publications
Latest Publications
- Open Access
- 5 November 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Sensor Networks
Capability and Explanation Boundaries of Large Language Models for Non-Intrusive Load Monitoring
Non-intrusive load monitoring (NILM) aims to disaggregate total electricity consumption into individual appliance usage. Although deep learning has advanced NILM, it remains limited by its dependence on labeled data, restricted generalization, and lack of explainability. This paper introduces a prompt-based NILM framework that uses large language models (LLMs) with in-context learning. We integrate appliance features, contextual history, and one-shot examples into a unified prompting strategy and evaluate it on the REDD and UK-DALE datasets. Prompt-only LLMs deliver basic NILM capability but lag behind specialized deep learning models in accuracy, while achieving zero-shot generalization across houses and regions by adapting the injected appliance features. To characterize the unique value of LLMs for NILM, we propose a three-layer faithfulness framework that evaluates LLM-generated explanations along factual validity, internal coherence, and external correctness. The framework reveals a traceable-but-misaligned pattern in which LLMs often cite numerical values supported by the input but fail to align the cited events with true appliance activity. Rather than replacing specialized models, LLMs are better positioned as a semantic interface for energy disaggregation. We release our framework to support follow-up research.
- 5 November 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Sensor Networks
Open AccessCapability and Explanation Boundaries of Large Language Models for Non-Intrusive Load Monitoring
Non-intrusive load monitoring (NILM) aims to disaggregate total electricity consumption into individual appliance usage. Although deep learning has advanced NILM, it remains limited by its dependence on labeled data, restricted generalization, and lack of explainability. This paper introduces a prompt-based NILM framework that uses large language models (LLMs) with in-context learning. We integrate appliance features, contextual history, and one-shot examples into a unified prompting strategy and evaluate it on the REDD and UK-DALE datasets. Prompt-only LLMs deliver basic NILM capability but lag behind specialized deep learning models in accuracy, while achieving zero-shot generalization across houses and regions by adapting the injected appliance features. To characterize the unique value of LLMs for NILM, we propose a three-layer faithfulness framework that evaluates LLM-generated explanations along factual validity, internal coherence, and external correctness. The framework reveals a traceable-but-misaligned pattern in which LLMs often cite numerical values supported by the input but fail to align the cited events with true appliance activity. Rather than replacing specialized models, LLMs are better positioned as a semantic interface for energy disaggregation. We release our framework to support follow-up research.
- Open Access
- 14 October 2026
- Published by Elsevier in SSRN Electronic Journal
A Delay-Aligned Redundant Scheduling Framework for Enhancing CQF Reliability in Time-Sensitive Networking
Industrial real-time systems require both reliable communication and predictable delivery timing. In Time-Sensitive Networking (TSN), Frame Replication and Elimination for Reliability (FRER) provides redundant transmission, but under Cyclic Queuing and Forwarding (CQF), replicas on paths with unequal hop counts experience substantially different forwarding durations. When a fault drops the earlier copy, the surviving copy arrives later and can cause a periodic control task to miss the current update. We propose ReCQF, a delay-aligned redundant scheduling approach that builds on the existing Multi-Instance CQF (MICQF) mechanism to jointly select cycle instances for asymmetric replica paths. ReCQF matches the forwarding granularity to path length and compensates the residual duration mismatch through bounded waiting. We model replica schedules as transmission windows, derive the conditions under which finite holding achieves exact compensation, and bound the resulting template durations and final-window endpoint differences. The Delay Alignment Scheduling Algorithm (DASA) coordinates cycle-pair selection, waiting compensation, and window placement under shared-link capacity constraints. On ring topologies with ten random seeds, ReCQF reduces the mean replica path-duration difference by 97.9% and 94.1% relative to CQF+FRER and CSQF+FRER, shortens the mean maximum path duration by 43%, and schedules 72.5% of flows within a 400us duration-difference bound, compared with 51.0% for CSQF+FRER. A hardware testbed confirms that assigning a finer cycle to the longer path narrows the measured inter-path delay gap.
- 14 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessA Delay-Aligned Redundant Scheduling Framework for Enhancing CQF Reliability in Time-Sensitive Networking
Industrial real-time systems require both reliable communication and predictable delivery timing. In Time-Sensitive Networking (TSN), Frame Replication and Elimination for Reliability (FRER) provides redundant transmission, but under Cyclic Queuing and Forwarding (CQF), replicas on paths with unequal hop counts experience substantially different forwarding durations. When a fault drops the earlier copy, the surviving copy arrives later and can cause a periodic control task to miss the current update. We propose ReCQF, a delay-aligned redundant scheduling approach that builds on the existing Multi-Instance CQF (MICQF) mechanism to jointly select cycle instances for asymmetric replica paths. ReCQF matches the forwarding granularity to path length and compensates the residual duration mismatch through bounded waiting. We model replica schedules as transmission windows, derive the conditions under which finite holding achieves exact compensation, and bound the resulting template durations and final-window endpoint differences. The Delay Alignment Scheduling Algorithm (DASA) coordinates cycle-pair selection, waiting compensation, and window placement under shared-link capacity constraints. On ring topologies with ten random seeds, ReCQF reduces the mean replica path-duration difference by 97.9% and 94.1% relative to CQF+FRER and CSQF+FRER, shortens the mean maximum path duration by 43%, and schedules 72.5% of flows within a 400us duration-difference bound, compared with 51.0% for CSQF+FRER. A hardware testbed confirms that assigning a finer cycle to the longer path narrows the measured inter-path delay gap.
- Open Access
- 21 July 2026
- Published by Elsevier in SSRN Electronic Journal
Effective Phase Alignment: Reducing Queuing Delay in 3-Queue Cyclic Queuing and Forwarding for Wide-Area Deterministic Networks
Cyclic Queuing and Forwarding (CQF) is a core mechanism of Time-Sensitive Networking (TSN)and provides deterministic latency without complex gating calculations. Scaling CQF to wide-area networks is difficult. Long link delays and imprecise synchronization shrink the usable partof each time slot and lower the usable bandwidth. 3-queue CQF addresses this problem with anadditional queue that absorbs link delay and decouples transmission timing from reception timing.The improvement has a cost. The per-hop queuing delay bound rises from 𝑇 in traditional CQF to 2𝑇 ,where 𝑇 is the slot length, and the extra delay accumulates along multi-hop paths.This paper presents Effective Phase Alignment (EPA), a mechanism that reduces the queuing delayof 3-queue CQF by aligning the effective phase, i.e., the phase difference between the sending andreceiving windows. EPA adjusts per-port transmission offsets and requires only frequency synchro-nization. The analysis shows that optimal alignment bounds the flow-weighted average queuing delayof each output port by (1.5 − 12𝑁 )𝑇 for any link delays and clock offsets, strictly below 1.5𝑇 , where𝑁 is the number of input ports of the output port. Perfect alignment recovers the delay of traditionalCQF. In simulations on linear, multi-layer ring, and random regular graph topologies, EPA reducesthe average queuing delay of standard 3-queue CQF by 22.78% to 32.66%.
- 21 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessEffective Phase Alignment: Reducing Queuing Delay in 3-Queue Cyclic Queuing and Forwarding for Wide-Area Deterministic Networks
Cyclic Queuing and Forwarding (CQF) is a core mechanism of Time-Sensitive Networking (TSN)and provides deterministic latency without complex gating calculations. Scaling CQF to wide-area networks is difficult. Long link delays and imprecise synchronization shrink the usable partof each time slot and lower the usable bandwidth. 3-queue CQF addresses this problem with anadditional queue that absorbs link delay and decouples transmission timing from reception timing.The improvement has a cost. The per-hop queuing delay bound rises from 𝑇 in traditional CQF to 2𝑇 ,where 𝑇 is the slot length, and the extra delay accumulates along multi-hop paths.This paper presents Effective Phase Alignment (EPA), a mechanism that reduces the queuing delayof 3-queue CQF by aligning the effective phase, i.e., the phase difference between the sending andreceiving windows. EPA adjusts per-port transmission offsets and requires only frequency synchro-nization. The analysis shows that optimal alignment bounds the flow-weighted average queuing delayof each output port by (1.5 − 12𝑁 )𝑇 for any link delays and clock offsets, strictly below 1.5𝑇 , where𝑁 is the number of input ports of the output port. Perfect alignment recovers the delay of traditionalCQF. In simulations on linear, multi-layer ring, and random regular graph topologies, EPA reducesthe average queuing delay of standard 3-queue CQF by 22.78% to 32.66%.
- Open Access
- 13 April 2026
- Published by Elsevier in SSRN Electronic Journal
Enhancing CQF Redundancy with Delay Alignment Scheduling in TSN
Time-Sensitive Networking (TSN) deployments that use Frame Replication and Elimination for Reliability (FRER) require redundant copies to arrive within a sufficiently small time window for effective elimination. However, conventional FRER deployments over Cyclic Queuing and Forwarding (CQF) suffer from large inter-path delay mismatches when redundant paths have different hop counts, which increases end-to-end delay and can inflate reserved resources at elimination points. To address this scheduling problem, we propose ReCQF, a delay-aligned redundant scheduling approach built on Multi-Instance CQF (MICQF). ReCQF adaptively maps redundant paths to CQF instances with different cycle durations and applies a hop-by-hop heuristic to reduce inter-path delay difference while maintaining bounded latency. Evaluations on representative industrial ring and grid topologies, together with a hardware proof-of-concept testbed, show that ReCQF reduces average maximum delay by 35.92% and 36.08% and reduces average delay difference by 98.06% and 94.33% compared with CQF+FRER and CSQF+FRER, respectively, under the evaluated ring-topology settings. These results indicate that ReCQF better supports timely FRER elimination in TSN scenarios.
- 13 April 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessEnhancing CQF Redundancy with Delay Alignment Scheduling in TSN
Time-Sensitive Networking (TSN) deployments that use Frame Replication and Elimination for Reliability (FRER) require redundant copies to arrive within a sufficiently small time window for effective elimination. However, conventional FRER deployments over Cyclic Queuing and Forwarding (CQF) suffer from large inter-path delay mismatches when redundant paths have different hop counts, which increases end-to-end delay and can inflate reserved resources at elimination points. To address this scheduling problem, we propose ReCQF, a delay-aligned redundant scheduling approach built on Multi-Instance CQF (MICQF). ReCQF adaptively maps redundant paths to CQF instances with different cycle durations and applies a hop-by-hop heuristic to reduce inter-path delay difference while maintaining bounded latency. Evaluations on representative industrial ring and grid topologies, together with a hardware proof-of-concept testbed, show that ReCQF reduces average maximum delay by 35.92% and 36.08% and reduces average delay difference by 98.06% and 94.33% compared with CQF+FRER and CSQF+FRER, respectively, under the evaluated ring-topology settings. These results indicate that ReCQF better supports timely FRER elimination in TSN scenarios.
- 27 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. PP (00189545) , 1-16
ReToCue: Reliability-Aware Task Offloading and Caching in UAV-assisted LEO Satellite Edge Computing
The integration of Space-Air-Ground Networks with mobile edge computing represents a key 6G paradigm enabled by advances in low Earth orbit (LEO) satellites, unmanned aerial vehicles (UAVs), and terrestrial networks. However, several critical challenges emerge when handling delay-sensitive services in rescue missions and infrastructure-less scenarios, where terrestrial network layers are absent. In such scenarios, the limited battery capacity of UAVs and satellites can be exhausted quickly. These factors jointly increase the probability of task delay violations and degrade the quality of service. To address these challenges, our work first investigates the joint optimization of edge caching and computation offloading in the UAV-assisted LEO Satellite edge computing (UAV-LEC) system. The primary objective in our formulation is to minimize total energy consumption while maintaining strict service reliability requirements. Given the distributed and stochastic characteristics of the UAVLEC system, we utilize a partially observable Markov decision process (POMDP) to model the system. Furthermore, martingale theory is incorporated to analyze the service reliability bounds of the tandem queue in the proposed framework. Building on these theoretical foundations, we propose a novel asynchronous interactive multi-agent reinforcement learning algorithm to address the distributed caching and computation optimization problem in the UAV-LEC system. Finally, we conduct comprehensive simulations to demonstrate that the proposed algorithm achieves significant improvements compared to the baseline.
- 27 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. PP (00189545) , 1-16
ReToCue: Reliability-Aware Task Offloading and Caching in UAV-assisted LEO Satellite Edge Computing
The integration of Space-Air-Ground Networks with mobile edge computing represents a key 6G paradigm enabled by advances in low Earth orbit (LEO) satellites, unmanned aerial vehicles (UAVs), and terrestrial networks. However, several critical challenges emerge when handling delay-sensitive services in rescue missions and infrastructure-less scenarios, where terrestrial network layers are absent. In such scenarios, the limited battery capacity of UAVs and satellites can be exhausted quickly. These factors jointly increase the probability of task delay violations and degrade the quality of service. To address these challenges, our work first investigates the joint optimization of edge caching and computation offloading in the UAV-assisted LEO Satellite edge computing (UAV-LEC) system. The primary objective in our formulation is to minimize total energy consumption while maintaining strict service reliability requirements. Given the distributed and stochastic characteristics of the UAVLEC system, we utilize a partially observable Markov decision process (POMDP) to model the system. Furthermore, martingale theory is incorporated to analyze the service reliability bounds of the tandem queue in the proposed framework. Building on these theoretical foundations, we propose a novel asynchronous interactive multi-agent reinforcement learning algorithm to address the distributed caching and computation optimization problem in the UAV-LEC system. Finally, we conduct comprehensive simulations to demonstrate that the proposed algorithm achieves significant improvements compared to the baseline.
- Open Access
SAGkit: A Python SAG Toolkit for Response Time Analysis of Hybrid-Triggered Jobs
For distributed control systems, modern latency-critical applications are increasingly demanding real-time guarantees and robustness. Response-time analysis (RTA) is useful for this purpose, as it helps analyze and guarantee timing bounds. However, conventional RTA methods struggle with the state-space explosion problem, especially in non-preemptive systems with release jitter and execution time variations. In this paper, we introduce SAGkit, a Python toolkit that implements the schedule-abstraction graph (SAG) framework. SAGkit novelly enables exact and sustainable RTA of hybrid-triggered jobs by allowing job absence on the SAG basis. Our experiments demonstrate that SAGkit achieves exactness with acceptable runtime and memory overhead. This lightweight toolkit empowers researchers to analyze complex distributed control systems and is open-access for further development.
Open AccessSAGkit: A Python SAG Toolkit for Response Time Analysis of Hybrid-Triggered Jobs
For distributed control systems, modern latency-critical applications are increasingly demanding real-time guarantees and robustness. Response-time analysis (RTA) is useful for this purpose, as it helps analyze and guarantee timing bounds. However, conventional RTA methods struggle with the state-space explosion problem, especially in non-preemptive systems with release jitter and execution time variations. In this paper, we introduce SAGkit, a Python toolkit that implements the schedule-abstraction graph (SAG) framework. SAGkit novelly enables exact and sustainable RTA of hybrid-triggered jobs by allowing job absence on the SAG basis. Our experiments demonstrate that SAGkit achieves exactness with acceptable runtime and memory overhead. This lightweight toolkit empowers researchers to analyze complex distributed control systems and is open-access for further development.
- Open Access
- 11 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 34-44
Prompting Large Language Models for Training-Free Non-Intrusive Load Monitoring
- 11 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 34-44
Open AccessPrompting Large Language Models for Training-Free Non-Intrusive Load Monitoring
- 2 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Networking
- Vol. PP, 1-16
Efficient Data Center Network Monitoring and Troubleshooting With LMon: Leveraging ECMP Hashing Linearity and Lightweight Probing
Network performance monitoring and troubleshooting are crucial yet challenging tasks in datacenter management. Despite the numerous solutions that have been proposed in recent years, their efforts are often hindered by high costs and unreliable failure localization, making it difficult to deploy them in real-world environments. In this paper, we present LMon, a highly reliable and efficient system for monitoring and troubleshooting in datacenter networks. LMon utilizes the characteristic of ECMP hashing linearity to control probe packet routing, enabling the monitoring of targeted paths without any modification of underlying protocols and devices. Additionally, LMon leverages a lightweight probing technique to reduce monitoring overhead, as well as integrates the improved LASSO regression and hypothesis testing for higher accuracy and faster processing in link failure localization. We evaluate the performance of LMon in our testing environment. Compared to the monitoring system Pingmesh, LMon generates only one-third probes while maintaining 99% accuracy and 1% false negatives.
- 2 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Networking
- Vol. PP, 1-16
Efficient Data Center Network Monitoring and Troubleshooting With LMon: Leveraging ECMP Hashing Linearity and Lightweight Probing
Network performance monitoring and troubleshooting are crucial yet challenging tasks in datacenter management. Despite the numerous solutions that have been proposed in recent years, their efforts are often hindered by high costs and unreliable failure localization, making it difficult to deploy them in real-world environments. In this paper, we present LMon, a highly reliable and efficient system for monitoring and troubleshooting in datacenter networks. LMon utilizes the characteristic of ECMP hashing linearity to control probe packet routing, enabling the monitoring of targeted paths without any modification of underlying protocols and devices. Additionally, LMon leverages a lightweight probing technique to reduce monitoring overhead, as well as integrates the improved LASSO regression and hypothesis testing for higher accuracy and faster processing in link failure localization. We evaluate the performance of LMon in our testing environment. Compared to the monitoring system Pingmesh, LMon generates only one-third probes while maintaining 99% accuracy and 1% false negatives.
- 9 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Informatics
- Vol. 21 (7) , 5390-5400
FlexTAS: Flexible Gating Control for Enhanced Time-Sensitive Networking Deployment
Time-sensitive networking (TSN), essential in industrial networks for its promise of reliable and deterministic data transmission, faces deployment challenges due to the limitations of existing time-aware shaper (TAS)-based scheduling algorithms. Specifically, the size of the generated gate control lists (GCLs) is usually too large to be deployed in actual devices. To bridge the gap between theory and practice, we propose FlexTAS, a flexible and practical solution for TSN. The key insight behind FlexTAS is that relaxing gating does not introduce uncertainty, as long as nonoverlap reserved time slots are guaranteed. FlexTAS is comprised of two main components: first, a novel gating model deviates from the conventional TAS model by incorporating selective relaxation of gating at certain nodes; and second, a deep reinforcement learning-based engine to rapidly generate valid schedules. We build a real testbed and validate the effectiveness of our proposed solution. Our evaluation demonstrates that FlexTAS effectively controls the number of gate entries within the GCL capacity of devices, while simultaneously meeting the Quality of Service(QoS) requirements of time-triggered streams. It significantly reduces the number of GCL entries by 60% to 80%, and facilitates deployment in heterogeneous networks, thus offering a practical solution for TSN.
- 9 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Informatics
- Vol. 21 (7) , 5390-5400
FlexTAS: Flexible Gating Control for Enhanced Time-Sensitive Networking Deployment
Time-sensitive networking (TSN), essential in industrial networks for its promise of reliable and deterministic data transmission, faces deployment challenges due to the limitations of existing time-aware shaper (TAS)-based scheduling algorithms. Specifically, the size of the generated gate control lists (GCLs) is usually too large to be deployed in actual devices. To bridge the gap between theory and practice, we propose FlexTAS, a flexible and practical solution for TSN. The key insight behind FlexTAS is that relaxing gating does not introduce uncertainty, as long as nonoverlap reserved time slots are guaranteed. FlexTAS is comprised of two main components: first, a novel gating model deviates from the conventional TAS model by incorporating selective relaxation of gating at certain nodes; and second, a deep reinforcement learning-based engine to rapidly generate valid schedules. We build a real testbed and validate the effectiveness of our proposed solution. Our evaluation demonstrates that FlexTAS effectively controls the number of gate entries within the GCL capacity of devices, while simultaneously meeting the Quality of Service(QoS) requirements of time-triggered streams. It significantly reduces the number of GCL entries by 60% to 80%, and facilitates deployment in heterogeneous networks, thus offering a practical solution for TSN.