Xi Chen
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Xi Chen has a total of 48 co-authors. Here are the top co-authors by number of shared publications:
- Xue Liu
School of Computer Science, McGill University, Montréal, Canada
7 shared publications - Chen Ma
City University of Hong Kong
3 shared publications - Guihai Chen
Shanghai Jiao Tong University
2 shared publications - Jun Yan
Concordia University
2 shared publications - Linghe Kong
Shanghai Jiao Tong University
2 shared publications Qiao Xiang
School of Informatics, Xiamen University, Xiamen, China
2 shared publications
Latest Publications
- Open Access
- 14 March 2026
- Published by Springer Nature in Machine Learning
- Vol. 115 (3) , 72
FedBNR: A Fully Global Federated Gaussian Process
- 14 March 2026
- Published by Springer Nature in Machine Learning
- Vol. 115 (3) , 72
Open AccessFedBNR: A Fully Global Federated Gaussian Process
- 26 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. 75 (6) , 12025-12030
Self-Refined Generative Foundation Models for Wireless Traffic Prediction
With a broad range of emerging applications in 6G networks, wireless traffic prediction has become a critical component of network management. However, the dynamically shifting distribution of wireless traffic in non-stationary 6G networks presents significant challenges to achieving accurate and stable predictions. Motivated by recent advancements in Generative AI (GenAI)-enabled 6G networks, this paper proposes a novel self-refined Large Language Model (LLM) for wireless traffic prediction, namely TrafficLLM, through in-context learning without parameter fine-tuning or model training. The proposed TrafficLLM harnesses the powerful few-shot learning abilities of LLMs to enhance the scalability of traffic prediction in dynamically changing wireless environments. Specifically, our proposed TrafficLLM embraces an LLM to iteratively refine its predictions through a three-step process: traffic prediction, feedback generation, and prediction refinement. Initially, the proposed TrafficLLM conducts traffic predictions using task-specific demonstration prompts. Recognizing that LLMs may generate incorrect predictions on the first attempt, this paper designs feedback demonstration prompts to provide multifaceted and valuable feedback related to these initial predictions. The validation scheme is further incorporated to systematically enhance the accuracy of mathematical calculations during the feedback generation process. Following this comprehensive feedback, our proposed TrafficLLM introduces refinement demonstration prompts, enabling the same LLM to further refine its predictions and thereby enhance prediction performance. Evaluations on two realistic datasets demonstrate that the proposed TrafficLLM outperforms LLM-based in-context learning methods, achieving performance improvements of 23.17% and 17.09%, respectively.
- 26 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. 75 (6) , 12025-12030
Self-Refined Generative Foundation Models for Wireless Traffic Prediction
With a broad range of emerging applications in 6G networks, wireless traffic prediction has become a critical component of network management. However, the dynamically shifting distribution of wireless traffic in non-stationary 6G networks presents significant challenges to achieving accurate and stable predictions. Motivated by recent advancements in Generative AI (GenAI)-enabled 6G networks, this paper proposes a novel self-refined Large Language Model (LLM) for wireless traffic prediction, namely TrafficLLM, through in-context learning without parameter fine-tuning or model training. The proposed TrafficLLM harnesses the powerful few-shot learning abilities of LLMs to enhance the scalability of traffic prediction in dynamically changing wireless environments. Specifically, our proposed TrafficLLM embraces an LLM to iteratively refine its predictions through a three-step process: traffic prediction, feedback generation, and prediction refinement. Initially, the proposed TrafficLLM conducts traffic predictions using task-specific demonstration prompts. Recognizing that LLMs may generate incorrect predictions on the first attempt, this paper designs feedback demonstration prompts to provide multifaceted and valuable feedback related to these initial predictions. The validation scheme is further incorporated to systematically enhance the accuracy of mathematical calculations during the feedback generation process. Following this comprehensive feedback, our proposed TrafficLLM introduces refinement demonstration prompts, enabling the same LLM to further refine its predictions and thereby enhance prediction performance. Evaluations on two realistic datasets demonstrate that the proposed TrafficLLM outperforms LLM-based in-context learning methods, achieving performance improvements of 23.17% and 17.09%, respectively.
- Open Access
- 11 September 2025
- Published by Institution of Engineering and Technology (IET) in Artificial Intelligence for Engineering
- Vol. 1 (1) , 7-18
Does Combining Parameter‐Efficient Modules Improve Few‐Shot Transfer Accuracy?
Parameter‐efficient fine‐tuning has become a key technique for adapting large language and vision models to diverse downstream tasks, including emerging applications in engineering, such as design optimisation, fault detection in complex systems, material property prediction and autonomous robotics control. Specifically, the efficiency of low‐rank adaptation has facilitated the creation and sharing of hundreds of custom low‐rank adaptation (LoRA) modules, each trained on distinct data from various downstream tasks. In this paper, we explore the composability of LoRA modules, examining if combining these pre‐trained modules enhances generalisation to unseen downstream tasks. Our investigation involves evaluating two approaches: (a) uniform composition , involving averaging upstream LoRA modules with equal weights and (b) learned composition , where we learn the weights for each upstream module and perform weighted averaging. Through experiments on both vision and language models, we demonstrate that in few‐shot settings—common in many engineering workflows such as structural design, material discovery and system diagnostics—both composition methods consistently outperform full fine‐tuning and training LoRA from scratch. Moreover, in full‐shot settings, learned composition matches the performance of standard LoRA training while requiring significantly fewer trainable parameters. These results highlight the potential of compositional parameter‐efficient adaptation as a scalable and flexible approach for AI‐driven engineering tasks, enabling faster deployment, reduced data requirements and improved model reusability across diverse domains.
- 11 September 2025
- Published by Institution of Engineering and Technology (IET) in Artificial Intelligence for Engineering
- Vol. 1 (1) , 7-18
Open AccessDoes Combining Parameter‐Efficient Modules Improve Few‐Shot Transfer Accuracy?
Parameter‐efficient fine‐tuning has become a key technique for adapting large language and vision models to diverse downstream tasks, including emerging applications in engineering, such as design optimisation, fault detection in complex systems, material property prediction and autonomous robotics control. Specifically, the efficiency of low‐rank adaptation has facilitated the creation and sharing of hundreds of custom low‐rank adaptation (LoRA) modules, each trained on distinct data from various downstream tasks. In this paper, we explore the composability of LoRA modules, examining if combining these pre‐trained modules enhances generalisation to unseen downstream tasks. Our investigation involves evaluating two approaches: (a) uniform composition , involving averaging upstream LoRA modules with equal weights and (b) learned composition , where we learn the weights for each upstream module and perform weighted averaging. Through experiments on both vision and language models, we demonstrate that in few‐shot settings—common in many engineering workflows such as structural design, material discovery and system diagnostics—both composition methods consistently outperform full fine‐tuning and training LoRA from scratch. Moreover, in full‐shot settings, learned composition matches the performance of standard LoRA training while requiring significantly fewer trainable parameters. These results highlight the potential of compositional parameter‐efficient adaptation as a scalable and flexible approach for AI‐driven engineering tasks, enabling faster deployment, reduced data requirements and improved model reusability across diverse domains.
- 8 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 25 (2) , 2238-2254
MaestroBot: Generalized Gesture-Driven Hierarchical Coordination for Robotic Formations
Robotic swarm coordination holds transformative potential for applications such as warehouse automation, search & rescue, and entertainment. However, approaches relying on wearable devices or vision-based systems are often constrained by hardware-intensive, high computational requirements, reliance on line-of-sight, and privacy concerns. Wireless sensing, particularly using Channel State Information (CSI), offers a promising alternative by translating environmental perturbations into CSI variation data. Nevertheless, existing CSI-based systems face significant challenges in domain adaptation, resource limitation, and scalability issues. This paper introduces MaestroBot, a hierarchical motion coordination system that combines distributed CSI-based wireless sensing with domain-adaptive learning to address these limitations. For leader robots, the system features a lightweight hand gesture recognition model, built on a “Hybrid-Single” knowledge distillation framework, achieving up to 95.87% accuracy while maintaining adaptability across diverse domains. For follower robots, the hierarchical motion propagation model leverages localized CSI analysis and dual-layer error correction mechanisms to deliver 97.2% accuracy with a low latency of 0.085 seconds, even in multi-row formations. Additionally, its cost-effective hardware design ensures practical scalability and real-world deployability. These results position MaestroBot as an efficient, robust, and privacy-preserving solution for large-scale robotic swarm coordination in dynamic environments.
- 8 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 25 (2) , 2238-2254
MaestroBot: Generalized Gesture-Driven Hierarchical Coordination for Robotic Formations
Robotic swarm coordination holds transformative potential for applications such as warehouse automation, search & rescue, and entertainment. However, approaches relying on wearable devices or vision-based systems are often constrained by hardware-intensive, high computational requirements, reliance on line-of-sight, and privacy concerns. Wireless sensing, particularly using Channel State Information (CSI), offers a promising alternative by translating environmental perturbations into CSI variation data. Nevertheless, existing CSI-based systems face significant challenges in domain adaptation, resource limitation, and scalability issues. This paper introduces MaestroBot, a hierarchical motion coordination system that combines distributed CSI-based wireless sensing with domain-adaptive learning to address these limitations. For leader robots, the system features a lightweight hand gesture recognition model, built on a “Hybrid-Single” knowledge distillation framework, achieving up to 95.87% accuracy while maintaining adaptability across diverse domains. For follower robots, the hierarchical motion propagation model leverages localized CSI analysis and dual-layer error correction mechanisms to deliver 97.2% accuracy with a low latency of 0.085 seconds, even in multi-row formations. Additionally, its cost-effective hardware design ensures practical scalability and real-world deployability. These results position MaestroBot as an efficient, robust, and privacy-preserving solution for large-scale robotic swarm coordination in dynamic environments.
- Open Access
Less or More From Teacher: Exploiting Trilateral Geometry For Knowledge Distillation
Knowledge distillation aims to train a compact student network using soft supervision from a larger teacher network and hard supervision from ground truths. However, determining an optimal knowledge fusion ratio that balances these supervisory signals remains challenging. Prior methods generally resort to a constant or heuristic-based fusion ratio, which often falls short of a proper balance. In this study, we introduce a novel adaptive method for learning a sample-wise knowledge fusion ratio, exploiting both the correctness of teacher and student, as well as how well the student mimics the teacher on each sample. Our method naturally leads to the intra-sample trilateral geometric relations among the student prediction ($S$), teacher prediction ($T$), and ground truth ($G$). To counterbalance the impact of outliers, we further extend to the inter-sample relations, incorporating the teacher's global average prediction $\bar{T}$ for samples within the same class. A simple neural network then learns the implicit mapping from the intra- and inter-sample relations to an adaptive, sample-wise knowledge fusion ratio in a bilevel-optimization manner. Our approach provides a simple, practical, and adaptable solution for knowledge distillation that can be employed across various architectures and model sizes. Extensive experiments demonstrate consistent improvements over other loss re-weighting methods on image classification, attack detection, and click-through rate prediction.Open AccessLess or More From Teacher: Exploiting Trilateral Geometry For Knowledge Distillation
Knowledge distillation aims to train a compact student network using soft supervision from a larger teacher network and hard supervision from ground truths. However, determining an optimal knowledge fusion ratio that balances these supervisory signals remains challenging. Prior methods generally resort to a constant or heuristic-based fusion ratio, which often falls short of a proper balance. In this study, we introduce a novel adaptive method for learning a sample-wise knowledge fusion ratio, exploiting both the correctness of teacher and student, as well as how well the student mimics the teacher on each sample. Our method naturally leads to the intra-sample trilateral geometric relations among the student prediction ($S$), teacher prediction ($T$), and ground truth ($G$). To counterbalance the impact of outliers, we further extend to the inter-sample relations, incorporating the teacher's global average prediction $\bar{T}$ for samples within the same class. A simple neural network then learns the implicit mapping from the intra- and inter-sample relations to an adaptive, sample-wise knowledge fusion ratio in a bilevel-optimization manner. Our approach provides a simple, practical, and adaptable solution for knowledge distillation that can be employed across various architectures and model sizes. Extensive experiments demonstrate consistent improvements over other loss re-weighting methods on image classification, attack detection, and click-through rate prediction. - 6 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 11 (5) , 8268-8287
FedSwarm: An Adaptive Federated Learning Framework for Scalable AIoT
Federated learning (FL) is a key solution for datadriven the Artificial Intelligence of Things (AIoT). Although much progress has been made, scalability remains a core challenge for real-world FL deployments. Existing solutions either suffer from accuracy loss or do not fully address the connectivity dynamicity of FL systems. In this article, we tackle the scalability issue with a novel, adaptive FL framework called FedSwarm, which improves system scalability for AIoT by deploying multiple collaborative edge servers. FedSwarm has two novel features: 1) adaptiveness on the number of local updates and 2) dynamicity of the synchronization between edge devices and edge servers. We formulate FedSwarm as a local update adaptation and perdevice dynamic server selection problem and prove FedSwarm‘s convergence bound. We further design a control mechanism consisting of a learning-based algorithm for collaboratively providing local update adaptation on the servers’ side and a bonus-based strategy for spurring dynamic per-device server selection on the devices’ side. Our extensive evaluation shows that FedSwarm significantly outperforms other studies with better scalability, lower energy consumption, and higher model accuracy.
- 6 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 11 (5) , 8268-8287
FedSwarm: An Adaptive Federated Learning Framework for Scalable AIoT
Federated learning (FL) is a key solution for datadriven the Artificial Intelligence of Things (AIoT). Although much progress has been made, scalability remains a core challenge for real-world FL deployments. Existing solutions either suffer from accuracy loss or do not fully address the connectivity dynamicity of FL systems. In this article, we tackle the scalability issue with a novel, adaptive FL framework called FedSwarm, which improves system scalability for AIoT by deploying multiple collaborative edge servers. FedSwarm has two novel features: 1) adaptiveness on the number of local updates and 2) dynamicity of the synchronization between edge devices and edge servers. We formulate FedSwarm as a local update adaptation and perdevice dynamic server selection problem and prove FedSwarm‘s convergence bound. We further design a control mechanism consisting of a learning-based algorithm for collaboratively providing local update adaptation on the servers’ side and a bonus-based strategy for spurring dynamic per-device server selection on the devices’ side. Our extensive evaluation shows that FedSwarm significantly outperforms other studies with better scalability, lower energy consumption, and higher model accuracy.
- Open Access
- 16 May 2023
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Sensor Networks
- Vol. 19 (4) , 1-20
Eliminating Space Scanning: Fast mmWave Beam Alignment with UWB Radios
Due to their large bandwidth and impressive data speed, millimeter-wave (mmWave) radios are expected to play a key role in the 5G and beyond (e.g., 6G) communication networks. Yet, to release mmWave’s true power, the highly directional mmWave beams need to be aligned perfectly. Most existing beam alignment methods adopt an exhaustive or semi-exhaustive space scanning, which introduces up to seconds of delays. To eliminate the need for complex space scanning, this article presents an Ultra-wideband (UWB)-assisted mmWave communication framework, which leverages the co-located UWB antennas to estimate the best angles for mmWave beam alignment. One major challenge of applying this idea in the real world is the barrier of limited antenna numbers. Commercial-Off-The-Shelf (COTS) devices are usually equipped with only a small number of UWB antennas, which are not enough for the existing algorithms to provide an accurate angle estimation. To solve this challenge, we design a novel Multi-Frequency MUltiple SIgnal Classification (MF-MUSIC) algorithm, which extends the classic MUltiple SIgnal Classification (MUSIC) algorithm to the frequency domain and overcomes the antenna limitation barrier in the spatial domain. Extensive real-world experiments and numerical simulations illustrate the advantage of the proposed MF-MUSIC algorithm. MF-MUSIC uses only three antennas to achieve an accurate angle estimation, which is a mere 0.15° (or a relative difference of 3.6%) different from the state-of-the-art 16-antenna-based angle estimation method.
- 16 May 2023
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Sensor Networks
- Vol. 19 (4) , 1-20
Open AccessEliminating Space Scanning: Fast mmWave Beam Alignment with UWB Radios
Due to their large bandwidth and impressive data speed, millimeter-wave (mmWave) radios are expected to play a key role in the 5G and beyond (e.g., 6G) communication networks. Yet, to release mmWave’s true power, the highly directional mmWave beams need to be aligned perfectly. Most existing beam alignment methods adopt an exhaustive or semi-exhaustive space scanning, which introduces up to seconds of delays. To eliminate the need for complex space scanning, this article presents an Ultra-wideband (UWB)-assisted mmWave communication framework, which leverages the co-located UWB antennas to estimate the best angles for mmWave beam alignment. One major challenge of applying this idea in the real world is the barrier of limited antenna numbers. Commercial-Off-The-Shelf (COTS) devices are usually equipped with only a small number of UWB antennas, which are not enough for the existing algorithms to provide an accurate angle estimation. To solve this challenge, we design a novel Multi-Frequency MUltiple SIgnal Classification (MF-MUSIC) algorithm, which extends the classic MUltiple SIgnal Classification (MUSIC) algorithm to the frequency domain and overcomes the antenna limitation barrier in the spatial domain. Extensive real-world experiments and numerical simulations illustrate the advantage of the proposed MF-MUSIC algorithm. MF-MUSIC uses only three antennas to achieve an accurate angle estimation, which is a mere 0.15° (or a relative difference of 3.6%) different from the state-of-the-art 16-antenna-based angle estimation method.
- 20 January 2023
- Published by MIT Press in Neural Computation
- Vol. 35 (2) , 228-248
Dynamic Consolidation for Continual Learning
Training deep learning models from a stream of nonstationary data is a critical problem to be solved to achieve general artificial intelligence. As a promising solution, the continual learning (CL) technique aims to build intelligent systems that have the plasticity to learn from new information without forgetting the previously obtained knowledge. Unfortunately, existing CL methods face two nontrivial limitations. First, when updating a model with new data, existing CL methods usually constrain the model parameters within the vicinity of the parameters optimized for old data, limiting the exploration ability of the model; second, the important strength of each parameter (used to consolidate the previously learned knowledge) is fixed and thus is suboptimal for the dynamic parameter updates. To address these limitations, we first relax the vicinity constraints with a global definition of the important strength, which allows us to explore the full parameter space. Specifically, we define the important strength as the sensitivity of the global loss function to the model parameters. Moreover, we propose adjusting the important strength adaptively to align it with the dynamic parameter updates. Through extensive experiments on popular data sets, we demonstrate that our proposed method outperforms the strong baselines by up to 24% in terms of average accuracy.
- 20 January 2023
- Published by MIT Press in Neural Computation
- Vol. 35 (2) , 228-248
Dynamic Consolidation for Continual Learning
Training deep learning models from a stream of nonstationary data is a critical problem to be solved to achieve general artificial intelligence. As a promising solution, the continual learning (CL) technique aims to build intelligent systems that have the plasticity to learn from new information without forgetting the previously obtained knowledge. Unfortunately, existing CL methods face two nontrivial limitations. First, when updating a model with new data, existing CL methods usually constrain the model parameters within the vicinity of the parameters optimized for old data, limiting the exploration ability of the model; second, the important strength of each parameter (used to consolidate the previously learned knowledge) is fixed and thus is suboptimal for the dynamic parameter updates. To address these limitations, we first relax the vicinity constraints with a global definition of the important strength, which allows us to explore the full parameter space. Specifically, we define the important strength as the sensitivity of the global loss function to the model parameters. Moreover, we propose adjusting the important strength adaptively to align it with the dynamic parameter updates. Through extensive experiments on popular data sets, we demonstrate that our proposed method outperforms the strong baselines by up to 24% in terms of average accuracy.
- Open Access
Hyperspherical Quantization: Toward Smaller and More Accurate Models
Model quantization enables the deployment of deep neural networks under resource-constrained devices. Vector quantization aims at reducing the model size by indexing model weights with full-precision embeddings, i.e., codewords, while the index needs to be restored to 32-bit during computation. Binary and other low-precision quantization methods can reduce the model size up to 32$\times$, however, at the cost of a considerable accuracy drop. In this paper, we propose an efficient framework for ternary quantization to produce smaller and more accurate compressed models. By integrating hyperspherical learning, pruning and reinitialization, our proposed Hyperspherical Quantization (HQ) method reduces the cosine distance between the full-precision and ternary weights, thus reducing the bias of the straight-through gradient estimator during ternary quantization. Compared with existing work at similar compression levels ($\sim$30$\times$, $\sim$40$\times$), our method significantly improves the test accuracy and reduces the model size.
Open AccessHyperspherical Quantization: Toward Smaller and More Accurate Models
Model quantization enables the deployment of deep neural networks under resource-constrained devices. Vector quantization aims at reducing the model size by indexing model weights with full-precision embeddings, i.e., codewords, while the index needs to be restored to 32-bit during computation. Binary and other low-precision quantization methods can reduce the model size up to 32$\times$, however, at the cost of a considerable accuracy drop. In this paper, we propose an efficient framework for ternary quantization to produce smaller and more accurate compressed models. By integrating hyperspherical learning, pruning and reinitialization, our proposed Hyperspherical Quantization (HQ) method reduces the cosine distance between the full-precision and ternary weights, thus reducing the bias of the straight-through gradient estimator during ternary quantization. Compared with existing work at similar compression levels ($\sim$30$\times$, $\sim$40$\times$), our method significantly improves the test accuracy and reduces the model size.
- 15 November 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Network
- Vol. 37 (6) , 150-157
Blockchain-Based Federated Learning: A Systematic Survey
Federated Learning (FL) is a distributed machine learning paradigm that trains models across multiple devices without exchanging users' data, thereby providing stronger data privacy guarantees. However, some research reveals that FL may face security and privacy issues, such as single point of failure, model poisoning, and parameter privacy disclosure. Recently, the field of combining blockchain and FL has trend to become a hot research topic. More and more researchers attempt to use blockchain to decentralize and secure FL frameworks. To further understand recent advances in blockchain-based FL (BFL) systems, this article aims to provide a systematic survey to deconstruct BFL systems. We propose a taxonomy of BFL systems following the lifecycle of FL tasks and divide them into three layers, i.e., the blockchain layer, the training layer, and the aggregation layer. We review and summarize representative work in each layer. We also discuss several open challenges for designing more secure and efficient BFL systems.
- 15 November 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Network
- Vol. 37 (6) , 150-157
Blockchain-Based Federated Learning: A Systematic Survey
Federated Learning (FL) is a distributed machine learning paradigm that trains models across multiple devices without exchanging users' data, thereby providing stronger data privacy guarantees. However, some research reveals that FL may face security and privacy issues, such as single point of failure, model poisoning, and parameter privacy disclosure. Recently, the field of combining blockchain and FL has trend to become a hot research topic. More and more researchers attempt to use blockchain to decentralize and secure FL frameworks. To further understand recent advances in blockchain-based FL (BFL) systems, this article aims to provide a systematic survey to deconstruct BFL systems. We propose a taxonomy of BFL systems following the lifecycle of FL tasks and divide them into three layers, i.e., the blockchain layer, the training layer, and the aggregation layer. We review and summarize representative work in each layer. We also discuss several open challenges for designing more secure and efficient BFL systems.