Kunchi 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
Kunchi Li has a total of 33 co-authors. Here are the top co-authors by number of shared publications:
- Jun Wan
Institute of Automation, Chinese Academy of Sciences, Beijing, China, China
3 shared publications - Shan Yu
Laboratory of Brain Atlas and Brain-inspired Intelligence, Institute of Automation, Chinese Academy of Sciences, Beijing, China
3 shared publications - Da-Han Wang
Xiamen University of Technology
2 shared publications - Hongyang Chen
Zhejiang Lab
2 shared publications - Shunzhi Zhu
Xiamen University of Technology, Xiamen, China
2 shared publications - Song-Liang Pan
Fujian Key Laboratory of Pattern Recognition and Image Understanding, School of Computer and Information Engineering, Xiamen University of Technology, Xiamen, China
2 shared publications
Latest Publications
- 15 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. PP (15209210) , 1-15
Debias Fine-Grained Radical Modeling via Global-Local Attention Mutual Learning for Zero-Shot Chinese Character Recognition
In zero-shot Chinese character recognition (ZSCCR), radicals play a critical bridge for transferring semantic knowledge from seen to unseen characters. However, existing radical-based approaches tend to coarsely map Chinese character images to radical embeddings, disregarding the imperative for fine-grained radical learning and the prevalent imbalanced issue among radicals. This leads to underemphasizing less frequent radicals or introducing bias towards more prevalent ones. To address this, we propose a novel framework of Debiasing fine-grained Radical Modeling via Global-Local radical attention mutual learning for ZSCCR, termed DeRMGL. DeRMGL features two branches for global and local radical modeling. The multi-local branch partitions learnable radical prototypes into non-overlapping blocks by frequency, each with a dedicated attention network. Since radicals in each block have similar occurrence frequencies, attention scores are less skewed by frequency disparities, effectively alleviating radical imbalance. The global branch maintains full radical coverage to coordinate holistic semantic scoring. During training, mutual learning loss optimizes both branches collaboratively, and during inference, their predictions are combined. Furthermore, an enhanced Radical Prototype-based Attention (RPA) network serves as the backbone to leverage visual-semantic knowledge. Extensive experiments on four ZSCCR benchmark datasets demonstrate that DeRMGL achieves state-of-the-art performance.
- 15 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. PP (15209210) , 1-15
Debias Fine-Grained Radical Modeling via Global-Local Attention Mutual Learning for Zero-Shot Chinese Character Recognition
In zero-shot Chinese character recognition (ZSCCR), radicals play a critical bridge for transferring semantic knowledge from seen to unseen characters. However, existing radical-based approaches tend to coarsely map Chinese character images to radical embeddings, disregarding the imperative for fine-grained radical learning and the prevalent imbalanced issue among radicals. This leads to underemphasizing less frequent radicals or introducing bias towards more prevalent ones. To address this, we propose a novel framework of Debiasing fine-grained Radical Modeling via Global-Local radical attention mutual learning for ZSCCR, termed DeRMGL. DeRMGL features two branches for global and local radical modeling. The multi-local branch partitions learnable radical prototypes into non-overlapping blocks by frequency, each with a dedicated attention network. Since radicals in each block have similar occurrence frequencies, attention scores are less skewed by frequency disparities, effectively alleviating radical imbalance. The global branch maintains full radical coverage to coordinate holistic semantic scoring. During training, mutual learning loss optimizes both branches collaboratively, and during inference, their predictions are combined. Furthermore, an enhanced Radical Prototype-based Attention (RPA) network serves as the backbone to leverage visual-semantic knowledge. Extensive experiments on four ZSCCR benchmark datasets demonstrate that DeRMGL achieves state-of-the-art performance.
- 26 August 2025
- Published by Elsevier in Expert Systems with Applications
- Vol. 297
Diverse feature generation for zero-shot Chinese character recognition
- 26 August 2025
- Published by Elsevier in Expert Systems with Applications
- Vol. 297
Diverse feature generation for zero-shot Chinese character recognition
- 2 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 36 (9) , 16171-16185
CKDF-V2: Effectively Alleviating Representation Shift for Continual Learning With Small Memory
In continual learning (CL), the newly arrived data are often out-of-distribution from the previous ones, causing drastic representation shift (RS) when updating the old model on the new data, leading to catastrophic forgetting. In this work, we propose feature boosting calibration (FBC) to tackle this problem. Specifically, an expanded module is trained to learn all the classes, including the old and new classes, discovering critical features missed by the original/old model. Then, an FBC network (FBCN) is trained to exploit these missed features to calibrate the old representations. As the missed features increase the information needed for distinguishing between the old and new classes, FBCN generates the calibrated ones with more transferable features, thus alleviating the RS. Next, given the limited memory to store samples of the old/learned classes, the data are severely imbalanced between the old and new classes. To cope with this problem, we propose blockwise knowledge distillation (BWKD), which splits the softmax layer into blocks according to class frequency and then distills each block separately, resolving data imbalance effectively. Building upon the two improvements, we propose a two-stage training framework for CL, named CKDF-V2, providing an enhanced version of the cascaded knowledge distillation framework (CKDF). Furthermore, we integrate it with a task-token expansion method to develop a novel approach for CL based on the vision transformer (ViT). Extensive experiments show that both a convolutional neural network (CNN) and ViT-based CKDF-V2 obtain favorable results across multiple CL benchmarks.
- 2 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 36 (9) , 16171-16185
CKDF-V2: Effectively Alleviating Representation Shift for Continual Learning With Small Memory
In continual learning (CL), the newly arrived data are often out-of-distribution from the previous ones, causing drastic representation shift (RS) when updating the old model on the new data, leading to catastrophic forgetting. In this work, we propose feature boosting calibration (FBC) to tackle this problem. Specifically, an expanded module is trained to learn all the classes, including the old and new classes, discovering critical features missed by the original/old model. Then, an FBC network (FBCN) is trained to exploit these missed features to calibrate the old representations. As the missed features increase the information needed for distinguishing between the old and new classes, FBCN generates the calibrated ones with more transferable features, thus alleviating the RS. Next, given the limited memory to store samples of the old/learned classes, the data are severely imbalanced between the old and new classes. To cope with this problem, we propose blockwise knowledge distillation (BWKD), which splits the softmax layer into blocks according to class frequency and then distills each block separately, resolving data imbalance effectively. Building upon the two improvements, we propose a two-stage training framework for CL, named CKDF-V2, providing an enhanced version of the cascaded knowledge distillation framework (CKDF). Furthermore, we integrate it with a task-token expansion method to develop a novel approach for CL based on the vision transformer (ViT). Extensive experiments show that both a convolutional neural network (CNN) and ViT-based CKDF-V2 obtain favorable results across multiple CL benchmarks.
- 24 April 2025
- Published by Elsevier in Neurocomputing
- Vol. 639
Enhance the old representations’ adaptability dynamically for exemplar-free continual learning
- 24 April 2025
- Published by Elsevier in Neurocomputing
- Vol. 639
Enhance the old representations’ adaptability dynamically for exemplar-free continual learning
- Open Access
- 20 June 2024
- Published by Wiley in Advanced Science
- Vol. 11 (32) , e2400929
Mapping Brain Synergy Dysfunction in Schizophrenia: Understanding Individual Differences and Underlying Molecular Mechanisms
To elucidate the brain‐wide information interactions that vary and contribute to individual differences in schizophrenia (SCZ), an information‐resolved method is employed to construct individual synergistic and redundant interaction matrices based on regional pairwise BOLD time‐series from 538 SCZ and 540 normal controls (NC). This analysis reveals a stable pattern of regionally‐specific synergy dysfunction in SCZ. Furthermore, a hierarchical Bayesian model is applied to deconstruct the patterns of whole‐brain synergy dysfunction into three latent factors that explain symptom heterogeneity in SCZ. Factor 1 exhibits a significant positive correlation with Positive and Negative Syndrome Scale (PANSS) positive scores, while factor 3 demonstrates significant negative correlations with PANSS negative and general scores. By integrating the neuroimaging data with normative gene expression information, this study identifies that each of these three factors corresponded to a subset of the SCZ risk gene set. Finally, by combining data from NeuroSynth and open molecular imaging sources, along with a spatially heterogeneous mean‐field model, this study delineates three SCZ synergy factors corresponding to distinct symptom profiles and implicating unique cognitive, neurodynamic, and neurobiological mechanisms.
- 20 June 2024
- Published by Wiley in Advanced Science
- Vol. 11 (32) , e2400929
Open AccessMapping Brain Synergy Dysfunction in Schizophrenia: Understanding Individual Differences and Underlying Molecular Mechanisms
To elucidate the brain‐wide information interactions that vary and contribute to individual differences in schizophrenia (SCZ), an information‐resolved method is employed to construct individual synergistic and redundant interaction matrices based on regional pairwise BOLD time‐series from 538 SCZ and 540 normal controls (NC). This analysis reveals a stable pattern of regionally‐specific synergy dysfunction in SCZ. Furthermore, a hierarchical Bayesian model is applied to deconstruct the patterns of whole‐brain synergy dysfunction into three latent factors that explain symptom heterogeneity in SCZ. Factor 1 exhibits a significant positive correlation with Positive and Negative Syndrome Scale (PANSS) positive scores, while factor 3 demonstrates significant negative correlations with PANSS negative and general scores. By integrating the neuroimaging data with normative gene expression information, this study identifies that each of these three factors corresponded to a subset of the SCZ risk gene set. Finally, by combining data from NeuroSynth and open molecular imaging sources, along with a spatially heterogeneous mean‐field model, this study delineates three SCZ synergy factors corresponding to distinct symptom profiles and implicating unique cognitive, neurodynamic, and neurobiological mechanisms.
- 14 March 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. 34 (8) , 7328-7343
ESDB: Expand the Shrinking Decision Boundary via One-to-Many Information Matching for Continual Learning With Small Memory
Rehearsal methods based on knowledge distillation (KD) have been widely used in continual learning (CL). However, given memory constraints, few exemplars contain limited variations of previously learned tasks, impeding the effectiveness of KD in retaining long-term knowledge. The decision boundaries learned by the typical KD strategy overfit the limited exemplars, leading to “shrunk boundaries" of the old classes. To tackle this problem, we propose a novel KD strategy, called One-to-Many Information Matching method (O2MIM), which generates interpolated data by mixing samples between old and new classes, disentangles the supervision information from them and assigns supervision information to them in favor of the old classes. By doing so, the supervision information from a single exemplar can be matched with multiple information from different interpolated images. Moreover, O2MIM utilizes one trainable parameter to create an adaptive KD loss, thereby facilitating a flexible matching process with the designated supervision information. Consequently, O2MIM exploits the exemplar corset more effectively, expanding the shrunk decision boundaries towards the new classes. Next, to incorporate new classes into our classification model, we apply an effective classification training strategy to train a debiased classifier. Combining it with O2MIM, we propose the method of Expanding the Shrinking Decision Boundaries (ESDB), which simultaneously transfers knowledge from the old model via O2MIM and learns new classes by the classification training strategy. Extensive experiments demonstrate that ESDB achieves state-of-the-art performance on diverse CL benchmarks. We also confirm that O2MIM can be used with various label-mixing methods to improve overall performance in CL. The code is available at: https://github.com/CSTiger77/ESDB.
- 14 March 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. 34 (8) , 7328-7343
ESDB: Expand the Shrinking Decision Boundary via One-to-Many Information Matching for Continual Learning With Small Memory
Rehearsal methods based on knowledge distillation (KD) have been widely used in continual learning (CL). However, given memory constraints, few exemplars contain limited variations of previously learned tasks, impeding the effectiveness of KD in retaining long-term knowledge. The decision boundaries learned by the typical KD strategy overfit the limited exemplars, leading to “shrunk boundaries" of the old classes. To tackle this problem, we propose a novel KD strategy, called One-to-Many Information Matching method (O2MIM), which generates interpolated data by mixing samples between old and new classes, disentangles the supervision information from them and assigns supervision information to them in favor of the old classes. By doing so, the supervision information from a single exemplar can be matched with multiple information from different interpolated images. Moreover, O2MIM utilizes one trainable parameter to create an adaptive KD loss, thereby facilitating a flexible matching process with the designated supervision information. Consequently, O2MIM exploits the exemplar corset more effectively, expanding the shrunk decision boundaries towards the new classes. Next, to incorporate new classes into our classification model, we apply an effective classification training strategy to train a debiased classifier. Combining it with O2MIM, we propose the method of Expanding the Shrinking Decision Boundaries (ESDB), which simultaneously transfers knowledge from the old model via O2MIM and learns new classes by the classification training strategy. Extensive experiments demonstrate that ESDB achieves state-of-the-art performance on diverse CL benchmarks. We also confirm that O2MIM can be used with various label-mixing methods to improve overall performance in CL. The code is available at: https://github.com/CSTiger77/ESDB.
- 24 May 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Image Processing
- Vol. 31 (10577149) , 3825-3837
CKDF: Cascaded Knowledge Distillation Framework for Robust Incremental Learning
Recently, owing to the superior performances, knowledge distillation-based (kd-based) methods with the exemplar rehearsal have been widely applied in class incremental learning (CIL). However, we discover that they suffer from the feature uncalibration problem, which is caused by directly transferring knowledge from the old model immediately to the new model when learning a new task. As the old model confuses the feature representations between the learned and new classes, the kd loss and the classification loss used in kd-based methods are heterogeneous. This is detrimental if we learn the existing knowledge from the old model directly in the way as in typical kd-based methods. To tackle this problem, the feature calibration network (FCN) is proposed, which is used to calibrate the existing knowledge to alleviate the feature representation confusion of the old model. In addition, to relieve the task-recency bias of FCN caused by the limited storage memory in CIL, we propose a novel image-feature hybrid sample rehearsal strategy to train FCN by splitting the memory budget to store the image-and-feature exemplars of the previous tasks. As feature embeddings of images have much lower-dimensions, this allows us to store more samples to train FCN. Based on these two improvements, we propose the Cascaded Knowledge Distillation Framework (CKDF) including three main stages. The first stage is used to train FCN to calibrate the existing knowledge of the old model. Then, the new model is trained simultaneously by transferring knowledge from the calibrated teacher model through the knowledge distillation strategy and learning new classes. Finally, after completing the new task learning, the feature exemplars of previous tasks are updated. Importantly, we demonstrate that the proposed CKDF is a general framework that can be applied to various kd-based methods. Experimental results show that our method achieves state-of-the-art performances on several CIL benchmarks.
- 24 May 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Image Processing
- Vol. 31 (10577149) , 3825-3837
CKDF: Cascaded Knowledge Distillation Framework for Robust Incremental Learning
Recently, owing to the superior performances, knowledge distillation-based (kd-based) methods with the exemplar rehearsal have been widely applied in class incremental learning (CIL). However, we discover that they suffer from the feature uncalibration problem, which is caused by directly transferring knowledge from the old model immediately to the new model when learning a new task. As the old model confuses the feature representations between the learned and new classes, the kd loss and the classification loss used in kd-based methods are heterogeneous. This is detrimental if we learn the existing knowledge from the old model directly in the way as in typical kd-based methods. To tackle this problem, the feature calibration network (FCN) is proposed, which is used to calibrate the existing knowledge to alleviate the feature representation confusion of the old model. In addition, to relieve the task-recency bias of FCN caused by the limited storage memory in CIL, we propose a novel image-feature hybrid sample rehearsal strategy to train FCN by splitting the memory budget to store the image-and-feature exemplars of the previous tasks. As feature embeddings of images have much lower-dimensions, this allows us to store more samples to train FCN. Based on these two improvements, we propose the Cascaded Knowledge Distillation Framework (CKDF) including three main stages. The first stage is used to train FCN to calibrate the existing knowledge of the old model. Then, the new model is trained simultaneously by transferring knowledge from the calibrated teacher model through the knowledge distillation strategy and learning new classes. Finally, after completing the new task learning, the feature exemplars of previous tasks are updated. Importantly, we demonstrate that the proposed CKDF is a general framework that can be applied to various kd-based methods. Experimental results show that our method achieves state-of-the-art performances on several CIL benchmarks.