Chao Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Chao Li has a total of 29 co-authors. Here are the top co-authors by number of shared publications:
- Chengwei Zhang
College of Information Science and Technology, Dalian Maritime University, Dalian, China
3 shared publications - Yaxi Liu
College of Computer and Information Engineering, Tianjin Normal University, Tianjin, China
3 shared publications - Feng Wang
College of Computer and Information Engineering, Tianjin Normal University, Tianjin, China
2 shared publications - Furui Zhan
Dalian Maritime University
2 shared publications Haishuai Wang
Department of Computer Science, Zhejiang University College, Hangzhou, China
2 shared publicationsWanting Liu
College of Information Science and Technology, Dalian Maritime University, Dalian, China
2 shared publications
Latest Publications
- 23 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (8) , 15621-15630
Structured Sparse Deep Reinforcement Learning for Accelerating Traffic Signal Decision-Making Under Limited Computational Resources
Current deep reinforcement learning (DRL)-based adaptive traffic signal control (ATSC) methods generally face the challenge of excessive demands for storage and computational resources, which severely restricts their deployment in resource-constrained scenarios. Model pruning techniques provide an essential solution to this challenge, among which fine-grained structured pruning has become a research hotspot due to its ability to compress the model while maintaining inference performance. To address this, this study proposes a two-stage lightweight structured pruning method (SPTS-Light), whose core innovation lies in the deep synergy between dynamic shuffled blocks (DSBs)-based fine-grained structured pruning and two-stage dynamic sparse training (DST). Specifically, DSB reorganizes model parameters into dynamically adjustable fine-grained blocks, providing underlying support for accelerated computation and, more importantly, enabling model parameters to precisely adapt to actual needs. The two-stage training forms a logical closed loop for fully leveraging this capability. In the first stage, a shared agent conducts joint DST across intersections, constructing a generalized and stable global sparse skeleton by learning the common laws of traffic flow at different intersections. On this basis, the second stage performs personalized training for each intersection with a dedicated agent targeting its unique traffic characteristics. DSB’s fine-grained dynamic nature perfectly matches the requirements of this stage: through dynamic shuffling and reorganizing of fine-grained blocks, it achieves precise fine-tuning and efficient optimization of local sparse structures, allowing the global skeleton to quickly adapt to the personalized scenario of individual intersections. The evaluation results in multiple scenarios of real-world and synthetic road networks demonstrate that SPTS-Light can maintain highly competitive performance under resource-constrained conditions.
- 23 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (8) , 15621-15630
Structured Sparse Deep Reinforcement Learning for Accelerating Traffic Signal Decision-Making Under Limited Computational Resources
Current deep reinforcement learning (DRL)-based adaptive traffic signal control (ATSC) methods generally face the challenge of excessive demands for storage and computational resources, which severely restricts their deployment in resource-constrained scenarios. Model pruning techniques provide an essential solution to this challenge, among which fine-grained structured pruning has become a research hotspot due to its ability to compress the model while maintaining inference performance. To address this, this study proposes a two-stage lightweight structured pruning method (SPTS-Light), whose core innovation lies in the deep synergy between dynamic shuffled blocks (DSBs)-based fine-grained structured pruning and two-stage dynamic sparse training (DST). Specifically, DSB reorganizes model parameters into dynamically adjustable fine-grained blocks, providing underlying support for accelerated computation and, more importantly, enabling model parameters to precisely adapt to actual needs. The two-stage training forms a logical closed loop for fully leveraging this capability. In the first stage, a shared agent conducts joint DST across intersections, constructing a generalized and stable global sparse skeleton by learning the common laws of traffic flow at different intersections. On this basis, the second stage performs personalized training for each intersection with a dedicated agent targeting its unique traffic characteristics. DSB’s fine-grained dynamic nature perfectly matches the requirements of this stage: through dynamic shuffling and reorganizing of fine-grained blocks, it achieves precise fine-tuning and efficient optimization of local sparse structures, allowing the global skeleton to quickly adapt to the personalized scenario of individual intersections. The evaluation results in multiple scenarios of real-world and synthetic road networks demonstrate that SPTS-Light can maintain highly competitive performance under resource-constrained conditions.
- Open Access
- 12 December 2025
- Published by Association for Computing Machinery (ACM)
- p. 43-48
DGConformer: Dynamic Graph Convolutional Network Combined with Conformer for EEG Emotion Recognition
- 12 December 2025
- Published by Association for Computing Machinery (ACM)
- p. 43-48
Open AccessDGConformer: Dynamic Graph Convolutional Network Combined with Conformer for EEG Emotion Recognition
- 9 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (5) , 8777-8788
A Two-Stage Framework Based on RL for Truck–Drone Collaborative Delivery Problem
With the explosive growth of e-commerce, efficient last-mile delivery has emerged as a critical challenge. Truck–drone collaborative delivery has garnered significant attention as a promising solution to this problem. In this work, we formulate the truck–drone collaborative delivery problem as a collaborative optimization problem, aiming to minimize the total completion time for delivering packages to customers by leveraging the complementary strengths of the drone’s speed and the truck’s endurance. We propose a two-stage framework to address this challenge. In the first stage, the Lin–Kernighan–Helsgaun (LKH) algorithm is used to generate a high-quality initial traveling salesman problem (TSP) solution, serving as a robust starting point. In the second stage, a sequence allocate policy (SAPPO), based on proximal policy optimization (PPO), refines the TSP solution by optimizing the truck–drone collaborative path using a specially designed action space. Extensive experiments conducted on both random dataset and TSPLIB benchmarks demonstrate that our method significantly outperforms existing algorithms regarding delivery time, while exhibiting improved scalability and less training time.
- 9 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (5) , 8777-8788
A Two-Stage Framework Based on RL for Truck–Drone Collaborative Delivery Problem
With the explosive growth of e-commerce, efficient last-mile delivery has emerged as a critical challenge. Truck–drone collaborative delivery has garnered significant attention as a promising solution to this problem. In this work, we formulate the truck–drone collaborative delivery problem as a collaborative optimization problem, aiming to minimize the total completion time for delivering packages to customers by leveraging the complementary strengths of the drone’s speed and the truck’s endurance. We propose a two-stage framework to address this challenge. In the first stage, the Lin–Kernighan–Helsgaun (LKH) algorithm is used to generate a high-quality initial traveling salesman problem (TSP) solution, serving as a robust starting point. In the second stage, a sequence allocate policy (SAPPO), based on proximal policy optimization (PPO), refines the TSP solution by optimizing the truck–drone collaborative path using a specially designed action space. Extensive experiments conducted on both random dataset and TSPLIB benchmarks demonstrate that our method significantly outperforms existing algorithms regarding delivery time, while exhibiting improved scalability and less training time.
- Open Access
- 10 January 2025
- Published by Association for Computing Machinery (ACM)
- p. 157-162
SECNN-LSTM: A CNN and efficient Attention Mechanism Based LSTM network for Electroencephalogram (EEG)-Based Emotion Recognition
- 10 January 2025
- Published by Association for Computing Machinery (ACM)
- p. 157-162
Open AccessSECNN-LSTM: A CNN and efficient Attention Mechanism Based LSTM network for Electroencephalogram (EEG)-Based Emotion Recognition
- Open Access
- 10 January 2025
- Published by Association for Computing Machinery (ACM)
- p. 535-539
DA-MTCN: Domain Adaptation-based Modern Temporal Convolutional Network for EEG Emotion Recognition
- 10 January 2025
- Published by Association for Computing Machinery (ACM)
- p. 535-539
Open AccessDA-MTCN: Domain Adaptation-based Modern Temporal Convolutional Network for EEG Emotion Recognition
- 2 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 28 (10) , 5755-5767
Attention-Based Temporal Graph Representation Learning for EEG-Based Emotion Recognition
Due to the objectivity of emotional expression in the central nervous system, EEG-based emotion recognition can effectively reflect humans' internal emotional states. In recent years, convolutional neural networks (CNNs) and recurrent neural networks (RNNs) have made significant strides in extracting local features and temporal dependencies from EEG signals. However, CNNs ignore spatial distribution information from EEG electrodes; moreover, RNNs may encounter issues such as exploding/vanishing gradients and high time consumption. To address these limitations, we propose an attention-based temporal graph representation network (ATGRNet) for EEG-based emotion recognition. Firstly, a hierarchical attention mechanism is introduced to integrate feature representations from both frequency bands and channels ordered by priority in EEG signals. Second, a graph convolutional neural network with top-k operation is utilized to capture internal relationships between EEG electrodes under different emotion patterns. Next, a residual-based graph readout mechanism is applied to accumulate the EEG feature node-level representations into graph-level representations. Finally, the obtained graph-level representations are fed into a temporal convolutional network (TCN) to extract the temporal dependencies between EEG frames. We evaluated our proposed ATGRNet on the SEED, DEAP and FACED datasets. The experimental findings show that the proposed ATGRNet surpasses the state-of-the-art graph-based mehtods for EEG-based emotion recognition.
- 2 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 28 (10) , 5755-5767
Attention-Based Temporal Graph Representation Learning for EEG-Based Emotion Recognition
Due to the objectivity of emotional expression in the central nervous system, EEG-based emotion recognition can effectively reflect humans' internal emotional states. In recent years, convolutional neural networks (CNNs) and recurrent neural networks (RNNs) have made significant strides in extracting local features and temporal dependencies from EEG signals. However, CNNs ignore spatial distribution information from EEG electrodes; moreover, RNNs may encounter issues such as exploding/vanishing gradients and high time consumption. To address these limitations, we propose an attention-based temporal graph representation network (ATGRNet) for EEG-based emotion recognition. Firstly, a hierarchical attention mechanism is introduced to integrate feature representations from both frequency bands and channels ordered by priority in EEG signals. Second, a graph convolutional neural network with top-k operation is utilized to capture internal relationships between EEG electrodes under different emotion patterns. Next, a residual-based graph readout mechanism is applied to accumulate the EEG feature node-level representations into graph-level representations. Finally, the obtained graph-level representations are fed into a temporal convolutional network (TCN) to extract the temporal dependencies between EEG frames. We evaluated our proposed ATGRNet on the SEED, DEAP and FACED datasets. The experimental findings show that the proposed ATGRNet surpasses the state-of-the-art graph-based mehtods for EEG-based emotion recognition.
- Open Access
- 19 January 2024
- Published by Association for Computing Machinery (ACM)
- p. 624-630
Multi-View Hierarchical Attention Graph Convolutional Network with Domain Adaptation for EEG Emotion Recognition
EEG signals have emerged as a potent modality for emotion recognition owing to their direct correspondence to brain activity. However, the considerable inter-individual variability of EEG signals and the intricacy of their spatial structural features pose significant challenges. To address these challenges, we introduce the Multi-View Hierarchical Attention Graph Convolutional Network (MAGCN) accompanied by Domain Adaptation, a novel architecture that effectively combines multi-view adjacency matrices with a hierarchical attention mechanism and domain adaptation for robust EEG-based emotion recognition. The MAGCN model leverages both functional and spatial connections within EEG data, facilitating dynamic graph convolution that readily adapts to the variability of emotional states. Our domain discriminator, inspired by adversarial learning, guarantees the extraction of domain-invariant features, subsequently augmenting the model's generalisation across subjects. Experiments on the SEED and SEED-IV datasets attest to the MAGCN model's superior performance over existing methods.
- 19 January 2024
- Published by Association for Computing Machinery (ACM)
- p. 624-630
Open AccessMulti-View Hierarchical Attention Graph Convolutional Network with Domain Adaptation for EEG Emotion Recognition
EEG signals have emerged as a potent modality for emotion recognition owing to their direct correspondence to brain activity. However, the considerable inter-individual variability of EEG signals and the intricacy of their spatial structural features pose significant challenges. To address these challenges, we introduce the Multi-View Hierarchical Attention Graph Convolutional Network (MAGCN) accompanied by Domain Adaptation, a novel architecture that effectively combines multi-view adjacency matrices with a hierarchical attention mechanism and domain adaptation for robust EEG-based emotion recognition. The MAGCN model leverages both functional and spatial connections within EEG data, facilitating dynamic graph convolution that readily adapts to the variability of emotional states. Our domain discriminator, inspired by adversarial learning, guarantees the extraction of domain-invariant features, subsequently augmenting the model's generalisation across subjects. Experiments on the SEED and SEED-IV datasets attest to the MAGCN model's superior performance over existing methods.
- 22 November 2023
- Published by Elsevier in Information Fusion
- Vol. 104
Multi-view domain-adaptive representation learning for EEG-based emotion recognition
- 22 November 2023
- Published by Elsevier in Information Fusion
- Vol. 104
Multi-view domain-adaptive representation learning for EEG-based emotion recognition
- 1 February 2023
- Published by International Society for Horticultural Science (ISHS) in Acta Horticulturae
- No. 1360,p. 211-218
A deep learning-based web application to quantify blueberry internal bruising
ISHS XXXI International Horticultural Congress (IHC2022): III International Symposium on Mechanization, Precision Horticulture, and Robotics: Precision and Digital Horticulture in Field Environments A deep learning-based web application to quantify blueberry internal bruising
- 1 February 2023
- Published by International Society for Horticultural Science (ISHS) in Acta Horticulturae
- No. 1360,p. 211-218
A deep learning-based web application to quantify blueberry internal bruising
ISHS XXXI International Horticultural Congress (IHC2022): III International Symposium on Mechanization, Precision Horticulture, and Robotics: Precision and Digital Horticulture in Field Environments A deep learning-based web application to quantify blueberry internal bruising