Dongdong Li
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
Dongdong Li has a total of 80 co-authors. Here are the top co-authors by number of shared publications:
- Zhe Wang
East China University of Science and Technology
31 shared publications - Hai Yang
East China University of Science and Technology
11 shared publications - Wenli Du
East China University of Science and Technology
6 shared publications - Jing Zhang
East China University of Science and Technology
3 shared publications - Ziqiu Chi
Department of Computer Science and Engineering, East China University of Science and Technology, Shanghai, China
3 shared publications - Jia Yao
Department of Computer Science and Engineering, East China University of Science and Technology, Shanghai, China
2 shared publications
Latest Publications
- 22 September 2026
- Published by Elsevier in Information Fusion
- Vol. 139
Structured information decomposition over heterogeneous EEG relations for emotion recognition
- 22 September 2026
- Published by Elsevier in Information Fusion
- Vol. 139
Structured information decomposition over heterogeneous EEG relations for emotion recognition
- 6 August 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 333
Dense multi-scale graph token transformer for local-global graph representation learning
- 6 August 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 333
Dense multi-scale graph token transformer for local-global graph representation learning
- 7 June 2026
- Published by Elsevier in Neural Networks
- Vol. 204, 109221
NEURAL-VOX: NEURal auditory language decoding for voice and text reconstruction
- 7 June 2026
- Published by Elsevier in Neural Networks
- Vol. 204, 109221
NEURAL-VOX: NEURal auditory language decoding for voice and text reconstruction
- 29 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network Science and Engineering
- Vol. 13, 6608-6621
MMFC-PFL: Multimodal Feature Collaborative Personalized Federated Learning
Federated learning (FL) aims to improve model performance while protecting user privacy. The amalgamation of multimodal data, encompassing diverse data types, introduces challenges in FL scenarios. To address these challenges, we propose an innovative multimodal feature collaborative personalized federated learning framework (MMFC-PFL) grounded in the unique characteristics of multimodal features. It includes multimodal collaborative loss and personalized parameter aggregation. The multimodal collaborative loss comprises intra-modal loss, aimed at proficiently representing unimodal features, and inter-modal loss, designed to seamlessly integrate collaborative global multimodal features. The process of personalized parameter aggregation entails the discerning selection of clients during initialization in each global round, with a specific emphasis on those that exhibit maximal separation from the current client for aggregation purposes. The distance between clients is facilitated through Canonical Correlation Analysis on inter-modal features. Additionally, we present an iterative client-wise personalized aggregation strategy that enhances the generalization capabilities of the personalized model. We validate the proposed MMFC-PFL on three datasets: AV-MNIST, MM-IMDB, and MIMIC III. The experimental results demonstrate the effective improvement of model performance and robustness by the proposed multimodal federated learning framework.
- 29 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network Science and Engineering
- Vol. 13, 6608-6621
MMFC-PFL: Multimodal Feature Collaborative Personalized Federated Learning
Federated learning (FL) aims to improve model performance while protecting user privacy. The amalgamation of multimodal data, encompassing diverse data types, introduces challenges in FL scenarios. To address these challenges, we propose an innovative multimodal feature collaborative personalized federated learning framework (MMFC-PFL) grounded in the unique characteristics of multimodal features. It includes multimodal collaborative loss and personalized parameter aggregation. The multimodal collaborative loss comprises intra-modal loss, aimed at proficiently representing unimodal features, and inter-modal loss, designed to seamlessly integrate collaborative global multimodal features. The process of personalized parameter aggregation entails the discerning selection of clients during initialization in each global round, with a specific emphasis on those that exhibit maximal separation from the current client for aggregation purposes. The distance between clients is facilitated through Canonical Correlation Analysis on inter-modal features. Additionally, we present an iterative client-wise personalized aggregation strategy that enhances the generalization capabilities of the personalized model. We validate the proposed MMFC-PFL on three datasets: AV-MNIST, MM-IMDB, and MIMIC III. The experimental results demonstrate the effective improvement of model performance and robustness by the proposed multimodal federated learning framework.
- 21 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Affective Computing
- Vol. 17 (2) , 2383-2396
Fine-Grained Emotion Adaptive Alignment Network for Image Emotion Distribution Transfer
Our research found that images dominated by a particular emotion typically exhibit certain commonalities, and mapping the features of images with similar emotion distribution to adjacent emotion feature space helps capture the emotion distribution of the images. Based on this, we propose a novel Fine-grained Emotion adaptive Alignment Network (FEA-Net) for image emotion distribution transfer, which utilizes multiple codebooks to store the latent distributions of images from different emotion categories for realizing the transfer of emotion distributions between images. To ensure the consistency between the emotion distributions of generated and target images, we propose an adaptive feature alignment module, which maps emotion features to different channels and computes the adaptive soft cross-entropy loss to reduce the gap in emotion features at the channel level. In addition, a feature-enhanced driven attention module is proposed to enhance the quality of generated images, which utilizes feature self-enhancement to enrich the information of discrete features, combines with a residual connection structure to retain volatile content information. Experiments on the widely used public dataset indicate that our proposed FEA-Net not only presents outstanding emotion consistency between source and target image after emotion transfer but also can generate content consistent and emotion changing high-quality images. The code is available at the following address: https://github.com/JingVIPLab/FEA-Net.
- 21 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Affective Computing
- Vol. 17 (2) , 2383-2396
Fine-Grained Emotion Adaptive Alignment Network for Image Emotion Distribution Transfer
Our research found that images dominated by a particular emotion typically exhibit certain commonalities, and mapping the features of images with similar emotion distribution to adjacent emotion feature space helps capture the emotion distribution of the images. Based on this, we propose a novel Fine-grained Emotion adaptive Alignment Network (FEA-Net) for image emotion distribution transfer, which utilizes multiple codebooks to store the latent distributions of images from different emotion categories for realizing the transfer of emotion distributions between images. To ensure the consistency between the emotion distributions of generated and target images, we propose an adaptive feature alignment module, which maps emotion features to different channels and computes the adaptive soft cross-entropy loss to reduce the gap in emotion features at the channel level. In addition, a feature-enhanced driven attention module is proposed to enhance the quality of generated images, which utilizes feature self-enhancement to enrich the information of discrete features, combines with a residual connection structure to retain volatile content information. Experiments on the widely used public dataset indicate that our proposed FEA-Net not only presents outstanding emotion consistency between source and target image after emotion transfer but also can generate content consistent and emotion changing high-quality images. The code is available at the following address: https://github.com/JingVIPLab/FEA-Net.
- 27 December 2025
- Published by Elsevier in Information Fusion
- Vol. 130
DepressInstruct: Instruction tuning of large speech-language models for depression detection
- 27 December 2025
- Published by Elsevier in Information Fusion
- Vol. 130
DepressInstruct: Instruction tuning of large speech-language models for depression detection
- 26 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Audio, Speech and Language Processing
- Vol. 34, 404-414
TCLP-KWS: A Triplet Contrastive Learning Framework for Efficient and Customizable Keyword Spotting
Keyword spotting (KWS) requires efficient recognition with minimal resources, posing challenges in real-world applications. To address this, we propose a customized KWS Framework based on Triplet Contrastive Learning Pretraining (TCLP-KWS). The framework follows a two-stage training pipeline. In the pretraining stage, lightweight speech and query encoders are trained on large-scale public speech datasets to capture robust speech-phoneme relationships. By utilizing phrase-level data instead of full sentences, TCLP-KWS accelerates training and enhances fine-grained recognition of short expressions. In the fine-tuning stage, the pre-trained encoders are integrated with a self-attention-based feature extraction module and classifier, optimizing performance on keyword recognition tasks. Additionally, triplet contrastive learning is employed to align and extract features from both audio and phoneme modalities. TCLP-KWS incorporates intra-modal contrastive learning, which automatically generates hard negative pairs. This strategy improves robustness on challenging keyword distinctions, further enhancing recognition performance. Experimental results demonstrate the effectiveness of the framework in achieving high-performance keyword spotting with minimal resource requirements.
- 26 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Audio, Speech and Language Processing
- Vol. 34, 404-414
TCLP-KWS: A Triplet Contrastive Learning Framework for Efficient and Customizable Keyword Spotting
Keyword spotting (KWS) requires efficient recognition with minimal resources, posing challenges in real-world applications. To address this, we propose a customized KWS Framework based on Triplet Contrastive Learning Pretraining (TCLP-KWS). The framework follows a two-stage training pipeline. In the pretraining stage, lightweight speech and query encoders are trained on large-scale public speech datasets to capture robust speech-phoneme relationships. By utilizing phrase-level data instead of full sentences, TCLP-KWS accelerates training and enhances fine-grained recognition of short expressions. In the fine-tuning stage, the pre-trained encoders are integrated with a self-attention-based feature extraction module and classifier, optimizing performance on keyword recognition tasks. Additionally, triplet contrastive learning is employed to align and extract features from both audio and phoneme modalities. TCLP-KWS incorporates intra-modal contrastive learning, which automatically generates hard negative pairs. This strategy improves robustness on challenging keyword distinctions, further enhancing recognition performance. Experimental results demonstrate the effectiveness of the framework in achieving high-performance keyword spotting with minimal resource requirements.
- 4 November 2025
- Published by Elsevier in Pattern Recognition
- Vol. 172
AT-PMF: Progressive multi-modal fusion with adversarial training for physiological emotion recognition
- 4 November 2025
- Published by Elsevier in Pattern Recognition
- Vol. 172
AT-PMF: Progressive multi-modal fusion with adversarial training for physiological emotion recognition
- 17 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. 13 (2) , 1529-1540
FAD3QN: A Brain-Inspired Deep Reinforcement Learning Model for Speech Depression Detection
In recent years, the high prevalence and severity of depression have highlighted the urgent need for early detection. Depressed patients exhibit noticeable emotional changes in their speech, but existing detection methods face significant challenges in modeling emotion perception mechanisms. In this article, inspired by the knowledge of reinforcement learning neuroscience and the theory of emotion perception in the limbic system, we propose a brain-inspired model frontal-amygdala double dueling deep Q network for depression detection based on speech. The model simulates the frontal lobe and amygdala-centred brain mechanisms for emotion perception through the reinforcement learning framework of double dueling deep Q-networks, and embeds the skip-connected 1-D convolutional neural network and bidirectional long short-term memory network neural networks to simulate the emotion perception process in the limbic system. In addition, we designed an adaptively tuned reward function to address the data imbalance in depression detection, and incorporated an additional step-size penalty factor to limit the number of incorrect decisions made by the agent during the early stages of training. Experimental results across multiple datasets demonstrate the effectiveness and generalizability of our approach. Meanwhile, the relevant ablation experiments conducted in this article validate the key roles of the frontal and limbic systems in the reward learning and emotion perception process, as well as the importance of the adaptive reward function in solving the data imbalance problem.
- 17 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. 13 (2) , 1529-1540
FAD3QN: A Brain-Inspired Deep Reinforcement Learning Model for Speech Depression Detection
In recent years, the high prevalence and severity of depression have highlighted the urgent need for early detection. Depressed patients exhibit noticeable emotional changes in their speech, but existing detection methods face significant challenges in modeling emotion perception mechanisms. In this article, inspired by the knowledge of reinforcement learning neuroscience and the theory of emotion perception in the limbic system, we propose a brain-inspired model frontal-amygdala double dueling deep Q network for depression detection based on speech. The model simulates the frontal lobe and amygdala-centred brain mechanisms for emotion perception through the reinforcement learning framework of double dueling deep Q-networks, and embeds the skip-connected 1-D convolutional neural network and bidirectional long short-term memory network neural networks to simulate the emotion perception process in the limbic system. In addition, we designed an adaptively tuned reward function to address the data imbalance in depression detection, and incorporated an additional step-size penalty factor to limit the number of incorrect decisions made by the agent during the early stages of training. Experimental results across multiple datasets demonstrate the effectiveness and generalizability of our approach. Meanwhile, the relevant ablation experiments conducted in this article validate the key roles of the frontal and limbic systems in the reward learning and emotion perception process, as well as the importance of the adaptive reward function in solving the data imbalance problem.
- 16 October 2025
- Published by Springer Nature in Circuits, Systems, and Signal Processing
- Vol. 45 (4) , 3139-3155
An End-to-End Speaker Recognition Method Based on Multi-network Stacking
- 16 October 2025
- Published by Springer Nature in Circuits, Systems, and Signal Processing
- Vol. 45 (4) , 3139-3155
An End-to-End Speaker Recognition Method Based on Multi-network Stacking