Wei-Long Zheng
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
Wei-Long Zheng has a total of 144 co-authors. Here are the top co-authors by number of shared publications:
- Bao-Liang Lu
Shanghai Jiao Tong University
29 shared publications - Ziyi Li
School of Computer Science, Shanghai Jiao Tong University, Shanghai, China
9 shared publications - Michael Brandon Westover
Department of Neurology, Stanford University, Stanford, CA, United States of America
6 shared publications Adithya Sivaraju
Yale School of Medicine
3 shared publicationsJeannette Hofmeijer
Rijnstate Hospital
3 shared publicationsNicolas Gaspard
Universite Libre de Bruxelles
3 shared publications
Latest Publications
- 8 May 2026
- Published by Springer Nature in Journal of Shanghai Jiaotong University (Science)
- Vol. 31 (3) , 783-799
EEG Foundation Models for Brain-Computer Interfaces: Progress and Future Directions
- 8 May 2026
- Published by Springer Nature in Journal of Shanghai Jiaotong University (Science)
- Vol. 31 (3) , 783-799
EEG Foundation Models for Brain-Computer Interfaces: Progress and Future Directions
- 6 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Affective Computing
- Vol. 17 (2) , 2257-2273
SEED-OLF: A Novel EEG Dataset With Olfactory Stimulation for Emotion Recognition
Olfactory stimuli offer a unique and underexplored modality for affective brain-computer interfaces (aBCIs), leveraging the direct projection of odors to limbic structures involved in emotion processing. In this paper, we introduce a novel emotion induction paradigm based on olfactory stimulation and present a new dataset called SEED-OLF, which is designed to investigate the neural patterns of odor-induced emotional states. SEED-OLF was collected from three emotion-related classification tasks: 1) distinguishing sniffing from non-sniffing EEG responses, 2) decoding subjectively reported valence, and 3) decoding the objective valence of odors. To advance the decoding of odor induced emotional states, we introduce DSEN, a dual-stream EEG network that jointly captures spatiotemporal and spectral characteristics of EEG signals. DSEN integrates a spatiotemporal stream that employs temporal attention, multi-scale dense feature extraction, and attention-based fusion modules, along with a spectral stream that leverages differential entropy features and Transformer-based spectral processing. Experimental results indicate that DSEN significantly outperforms the existing state of-the-art models across all three classification tasks. Topographic EEG analyses further reveal spectral-specific and region-specific cortical activations associated with olfactory-induced emotional processing. The SEED-OLF dataset and codes will be public upon paper acceptance.
- 6 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Affective Computing
- Vol. 17 (2) , 2257-2273
SEED-OLF: A Novel EEG Dataset With Olfactory Stimulation for Emotion Recognition
Olfactory stimuli offer a unique and underexplored modality for affective brain-computer interfaces (aBCIs), leveraging the direct projection of odors to limbic structures involved in emotion processing. In this paper, we introduce a novel emotion induction paradigm based on olfactory stimulation and present a new dataset called SEED-OLF, which is designed to investigate the neural patterns of odor-induced emotional states. SEED-OLF was collected from three emotion-related classification tasks: 1) distinguishing sniffing from non-sniffing EEG responses, 2) decoding subjectively reported valence, and 3) decoding the objective valence of odors. To advance the decoding of odor induced emotional states, we introduce DSEN, a dual-stream EEG network that jointly captures spatiotemporal and spectral characteristics of EEG signals. DSEN integrates a spatiotemporal stream that employs temporal attention, multi-scale dense feature extraction, and attention-based fusion modules, along with a spectral stream that leverages differential entropy features and Transformer-based spectral processing. Experimental results indicate that DSEN significantly outperforms the existing state of-the-art models across all three classification tasks. Topographic EEG analyses further reveal spectral-specific and region-specific cortical activations associated with olfactory-induced emotional processing. The SEED-OLF dataset and codes will be public upon paper acceptance.
- Open Access
MindCine: Multimodal EEG-to-Video Reconstruction with Large-Scale Pretrained Models
Reconstructing human dynamic visual perception from electroencephalography (EEG) signals is of great research significance since EEG's non-invasiveness and high temporal resolution. However, EEG-to-video reconstruction remains challenging due to: 1) Single Modality: existing studies solely align EEG signals with the text modality, which ignores other modalities and are prone to suffer from overfitting problems; 2) Data Scarcity: current methods often have difficulty training to converge with limited EEG-video data. To solve the above problems, we propose a novel framework MindCine to achieve high-fidelity video reconstructions on limited data. We employ a multimodal joint learning strategy to incorporate beyond-text modalities in the training stage and leverage a pre-trained large EEG model to relieve the data scarcity issue for decoding semantic information, while a Seq2Seq model with causal attention is specifically designed for decoding perceptual information. Extensive experiments demonstrate that our model outperforms state-of-the-art methods both qualitatively and quantitatively. Additionally, the results underscore the effectiveness of the complementary strengths of different modalities and demonstrate that leveraging a large-scale EEG model can further enhance reconstruction performance by alleviating the challenges associated with limited data.
Open AccessMindCine: Multimodal EEG-to-Video Reconstruction with Large-Scale Pretrained Models
Reconstructing human dynamic visual perception from electroencephalography (EEG) signals is of great research significance since EEG's non-invasiveness and high temporal resolution. However, EEG-to-video reconstruction remains challenging due to: 1) Single Modality: existing studies solely align EEG signals with the text modality, which ignores other modalities and are prone to suffer from overfitting problems; 2) Data Scarcity: current methods often have difficulty training to converge with limited EEG-video data. To solve the above problems, we propose a novel framework MindCine to achieve high-fidelity video reconstructions on limited data. We employ a multimodal joint learning strategy to incorporate beyond-text modalities in the training stage and leverage a pre-trained large EEG model to relieve the data scarcity issue for decoding semantic information, while a Seq2Seq model with causal attention is specifically designed for decoding perceptual information. Extensive experiments demonstrate that our model outperforms state-of-the-art methods both qualitatively and quantitatively. Additionally, the results underscore the effectiveness of the complementary strengths of different modalities and demonstrate that leveraging a large-scale EEG model can further enhance reconstruction performance by alleviating the challenges associated with limited data.
- 28 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Affective Computing
- Vol. 17 (1) , 841-855
Gram: A Large General EEG Model for Raw Data Classification and Restoration
Drawing insights from Large Language Models, researchers have developed several Large Electroencephalogram (EEG) models (LEMs) to learn a generalized representation adaptable to various tasks. However, such LEMs are scarce and neglecting the potential in data restoration tasks. Meanwhile, how to efficiently integrate temporal view and spectral view of EEG data has always been a focal point. In this paper, we propose Gram, a large general EEG model for raw EEG data classification and restoration tasks. Gram consists of two stages. 1) The initial stage quantizes raw EEG patches into base classes rich in temporal information. 2) The second stage features a multi-view layer-fusion masked autoencoder that exploits EEG's complex Temporal and Spectral views through dual training objectives: a spectral mimic target after layer-fusion encoder for visible patches and a base-class classification target after decoder for masked patches. Pretrained on 7000 hours of EEG data, Gram achieves state-of-the-art performance on four cross-subject classification tasks including motor imagery as well as event, emotion, and sleep stage classification. For EEG data restoration, our model significantly improves classification performance by repairing corrupted data in comparison to using noisy data.
- 28 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Affective Computing
- Vol. 17 (1) , 841-855
Gram: A Large General EEG Model for Raw Data Classification and Restoration
Drawing insights from Large Language Models, researchers have developed several Large Electroencephalogram (EEG) models (LEMs) to learn a generalized representation adaptable to various tasks. However, such LEMs are scarce and neglecting the potential in data restoration tasks. Meanwhile, how to efficiently integrate temporal view and spectral view of EEG data has always been a focal point. In this paper, we propose Gram, a large general EEG model for raw EEG data classification and restoration tasks. Gram consists of two stages. 1) The initial stage quantizes raw EEG patches into base classes rich in temporal information. 2) The second stage features a multi-view layer-fusion masked autoencoder that exploits EEG's complex Temporal and Spectral views through dual training objectives: a spectral mimic target after layer-fusion encoder for visible patches and a base-class classification target after decoder for masked patches. Pretrained on 7000 hours of EEG data, Gram achieves state-of-the-art performance on four cross-subject classification tasks including motor imagery as well as event, emotion, and sleep stage classification. For EEG data restoration, our model significantly improves classification performance by repairing corrupted data in comparison to using noisy data.
- 22 November 2025
- Published by Springer Nature
- p. 64-75
Hierarchical Emotion Transformer for Multimodal Joint Emotion Category and Intensity Recognition
- 22 November 2025
- Published by Springer Nature
- p. 64-75
Hierarchical Emotion Transformer for Multimodal Joint Emotion Category and Intensity Recognition
- 9 November 2025
- Published by Springer Nature
- p. 441-456
Multi-session Meditative EEG Classification with Noise-Robust Self-supervision
- 9 November 2025
- Published by Springer Nature
- p. 441-456
Multi-session Meditative EEG Classification with Noise-Robust Self-supervision
- Open Access
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 5647-5656
Human vs AI: How Digital Human News Anchors Affect Our Cognitive Processes?
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 5647-5656
Open AccessHuman vs AI: How Digital Human News Anchors Affect Our Cognitive Processes?
- Open Access
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 5717-5725
Multimodal Emotion Recognition with Missing Modality via a Unified Multi-task Pre-training Framework
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 5717-5725
Open AccessMultimodal Emotion Recognition with Missing Modality via a Unified Multi-task Pre-training Framework
- 25 October 2025
- Vol. 42 (5) , 977-986
[A method for emotion transition recognition using cross-modal feature fusion and global perception].
Current studies on electroencephalogram (EEG) emotion recognition primarily concentrate on discrete stimulus paradigms under controlled laboratory settings, which cannot adequately represent the dynamic transition characteristics of emotional states during multi-context interactions. To address this issue, this paper proposes a novel method for emotion transition recognition that leverages a cross-modal feature fusion and global perception network (CFGPN). Firstly, an experimental paradigm encompassing six types of emotion transition scenarios was designed, and EEG and eye movement data were simultaneously collected from 20 participants, each annotated with dynamic continuous emotion labels. Subsequently, deep canonical correlation analysis integrated with a cross-modal attention mechanism was employed to fuse features from EEG and eye movement signals, resulting in multimodal feature vectors enriched with highly discriminative emotional information. These vectors are then input into a parallel hybrid architecture that combines convolutional neural networks (CNNs) and Transformers. The CNN is employed to capture local time-series features, whereas the Transformer leverages its robust global perception capabilities to effectively model long-range temporal dependencies, enabling accurate dynamic emotion transition recognition. The results demonstrate that the proposed method achieves the lowest mean square error in both valence and arousal recognition tasks on the dynamic emotion transition dataset and a classic multimodal emotion dataset. It exhibits superior recognition accuracy and stability when compared with five existing unimodal and six multimodal deep learning models. The approach enhances both adaptability and robustness in recognizing emotional state transitions in real-world scenarios, showing promising potential for applications in the field of biomedical engineering.
- 25 October 2025
- Vol. 42 (5) , 977-986
[A method for emotion transition recognition using cross-modal feature fusion and global perception].
Current studies on electroencephalogram (EEG) emotion recognition primarily concentrate on discrete stimulus paradigms under controlled laboratory settings, which cannot adequately represent the dynamic transition characteristics of emotional states during multi-context interactions. To address this issue, this paper proposes a novel method for emotion transition recognition that leverages a cross-modal feature fusion and global perception network (CFGPN). Firstly, an experimental paradigm encompassing six types of emotion transition scenarios was designed, and EEG and eye movement data were simultaneously collected from 20 participants, each annotated with dynamic continuous emotion labels. Subsequently, deep canonical correlation analysis integrated with a cross-modal attention mechanism was employed to fuse features from EEG and eye movement signals, resulting in multimodal feature vectors enriched with highly discriminative emotional information. These vectors are then input into a parallel hybrid architecture that combines convolutional neural networks (CNNs) and Transformers. The CNN is employed to capture local time-series features, whereas the Transformer leverages its robust global perception capabilities to effectively model long-range temporal dependencies, enabling accurate dynamic emotion transition recognition. The results demonstrate that the proposed method achieves the lowest mean square error in both valence and arousal recognition tasks on the dynamic emotion transition dataset and a classic multimodal emotion dataset. It exhibits superior recognition accuracy and stability when compared with five existing unimodal and six multimodal deep learning models. The approach enhances both adaptability and robustness in recognizing emotional state transitions in real-world scenarios, showing promising potential for applications in the field of biomedical engineering.
- 25 August 2025
- Vol. 42 (4) , 668-677
[Multi-source adversarial adaptation with calibration for electroencephalogram-based classification of meditation and resting states].
Meditation aims to guide individuals into a state of deep calm and focused attention, and in recent years, it has shown promising potential in the field of medical treatment. Numerous studies have demonstrated that electroencephalogram (EEG) patterns change during meditation, suggesting the feasibility of using deep learning techniques to monitor meditation states. However, significant inter-subject differences in EEG signals poses challenges to the performance of such monitoring systems. To address this issue, this study proposed a novel model-calibrated multi-source adversarial adaptation network (CMAAN). The model first trained multiple domain-adversarial neural networks in a pairwise manner between various source-domain individuals and the target-domain individual. These networks were then integrated through a calibration process using a small amount of labeled data from the target domain to enhance performance. We evaluated the proposed model on an EEG dataset collected from 18 subjects undergoing methamphetamine rehabilitation. The model achieved a classification accuracy of 73.09%. Additionally, based on the learned model, we analyzed the key EEG frequency bands and brain regions involved in the meditation process. The proposed multi-source domain adaptation framework improves both the performance and robustness of EEG-based meditation monitoring and holds great promise for applications in biomedical informatics and clinical practice.
- 25 August 2025
- Vol. 42 (4) , 668-677
[Multi-source adversarial adaptation with calibration for electroencephalogram-based classification of meditation and resting states].
Meditation aims to guide individuals into a state of deep calm and focused attention, and in recent years, it has shown promising potential in the field of medical treatment. Numerous studies have demonstrated that electroencephalogram (EEG) patterns change during meditation, suggesting the feasibility of using deep learning techniques to monitor meditation states. However, significant inter-subject differences in EEG signals poses challenges to the performance of such monitoring systems. To address this issue, this study proposed a novel model-calibrated multi-source adversarial adaptation network (CMAAN). The model first trained multiple domain-adversarial neural networks in a pairwise manner between various source-domain individuals and the target-domain individual. These networks were then integrated through a calibration process using a small amount of labeled data from the target domain to enhance performance. We evaluated the proposed model on an EEG dataset collected from 18 subjects undergoing methamphetamine rehabilitation. The model achieved a classification accuracy of 73.09%. Additionally, based on the learned model, we analyzed the key EEG frequency bands and brain regions involved in the meditation process. The proposed multi-source domain adaptation framework improves both the performance and robustness of EEG-based meditation monitoring and holds great promise for applications in biomedical informatics and clinical practice.