Feilong Tang
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
Feilong Tang has a total of 153 co-authors. Here are the top co-authors by number of shared publications:
- Jiadi Yu
Shanghai Jiao Tong University, Shanghai, China
26 shared publications - Yanmin Zhu
Shanghai Jiao Tong University
25 shared publications - Minyi Guo
Shanghai Jiao Tong University, Shanghai, China
17 shared publications - Chunyang Wang
East China Normal University
7 shared publications - Ilsun You
Kookmin University
7 shared publications - Long Chen
Simon Fraser University, Burnaby, BC, Canada
7 shared publications
Latest Publications
- Open Access
- 6 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 6460-6471
Privacy-Aware Representation Decoupling in Federated Recommendation against Attribute Inference Attacks
- 6 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 6460-6471
Open AccessPrivacy-Aware Representation Decoupling in Federated Recommendation against Attribute Inference Attacks
- Open Access
- 4 February 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Knowledge Discovery From Data
- Vol. 20 (2) , 1-19
MESA: Plugin Meta-Modulation for Transformer-Based Cold-Start Sequential Recommendation
Sequential recommenders aim to enhance prediction accuracy by leveraging user interaction sequences, with transformer-based models showing particularly strong performance. Among them, cold-start sequential recommenders are particularly challenging because these models typically require extensive historical data to perform optimally. Some works attempt to address this issue by enhancing the adaptive ability of the sequence recommenders with meta-learning approaches. However, they are unsuitable for enhancing the popular Transformer-based sequence recommenders: MAML-based models cannot adapt the large number of parameters of Transformers, while transition-based and metric-based meta-learning models rely on unique architectures that are incompatible with Transformer-based frameworks. Also, they usually lack mechanisms to recognize and cater to multiple interests within short interaction sequences. To address these limitations, we propose MESA, a meta-modulation plugin module specifically designed to enhance the cold-start recommendation of Transformer-based sequential recommender systems. (1) We design a meta-modulation method to directly modulate the parameters in Transformer-based sequence encoders, thus enabling the model to adapt more effectively to new users in cold-start scenarios. (2) Additionally, MESA integrates the Mixture of Experts (MoE) mechanism, which refines sequence representations by utilizing multiple experts, each focusing on different aspects of user interests. This structure enhances the personalization of the recommendation by effectively handling diverse user interests within the sequences. Experiments demonstrate the effectiveness of MESA in cold-start scenarios. Our codes are available here : https://github.com/Mushroom-cat/MESA .
- 4 February 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Knowledge Discovery From Data
- Vol. 20 (2) , 1-19
Open AccessMESA: Plugin Meta-Modulation for Transformer-Based Cold-Start Sequential Recommendation
Sequential recommenders aim to enhance prediction accuracy by leveraging user interaction sequences, with transformer-based models showing particularly strong performance. Among them, cold-start sequential recommenders are particularly challenging because these models typically require extensive historical data to perform optimally. Some works attempt to address this issue by enhancing the adaptive ability of the sequence recommenders with meta-learning approaches. However, they are unsuitable for enhancing the popular Transformer-based sequence recommenders: MAML-based models cannot adapt the large number of parameters of Transformers, while transition-based and metric-based meta-learning models rely on unique architectures that are incompatible with Transformer-based frameworks. Also, they usually lack mechanisms to recognize and cater to multiple interests within short interaction sequences. To address these limitations, we propose MESA, a meta-modulation plugin module specifically designed to enhance the cold-start recommendation of Transformer-based sequential recommender systems. (1) We design a meta-modulation method to directly modulate the parameters in Transformer-based sequence encoders, thus enabling the model to adapt more effectively to new users in cold-start scenarios. (2) Additionally, MESA integrates the Mixture of Experts (MoE) mechanism, which refines sequence representations by utilizing multiple experts, each focusing on different aspects of user interests. This structure enhances the personalization of the recommendation by effectively handling diverse user interests within the sequences. Experiments demonstrate the effectiveness of MESA in cold-start scenarios. Our codes are available here : https://github.com/Mushroom-cat/MESA .
- Open Access
- 3 September 2025
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 9 (3) , 1-22
OccMesh: Occlusion-aware Multi-user 3D Human Mesh Reconstruction Using mmWave Signals
Nowadays, 3D human mesh reconstruction technology has drawn great attention in the metaverse for building digital humans. As current mainstream solutions, vision-based systems suffer from privacy leakage and depend on lighting conditions. Towards a more privacy-preserving and illumination-robust manner, recent works have exploited radio frequency signals to realize 3D human mesh reconstruction. However, these studies cannot handle occlusion scenarios with proximity or overlap in multi-user scenarios. This paper presents an occlusion-aware 3D human mesh reconstruction system, OccMesh, which uses mmWave signals to estimate body skeletons and reconstruct human meshes of users with proximity or overlap. In this paper, OccMesh first detects subjects and generates point clouds through mmWave signals, and then segments point clouds to identify and locate each user. OccMesh next estimates human body keypoints of multiple users through a designed RBF Value Estimator. When there are multiple users in close proximity or overlapped, OccMesh infers the occluded human body parts and generates the skeletons of occluded users with a ResGCN-Attention-Block-based Skeleton Inferer. Based on the generated skeletons, OccMesh further reconstructs human meshes for users in occlusion scenarios. Experiments conducted in occlusion scenarios validate the accuracy and robustness of OccMesh in multi-user 3D human mesh reconstruction.
- 3 September 2025
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 9 (3) , 1-22
Open AccessOccMesh: Occlusion-aware Multi-user 3D Human Mesh Reconstruction Using mmWave Signals
Nowadays, 3D human mesh reconstruction technology has drawn great attention in the metaverse for building digital humans. As current mainstream solutions, vision-based systems suffer from privacy leakage and depend on lighting conditions. Towards a more privacy-preserving and illumination-robust manner, recent works have exploited radio frequency signals to realize 3D human mesh reconstruction. However, these studies cannot handle occlusion scenarios with proximity or overlap in multi-user scenarios. This paper presents an occlusion-aware 3D human mesh reconstruction system, OccMesh, which uses mmWave signals to estimate body skeletons and reconstruct human meshes of users with proximity or overlap. In this paper, OccMesh first detects subjects and generates point clouds through mmWave signals, and then segments point clouds to identify and locate each user. OccMesh next estimates human body keypoints of multiple users through a designed RBF Value Estimator. When there are multiple users in close proximity or overlapped, OccMesh infers the occluded human body parts and generates the skeletons of occluded users with a ResGCN-Attention-Block-based Skeleton Inferer. Based on the generated skeletons, OccMesh further reconstructs human meshes for users in occlusion scenarios. Experiments conducted in occlusion scenarios validate the accuracy and robustness of OccMesh in multi-user 3D human mesh reconstruction.
- Open Access
- 13 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 979-988
Generating Difficulty-aware Negative Samples via Conditional Diffusion for Multi-modal Recommendation
- 13 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 979-988
Open AccessGenerating Difficulty-aware Negative Samples via Conditional Diffusion for Multi-modal Recommendation
- Open Access
- 13 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 1728-1737
Social Relation-Level Privacy Risks and Preservation in Social Recommender Systems
- 13 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 1728-1737
Open AccessSocial Relation-Level Privacy Risks and Preservation in Social Recommender Systems
- Open Access
- 9 July 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Information Systems
Dual-Adaptive Update Strategies Enhanced Meta-Optimization for User Cold-Start Recommendation
User cold-start recommendation presents a significant challenge for recommender systems, affecting their overall effectiveness. Meta-learning-based methods have been introduced to address this issue. These methods treat the user cold-start recommendation problem as a few-shot learning task, where each user represents a unique task. The objective is to acquire shared initialization parameters that can be effectively applied across all cold-start users. Subsequently, these shared parameters are fine-tuned into personalized parameters using individual interaction data. Recent studies argue that shared parameters are unsuitable for all users with an implicit grouping distribution of user preference. Therefore, they propose adaptive-initialization-based methods, which first differentiate tasks based on user preferences and then generate task-adaptive initialization parameters using task representations. However, both the meta-learning and adaptive-initialization-based manners ignore discovering the adaptive capability of update strategies in the process of transferring initialization parameters to personalized parameters. Instead, they rely on task-shared optimization strategies, leading the model to fall into an overfitting or underfitting situation. In response to this, we propose a dual-adaptive update strategies enhanced meta-optimization framework (DAUS) for user cold-start recommendation. 1 First, we integrate dual-adaptive update strategies to enhance the adaptive capability of transferring initialization parameters. This involves incorporating both task-adaptive optimization hyperparameters and objectives. Second, we design a multifaceted task encoder , which can provide diverse task information to differentiate between tasks, including explicit task features (task relevance, training signals, etc.) and other implicit task information. Extensive experiments based on three real-world datasets demonstrate that our DAUS outperforms the state-of-the-art methods. The source code is available at https://github.com/XuHao-bit/DAUS .
- 9 July 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Information Systems
Open AccessDual-Adaptive Update Strategies Enhanced Meta-Optimization for User Cold-Start Recommendation
User cold-start recommendation presents a significant challenge for recommender systems, affecting their overall effectiveness. Meta-learning-based methods have been introduced to address this issue. These methods treat the user cold-start recommendation problem as a few-shot learning task, where each user represents a unique task. The objective is to acquire shared initialization parameters that can be effectively applied across all cold-start users. Subsequently, these shared parameters are fine-tuned into personalized parameters using individual interaction data. Recent studies argue that shared parameters are unsuitable for all users with an implicit grouping distribution of user preference. Therefore, they propose adaptive-initialization-based methods, which first differentiate tasks based on user preferences and then generate task-adaptive initialization parameters using task representations. However, both the meta-learning and adaptive-initialization-based manners ignore discovering the adaptive capability of update strategies in the process of transferring initialization parameters to personalized parameters. Instead, they rely on task-shared optimization strategies, leading the model to fall into an overfitting or underfitting situation. In response to this, we propose a dual-adaptive update strategies enhanced meta-optimization framework (DAUS) for user cold-start recommendation. 1 First, we integrate dual-adaptive update strategies to enhance the adaptive capability of transferring initialization parameters. This involves incorporating both task-adaptive optimization hyperparameters and objectives. Second, we design a multifaceted task encoder , which can provide diverse task information to differentiate between tasks, including explicit task features (task relevance, training signals, etc.) and other implicit task information. Extensive experiments based on three real-world datasets demonstrate that our DAUS outperforms the state-of-the-art methods. The source code is available at https://github.com/XuHao-bit/DAUS .
- Open Access
- 30 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 991-999
Towards Effective and Consistent Information Extraction for Social Recommendation: A Minimum and Sufficiency Perspective
- 30 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 991-999
Open AccessTowards Effective and Consistent Information Extraction for Social Recommendation: A Minimum and Sufficiency Perspective
- 4 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Services Computing
- Vol. 18 (4) , 2238-2253
On-Demand and Scalable Topology Control Service for LEO Satellite Network Evolving
Inter-Satellite Links (ISLs) are pivotal for delivering global connectivity services and optimizing resource utilization in 6 G and beyond. However, delivering effective topology control services through ISL provisioning faces critical challenges in sustainability and reliability. Reducing ISLs can conserve energy and extend satellite battery life for Low-Earth-Orbit (LEO) satellites where replacing batteries is impractical. Conversely, increasing ISLs can enhance service reliability but may lead to uneven traffic distribution, overloading nodes, and accelerating battery degradation, ultimately degrading the quality of 6 G services. To tackle this dilemma, we propose TASRI—a service-oriented framework for Traffic-Aware, Sustainable, and Reliable ISL provisioning. TASRI provides a dynamic topology control service by partitioning network topologies into logical zones, enabling flexible ISL activation and deactivation to adapt to varying service demands, ensuring efficient resource utilization and dynamic service orchestration. Using a sustainability-oriented weight model, we formulate the topology control service optimization problem and introduce a scalable on-demand topology evolving algorithm with a bounded approximation ratio. Extensive real-world deployment-based simulation results show that, compared to the state-of-the-art, our TASRI can substantially reduce battery life consumption, while achieving comparable reliability and excellent scalability with considerably fewer ISLs or ISL handovers.
- 4 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Services Computing
- Vol. 18 (4) , 2238-2253
On-Demand and Scalable Topology Control Service for LEO Satellite Network Evolving
Inter-Satellite Links (ISLs) are pivotal for delivering global connectivity services and optimizing resource utilization in 6 G and beyond. However, delivering effective topology control services through ISL provisioning faces critical challenges in sustainability and reliability. Reducing ISLs can conserve energy and extend satellite battery life for Low-Earth-Orbit (LEO) satellites where replacing batteries is impractical. Conversely, increasing ISLs can enhance service reliability but may lead to uneven traffic distribution, overloading nodes, and accelerating battery degradation, ultimately degrading the quality of 6 G services. To tackle this dilemma, we propose TASRI—a service-oriented framework for Traffic-Aware, Sustainable, and Reliable ISL provisioning. TASRI provides a dynamic topology control service by partitioning network topologies into logical zones, enabling flexible ISL activation and deactivation to adapt to varying service demands, ensuring efficient resource utilization and dynamic service orchestration. Using a sustainability-oriented weight model, we formulate the topology control service optimization problem and introduce a scalable on-demand topology evolving algorithm with a bounded approximation ratio. Extensive real-world deployment-based simulation results show that, compared to the state-of-the-art, our TASRI can substantially reduce battery life consumption, while achieving comparable reliability and excellent scalability with considerably fewer ISLs or ISL handovers.
- 29 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 24 (10) , 9368-9382
BAT: A Versatile Bipartite Attention-Based Approach for Comprehensive Truth Inference in Mobile Crowdsourcing
The proliferation of smart mobile devices has catalyzed the growth of Mobile CrowdSourcing (MCS) as a distributed problem-solving paradigm. MCS platforms heavily rely on advanced truth inference techniques to extract reliable information from diverse and potentially noisy crowd-contributed data. Existing truth inference models often made simplified assumptions about workers or tasks, employing complex Bayesian models or stringent data aggregation methods. These approaches tend to be task-specific, primarily limited to categorical labeling, making adaptations to other mobile computing scenarios labor-intensive. To address these limitations, we introduce the Bipartite Attention-driven Truth (BAT), a versatile approach tailored for mobile computing environments. BAT utilizes an Attributed Bipartite Graph (ABG) to holistically model the MCS process, with workers and tasks as nodes connected by edges representing answer-specific attributes. The approach employs a bipartite graph neural network with an innovative attention mechanism to assess the importance of different answers. BAT extends beyond categorical tasks to support numerical ones by incorporating novel feature representations and model extensions. Theoretical analyses clarify the link between answer similarity and worker expertise. Extensive experiments using diverse real-world datasets demonstrate BAT's superior performance compared to state-of-the-art categorical and numerical truth inference models, highlighting its effectiveness in mobile computing scenarios.
- 29 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 24 (10) , 9368-9382
BAT: A Versatile Bipartite Attention-Based Approach for Comprehensive Truth Inference in Mobile Crowdsourcing
The proliferation of smart mobile devices has catalyzed the growth of Mobile CrowdSourcing (MCS) as a distributed problem-solving paradigm. MCS platforms heavily rely on advanced truth inference techniques to extract reliable information from diverse and potentially noisy crowd-contributed data. Existing truth inference models often made simplified assumptions about workers or tasks, employing complex Bayesian models or stringent data aggregation methods. These approaches tend to be task-specific, primarily limited to categorical labeling, making adaptations to other mobile computing scenarios labor-intensive. To address these limitations, we introduce the Bipartite Attention-driven Truth (BAT), a versatile approach tailored for mobile computing environments. BAT utilizes an Attributed Bipartite Graph (ABG) to holistically model the MCS process, with workers and tasks as nodes connected by edges representing answer-specific attributes. The approach employs a bipartite graph neural network with an innovative attention mechanism to assess the importance of different answers. BAT extends beyond categorical tasks to support numerical ones by incorporating novel feature representations and model extensions. Theoretical analyses clarify the link between answer similarity and worker expertise. Extensive experiments using diverse real-world datasets demonstrate BAT's superior performance compared to state-of-the-art categorical and numerical truth inference models, highlighting its effectiveness in mobile computing scenarios.
- 25 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 24 (7) , 5964-5980
Realistic Facial Expression Reconstruction Using Millimeter Wave
The technology of facial expression reconstruction has paved the way for various face-centric applications such as virtual reality (VR) modeling, human-computer interaction, and affective computing. Existing vision-based solutions present challenges in privacy leakage and poor lighting conditions. In this paper, we introduce a nonintrusive facial expression reconstruction system, mm3DFace, which uses a millimeter wave (mmWave) radar to reconstruct facial expressions in a privacy-preserving and passive manner. mm3DFace first captures and pre-processes mmWave signals reflected by a human face, and extracts intricate facial geometric features using a ConvNeXt model integrated with triple loss embedding. Subsequently, mm3DFace derives pose-invariant facial representations utilizing region-divided affine transformation, and further generates individual facial shapes with 68 facial landmarks. Then, dynamic facial expressions with 3D facial avatars are reconstructed to exhibit realistic facial expressions. Finally, mm3DFace enables micro-expression recognition with mmWave signals, which ensures the capability of describing tiny facial changes. Through extensive real-world experiments involving 15 participants, mm3DFace achieves a normalized mean error of 3.94%, a mean absolute error of 2.30 mm, and a 3D-mean absolute error of 4.10 mm in tracking 68 facial landmarks, which demonstrates the efficacy and practicality of mm3DFace in real-world 3D facial reconstruction scenarios.
- 25 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 24 (7) , 5964-5980
Realistic Facial Expression Reconstruction Using Millimeter Wave
The technology of facial expression reconstruction has paved the way for various face-centric applications such as virtual reality (VR) modeling, human-computer interaction, and affective computing. Existing vision-based solutions present challenges in privacy leakage and poor lighting conditions. In this paper, we introduce a nonintrusive facial expression reconstruction system, mm3DFace, which uses a millimeter wave (mmWave) radar to reconstruct facial expressions in a privacy-preserving and passive manner. mm3DFace first captures and pre-processes mmWave signals reflected by a human face, and extracts intricate facial geometric features using a ConvNeXt model integrated with triple loss embedding. Subsequently, mm3DFace derives pose-invariant facial representations utilizing region-divided affine transformation, and further generates individual facial shapes with 68 facial landmarks. Then, dynamic facial expressions with 3D facial avatars are reconstructed to exhibit realistic facial expressions. Finally, mm3DFace enables micro-expression recognition with mmWave signals, which ensures the capability of describing tiny facial changes. Through extensive real-world experiments involving 15 participants, mm3DFace achieves a normalized mean error of 3.94%, a mean absolute error of 2.30 mm, and a 3D-mean absolute error of 4.10 mm in tracking 68 facial landmarks, which demonstrates the efficacy and practicality of mm3DFace in real-world 3D facial reconstruction scenarios.