Pengzhan Zhou
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
Pengzhan Zhou has a total of 63 co-authors. Here are the top co-authors by number of shared publications:
- Songtao Guo
College of Computer Science, Chongqing University, Chongqing, China
12 shared publications - Zhida Qin
School of Computer Science, Beijing Institute of Technology, Beijing, China
8 shared publications - Mingyan Li
Key Laboratory of Dependable Service Computing in Cyber-Physical-Society (Ministry of Education) and College of Computer Science, Chongqing University, Chongqing, China
6 shared publications - Tianyu Huang
School of Computer Science, Beijing Institute of Technology, Beijing, China
5 shared publications - Yuanyuan Yang
Department of ECE, Stony Brook University, Stony Brook, NY, USA
5 shared publications - Lin Tan
School of Computer and Information Engineering, Central South University of Forestry and Technology, Changsha, China
3 shared publications
Latest Publications
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 7776-7785
Counterfactual-Guided Diversified Path Integration for Transferable Neuron Attribution Attacks
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 7776-7785
Open AccessCounterfactual-Guided Diversified Path Integration for Transferable Neuron Attribution Attacks
- Open Access
- 18 May 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Intelligent Systems and Technology
- Vol. 17 (4) , 1-27
Causal Disentanglement-Enhanced Diffusion Denoising for Social Recommendation
In recent years, social recommendation systems have emerged as a pivotal technology for enhancing recommendation accuracy by leveraging user social homophily and influence. Although many works have been devoted to this area, existing works still struggle to extract the beneficial structural information from social relationships that is beneficial for recommendations and neglect the inherent popularity bias in the social networks, which leads to suboptimal recommendation performances. To address these challenges, we propose a novel framework termed Causal Disentanglement-Enhanced Diffusion Denoising for Social Recommendation (CaDDiSR). This framework first employs causal graphs to disentangle the complexities of social relationships, generating user representations with high-order structures, which are subsequently used as inputs to a diffusion process to effectively denoise social networks and retain social signals beneficial for recommendation tasks. Furthermore, the framework integrates a bidirectional knowledge distillation mechanism, which balances user representations between social and recommendation contexts, thereby facilitating the effective fusion of their respective advantages while simultaneously mitigating noise interference and enhancing overall system performance. Finally, cross-domain contrastive learning is utilized to optimize user and item representations, ensuring consistency in recommendation performance across diverse scenarios. Experimental results on multiple real-world datasets demonstrate that CaDDiSR significantly outperforms existing baseline models, substantiating its superior performance.
- 18 May 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Intelligent Systems and Technology
- Vol. 17 (4) , 1-27
Open AccessCausal Disentanglement-Enhanced Diffusion Denoising for Social Recommendation
In recent years, social recommendation systems have emerged as a pivotal technology for enhancing recommendation accuracy by leveraging user social homophily and influence. Although many works have been devoted to this area, existing works still struggle to extract the beneficial structural information from social relationships that is beneficial for recommendations and neglect the inherent popularity bias in the social networks, which leads to suboptimal recommendation performances. To address these challenges, we propose a novel framework termed Causal Disentanglement-Enhanced Diffusion Denoising for Social Recommendation (CaDDiSR). This framework first employs causal graphs to disentangle the complexities of social relationships, generating user representations with high-order structures, which are subsequently used as inputs to a diffusion process to effectively denoise social networks and retain social signals beneficial for recommendation tasks. Furthermore, the framework integrates a bidirectional knowledge distillation mechanism, which balances user representations between social and recommendation contexts, thereby facilitating the effective fusion of their respective advantages while simultaneously mitigating noise interference and enhancing overall system performance. Finally, cross-domain contrastive learning is utilized to optimize user and item representations, ensuring consistency in recommendation performance across diverse scenarios. Experimental results on multiple real-world datasets demonstrate that CaDDiSR significantly outperforms existing baseline models, substantiating its superior performance.
- Open Access
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 6091-6102
Bridging Time and Domains: A Time-aware Framework for Cross-Domain Sequential Recommendation
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 6091-6102
Open AccessBridging Time and Domains: A Time-aware Framework for Cross-Domain Sequential Recommendation
- 3 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Consumer Electronics
- Vol. PP (00983063) , 1
RFM: Refined Feature Modeling for Multimodal Recommendation
This paper addresses the challenge of multimodal recommendation in consumer-facing e-commerce platforms, where users interact with item information across multiple modalities—such as images, textual descriptions, and specifications. Effectively modeling these items and user preferences requires robust techniques for extracting meaningful multimodal features. To address this, we propose RFM (Refined Feature for E-commerce Multimodal Recommendation), a framework that improves user and item modeling through attention-based cross-layer knowledge distillation. RFM enables a student network to selectively learn from semantically aligned teacher layers, capturing features most relevant to user preferences while avoiding noisy or mismatched signals. These refined features are then used as item embeddings in a dual-graph learning structure, which includes both an item-item similarity graph and a user-item interaction graph. This design enhances the expressive power of item representations while preserving collaborative signals. RFM is particularly applicable to consumer-facing e-commerce platforms like Amazon, where users interact with multimodal content to evaluate a wide range of consumer electronics products. Extensive experiments on five real-world Amazon datasets demonstrate that RFM consistently outperforms strong baselines, achieving an average improvement of 5.8% in Recall@20. These results highlight the practical effectiveness of RFM for enhancing recommendation quality in online retail environments.
- 3 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Consumer Electronics
- Vol. PP (00983063) , 1
RFM: Refined Feature Modeling for Multimodal Recommendation
This paper addresses the challenge of multimodal recommendation in consumer-facing e-commerce platforms, where users interact with item information across multiple modalities—such as images, textual descriptions, and specifications. Effectively modeling these items and user preferences requires robust techniques for extracting meaningful multimodal features. To address this, we propose RFM (Refined Feature for E-commerce Multimodal Recommendation), a framework that improves user and item modeling through attention-based cross-layer knowledge distillation. RFM enables a student network to selectively learn from semantically aligned teacher layers, capturing features most relevant to user preferences while avoiding noisy or mismatched signals. These refined features are then used as item embeddings in a dual-graph learning structure, which includes both an item-item similarity graph and a user-item interaction graph. This design enhances the expressive power of item representations while preserving collaborative signals. RFM is particularly applicable to consumer-facing e-commerce platforms like Amazon, where users interact with multimodal content to evaluate a wide range of consumer electronics products. Extensive experiments on five real-world Amazon datasets demonstrate that RFM consistently outperforms strong baselines, achieving an average improvement of 5.8% in Recall@20. These results highlight the practical effectiveness of RFM for enhancing recommendation quality in online retail environments.
- 23 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 25 (6) , 8177-8191
HierTFL: Hierarchical Scheduling for Industrial TSN-Enhanced Federated Learning System
The adoption of federated learning (FL) in industrial IoT (IIoT) facilitates the deployment of field-level industrial intelligence by multi-node collaborative and distributed learning. Numerous studies on FL primarily concentrate on enhancing model accuracy under non-independent and identically distributed data. Nevertheless, this focus is inadequate for time-critical industrial systems, as these systems necessitate real-time processing during the FL model training phase and any strategy that prioritizes incremental accuracy gains at the expense of latency risks violating real-time deadlines. However, the system heterogeneity of computing and communication capabilities will impose a formidable bottleneck to the overall time consumed for FL model training. In this paper, we present a FL-enabled Time-Sensitive IIoT (FETI) framework that integrates FL with Time-Sensitive Networking (TSN) to support the deterministic forwarding of FL flows in industry. Aiming to speed up FL convergence within a targeted accuracy gap, we formulate a heterogeneity-aware FL-TSN joint optimization problem, which is theoretically transformed into a stochastic mixed-integer programming problem solved at each FL round. To address this problem, we propose a hierarchical reinforcement learning-based scheduling scheme, called HierTFL, with two interacting layers of policies. With the assistance of a proposed spatial-temporal state encoder, the high-level policy dynamically selects client participants based on data quality and resource availability in each FL round, while the low-level policy optimizes TSN flow scheduling of each selected client using non-cumulative Bellman updates. Experimental results under industrial monitoring datasets have shown the effectivity of HierTFL in achieving a balanced trade-off between model precision and convergence time compared to existing benchmarks.
- 23 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 25 (6) , 8177-8191
HierTFL: Hierarchical Scheduling for Industrial TSN-Enhanced Federated Learning System
The adoption of federated learning (FL) in industrial IoT (IIoT) facilitates the deployment of field-level industrial intelligence by multi-node collaborative and distributed learning. Numerous studies on FL primarily concentrate on enhancing model accuracy under non-independent and identically distributed data. Nevertheless, this focus is inadequate for time-critical industrial systems, as these systems necessitate real-time processing during the FL model training phase and any strategy that prioritizes incremental accuracy gains at the expense of latency risks violating real-time deadlines. However, the system heterogeneity of computing and communication capabilities will impose a formidable bottleneck to the overall time consumed for FL model training. In this paper, we present a FL-enabled Time-Sensitive IIoT (FETI) framework that integrates FL with Time-Sensitive Networking (TSN) to support the deterministic forwarding of FL flows in industry. Aiming to speed up FL convergence within a targeted accuracy gap, we formulate a heterogeneity-aware FL-TSN joint optimization problem, which is theoretically transformed into a stochastic mixed-integer programming problem solved at each FL round. To address this problem, we propose a hierarchical reinforcement learning-based scheduling scheme, called HierTFL, with two interacting layers of policies. With the assistance of a proposed spatial-temporal state encoder, the high-level policy dynamically selects client participants based on data quality and resource availability in each FL round, while the low-level policy optimizes TSN flow scheduling of each selected client using non-cumulative Bellman updates. Experimental results under industrial monitoring datasets have shown the effectivity of HierTFL in achieving a balanced trade-off between model precision and convergence time compared to existing benchmarks.
- Open Access
- 13 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 1944-1953
Large Language Models Enhanced Hyperbolic Space Recommender Systems
- 13 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 1944-1953
Open AccessLarge Language Models Enhanced Hyperbolic Space Recommender Systems
- 26 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Wireless Communications
- Vol. 25 (15361276) , 90-106
SkyLink: Joint Deployment and Scheduling in Collaborative Integrated Ground-Air-Space Network
Low Earth Orbit (LEO) satellite networks hold great promise in the field of wireless communication due to their global coverage. However, the long communication distances and massive data computations present significant challenges for current satellite networks. To overcome these barriers, we propose SkyLink, a universal Integrated Ground-Air-Space Collaborative Edge Computing system that leverages horizontal collaboration among aerial platforms (AirXs) as well as vertical collaboration among Ground-Air-Space. We propose a bi-level optimization framework based on a Multi-Agent Twin Delayed Deep Deterministic policy gradient (MATD3) with Hybrid Action Space and constructe a latent representation space for each agent to allow the agent to learn the latent policy. This enabling each AirX to act as an agent and autonomously optimize its hybrid action decisions to improve system efficiency in real-time based on the dynamic network environment, a capability not achievable by conventional DRL methods. This includes continuous optimization variables such as AirX deployment (location changes) and resource allocation, as well as discrete optimization variables for collaborative task offloading decisions. Extensive experiments against state-of-the-art algorithms (e.g., MADDPG, QMIX) demonstrate that the proposed system improves energy efficiency by 27.2% and task completion rate by 6.8% compared to traditional Integrated Ground-Air-Space (IG) Network.
- 26 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Wireless Communications
- Vol. 25 (15361276) , 90-106
SkyLink: Joint Deployment and Scheduling in Collaborative Integrated Ground-Air-Space Network
Low Earth Orbit (LEO) satellite networks hold great promise in the field of wireless communication due to their global coverage. However, the long communication distances and massive data computations present significant challenges for current satellite networks. To overcome these barriers, we propose SkyLink, a universal Integrated Ground-Air-Space Collaborative Edge Computing system that leverages horizontal collaboration among aerial platforms (AirXs) as well as vertical collaboration among Ground-Air-Space. We propose a bi-level optimization framework based on a Multi-Agent Twin Delayed Deep Deterministic policy gradient (MATD3) with Hybrid Action Space and constructe a latent representation space for each agent to allow the agent to learn the latent policy. This enabling each AirX to act as an agent and autonomously optimize its hybrid action decisions to improve system efficiency in real-time based on the dynamic network environment, a capability not achievable by conventional DRL methods. This includes continuous optimization variables such as AirX deployment (location changes) and resource allocation, as well as discrete optimization variables for collaborative task offloading decisions. Extensive experiments against state-of-the-art algorithms (e.g., MADDPG, QMIX) demonstrate that the proposed system improves energy efficiency by 27.2% and task completion rate by 6.8% compared to traditional Integrated Ground-Air-Space (IG) Network.
- 25 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. PP (15361233) , 1-18
FedRAV: Hierarchically Federated Region-Learning for Traffic Object Classification of Personalized Autonomous Vehicles with Guaranteed Efficiency
The emerging federated learning enables distributed autonomous vehicles to train equipped deep learning models collaboratively without exposing their raw data, providing great potential for utilizing explosively growing autonomous driving data. However, considering the complicated traffic environments and driving scenarios, deploying federated learning for autonomous vehicles is inevitably challenged by non-independent and identically distributed (Non-IID) data of vehicles, which may lead to failed convergence and low training accuracy. In this paper, we propose a novel hierarchically Federated Region-learning framework of Autonomous Vehicles (FedRAV) that adaptively divides a large area containing vehicles into sub-regions based on the defined region-wise distance, and achieves personalized vehicular models and regional models. Specifically, the architecture employs a designated hypernetwork to learn personalized mask vectors per vehicle used in the linear combination of models shared by vehicles in the same region. This approach ensures that the updated vehicular model adopts the beneficial models while discarding the unprofitable ones. We validate our FedRAV framework against existing federated learning algorithms on four real-world autonomous driving datasets in various heterogeneous settings. Extensive experiment results demonstrate that FedRAV framework achieves superior performance than the state-of-the-art algorithms, and improves the accuracy by 9.36%. The source code of FedRAV is available at: https://github.com/yjzhai-cs/FedRAV.
- 25 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. PP (15361233) , 1-18
FedRAV: Hierarchically Federated Region-Learning for Traffic Object Classification of Personalized Autonomous Vehicles with Guaranteed Efficiency
The emerging federated learning enables distributed autonomous vehicles to train equipped deep learning models collaboratively without exposing their raw data, providing great potential for utilizing explosively growing autonomous driving data. However, considering the complicated traffic environments and driving scenarios, deploying federated learning for autonomous vehicles is inevitably challenged by non-independent and identically distributed (Non-IID) data of vehicles, which may lead to failed convergence and low training accuracy. In this paper, we propose a novel hierarchically Federated Region-learning framework of Autonomous Vehicles (FedRAV) that adaptively divides a large area containing vehicles into sub-regions based on the defined region-wise distance, and achieves personalized vehicular models and regional models. Specifically, the architecture employs a designated hypernetwork to learn personalized mask vectors per vehicle used in the linear combination of models shared by vehicles in the same region. This approach ensures that the updated vehicular model adopts the beneficial models while discarding the unprofitable ones. We validate our FedRAV framework against existing federated learning algorithms on four real-world autonomous driving datasets in various heterogeneous settings. Extensive experiment results demonstrate that FedRAV framework achieves superior performance than the state-of-the-art algorithms, and improves the accuracy by 9.36%. The source code of FedRAV is available at: https://github.com/yjzhai-cs/FedRAV.
- 10 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (9) , 13061-13074
EMAFL: Evolutionary Momentum Auxiliary Adaptive Accelerating Federated Learning
The utilization of federated learning (FL) has witnessed notable advancements in the domain of edge computing (EC). However, limited edge resources and heterogeneous devices restrict the accelerated training of the FL model. To address this issue, we introduce the biological evolutionary mechanism and momentum gradient descent (MGD) update approach into FL, called the EMAFL scheme, aiming to achieve accelerated model training and maximize resource utilization, simultaneously. Specifically, we first update the local model with particle swarm optimization (PSO) for each device and perform MGD on the updated local model. Next, by a toy example, we illustrate the necessity of adopting the different number of local iterations for heterogeneous devices in a resource-limited environment. Analytical convergence of the EMAFL scheme, premised on a delineated resource budget is subsequently explored. This yields a mathematical delineation correlating the quantity of local iterations for heterogeneous devices with the optimal model parameters. Predicated on the prior theoretical examinations, an adaptive control algorithm is devised to ascertain the local iteration count pertinent to each device following every communication round. Finally, through a lot of experiments compared with the benchmarks, the advantages of EMAFL in model accuracy, resource consumption, and Non-IID issues are verified.
- 10 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (9) , 13061-13074
EMAFL: Evolutionary Momentum Auxiliary Adaptive Accelerating Federated Learning
The utilization of federated learning (FL) has witnessed notable advancements in the domain of edge computing (EC). However, limited edge resources and heterogeneous devices restrict the accelerated training of the FL model. To address this issue, we introduce the biological evolutionary mechanism and momentum gradient descent (MGD) update approach into FL, called the EMAFL scheme, aiming to achieve accelerated model training and maximize resource utilization, simultaneously. Specifically, we first update the local model with particle swarm optimization (PSO) for each device and perform MGD on the updated local model. Next, by a toy example, we illustrate the necessity of adopting the different number of local iterations for heterogeneous devices in a resource-limited environment. Analytical convergence of the EMAFL scheme, premised on a delineated resource budget is subsequently explored. This yields a mathematical delineation correlating the quantity of local iterations for heterogeneous devices with the optimal model parameters. Predicated on the prior theoretical examinations, an adaptive control algorithm is devised to ascertain the local iteration count pertinent to each device following every communication round. Finally, through a lot of experiments compared with the benchmarks, the advantages of EMAFL in model accuracy, resource consumption, and Non-IID issues are verified.
- 2 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 24 (4) , 3390-3406
ASMAFL: Adaptive Staleness-Aware Momentum Asynchronous Federated Learning in Edge Computing
Compared with synchronous federated learning (FL), asynchronous FL (AFL) has attracted more and more attention in edge computing (EC) fields because of its strong adaptability to heterogeneous application scenarios. However, the non-independent and identically distributed (Non-IID) data across devices and the staleness-aware estimation of unreliable wireless connections and limited edge resources make it much more difficult to achieve better AFL-related applications. To handle this problem, we propose an Adaptive Staleness-aware Momentum Accelerated AFL (ASMAFL) algorithm to reduce the resources consumption of heterogeneous wireless communication EC (WCEC) scenarios, as well as decrease the negative impact of Non-IID data for model training. Specifically, we first introduce the staleness-aware parameter and a unified momentum gradient descent (GD) framework to reformulate AFL. Then, we establish global convergence properties of AFL, derive an upper bound on AFL convergence rate, and find that the bound is related to the staleness-aware parameter and Non-IIDness. Next, we formulate the bound into a minimization problem of resource consumption under given model accuracy, and the corresponding staleness-aware parameter of devices will be recomputed after each asynchronous aggregation to eliminate the differences of local models’ contribution to global model aggregation. Finally, extensive experiments are carried out to validate the superiority of ASMAFL in model accuracy, convergence rate, resources consumption, Non-IID issue, etc.
- 2 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 24 (4) , 3390-3406
ASMAFL: Adaptive Staleness-Aware Momentum Asynchronous Federated Learning in Edge Computing
Compared with synchronous federated learning (FL), asynchronous FL (AFL) has attracted more and more attention in edge computing (EC) fields because of its strong adaptability to heterogeneous application scenarios. However, the non-independent and identically distributed (Non-IID) data across devices and the staleness-aware estimation of unreliable wireless connections and limited edge resources make it much more difficult to achieve better AFL-related applications. To handle this problem, we propose an Adaptive Staleness-aware Momentum Accelerated AFL (ASMAFL) algorithm to reduce the resources consumption of heterogeneous wireless communication EC (WCEC) scenarios, as well as decrease the negative impact of Non-IID data for model training. Specifically, we first introduce the staleness-aware parameter and a unified momentum gradient descent (GD) framework to reformulate AFL. Then, we establish global convergence properties of AFL, derive an upper bound on AFL convergence rate, and find that the bound is related to the staleness-aware parameter and Non-IIDness. Next, we formulate the bound into a minimization problem of resource consumption under given model accuracy, and the corresponding staleness-aware parameter of devices will be recomputed after each asynchronous aggregation to eliminate the differences of local models’ contribution to global model aggregation. Finally, extensive experiments are carried out to validate the superiority of ASMAFL in model accuracy, convergence rate, resources consumption, Non-IID issue, etc.