Zekun 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
Zekun Li has a total of 18 co-authors. Here are the top co-authors by number of shared publications:
- Yinghuan Shi
Nanjing University, Nanjing, Jiangsu, China
4 shared publications - Lei Qi
Southeast University, Nanjing, Jiangsu, China
3 shared publications - Yang Gao
Nanjing University
3 shared publications - Bo Zhang
Nanjing Normal University, Nanjing, China
2 shared publications - Jian Zhang
Nanjing University, Nanjing, China
2 shared publications - Ming Yang
Nanjing Normal University, Nanjing, China
2 shared publications
Latest Publications
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 185-194
Training Medical Volumetric Super-Resolution Model in ONLY One Epoch
Low-resolution medical volumes can lead to missed diagnoses, making medical volumetric super-resolution (MVSR) essential for reliable clinical decision-making. However, real-world scenarios, such as multi-center studies and large-scale medical emergencies, require rapid model development, whereas traditional MVSR methods rely on time-consuming training procedures, thereby increasing development time and delaying deployment. In this paper, we present ESA-one, a foundation model-assisted MVSR framework that significantly accelerates training. The key idea is to leverage the strong shape representation capability of the Segment Anything Model (SAM) to guide early-stage reconstruction, thereby avoiding inefficient shape learning from low-resolution inputs. In addition, tri-axial frequency features are introduced to enhance texture refinement, and a shape-prior distillation strategy enables the removal of SAM during inference for efficient deployment. Experiments demonstrate that ESA-one achieves high-resolution synthesis in only one training epoch—73 seconds on an NVIDIA A800 or 95 seconds on an NVIDIA RTX 3090—outperforming state-of-the-art methods by 10.06 dB at ×2 upsampling on the T2 modality. Our code is available at: https://github.com/caoluyang0830/ESA-one.git.
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 185-194
Open AccessTraining Medical Volumetric Super-Resolution Model in ONLY One Epoch
Low-resolution medical volumes can lead to missed diagnoses, making medical volumetric super-resolution (MVSR) essential for reliable clinical decision-making. However, real-world scenarios, such as multi-center studies and large-scale medical emergencies, require rapid model development, whereas traditional MVSR methods rely on time-consuming training procedures, thereby increasing development time and delaying deployment. In this paper, we present ESA-one, a foundation model-assisted MVSR framework that significantly accelerates training. The key idea is to leverage the strong shape representation capability of the Segment Anything Model (SAM) to guide early-stage reconstruction, thereby avoiding inefficient shape learning from low-resolution inputs. In addition, tri-axial frequency features are introduced to enhance texture refinement, and a shape-prior distillation strategy enables the removal of SAM during inference for efficient deployment. Experiments demonstrate that ESA-one achieves high-resolution synthesis in only one training epoch—73 seconds on an NVIDIA A800 or 95 seconds on an NVIDIA RTX 3090—outperforming state-of-the-art methods by 10.06 dB at ×2 upsampling on the T2 modality. Our code is available at: https://github.com/caoluyang0830/ESA-one.git.
- Open Access
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 5742-5753
HiLLM-CD: LLM-Enhanced Hierarchical Cognitive Diagnosis
Cognitive diagnosis (CD) aims to infer students' latent proficiencies from their exercise responses. While most existing approaches are optimized primarily for response prediction, such prediction-oriented training can yield less reliable proficiency. Moreover, representing proficiency as independent concept-wise variables overlooks the relations among concepts, often resulting in fragmented and incoherent diagnoses. To address these issues, we represent each student as an explicit concept tree with node-wise proficiencies, enabling coarse-to-fine diagnosis with explicit concept relations. In this work, we propose LLM-Enhanced Hierarchical Cognitive Diagnosis (HiLLM-CD), a tree-structured encoder-decoder framework that uses the concept tree as an interpretable proficiency representation. Specifically, HiLLM-CD first encodes a student's response history into node-wise proficiencies on the concept tree, and then decodes the tree-structured proficiencies to predict responses via a GCN-based cognitive decoder. Moreover, HiLLM-CD employs a multi-agent LLM pipeline to generate exercise-concept links and automatically construct the concept tree from educational text, substantially reducing reliance on expert annotations and making the overall diagnosis pipeline automated; it further uses an LLM teacher to guide the model to learn more accurate proficiencies. We evaluate HiLLM-CD on MOOCCubeX across four subject domains (Computer Science, Mathematics, Physics, and Biology). In fully unannotated settings, HiLLM-CD improves response prediction by 11.1% AUC on average and improves proficiency reliability by 0.597 in Kendall's tau. We further validate its utility on computerized adaptive testing, where HiLLM-CD improves AUC by 11.4% on average. Our code is available at: https://github.com/xyq20011011/HiLLM.
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 5742-5753
Open AccessHiLLM-CD: LLM-Enhanced Hierarchical Cognitive Diagnosis
Cognitive diagnosis (CD) aims to infer students' latent proficiencies from their exercise responses. While most existing approaches are optimized primarily for response prediction, such prediction-oriented training can yield less reliable proficiency. Moreover, representing proficiency as independent concept-wise variables overlooks the relations among concepts, often resulting in fragmented and incoherent diagnoses. To address these issues, we represent each student as an explicit concept tree with node-wise proficiencies, enabling coarse-to-fine diagnosis with explicit concept relations. In this work, we propose LLM-Enhanced Hierarchical Cognitive Diagnosis (HiLLM-CD), a tree-structured encoder-decoder framework that uses the concept tree as an interpretable proficiency representation. Specifically, HiLLM-CD first encodes a student's response history into node-wise proficiencies on the concept tree, and then decodes the tree-structured proficiencies to predict responses via a GCN-based cognitive decoder. Moreover, HiLLM-CD employs a multi-agent LLM pipeline to generate exercise-concept links and automatically construct the concept tree from educational text, substantially reducing reliance on expert annotations and making the overall diagnosis pipeline automated; it further uses an LLM teacher to guide the model to learn more accurate proficiencies. We evaluate HiLLM-CD on MOOCCubeX across four subject domains (Computer Science, Mathematics, Physics, and Biology). In fully unannotated settings, HiLLM-CD improves response prediction by 11.1% AUC on average and improves proficiency reliability by 0.597 in Kendall's tau. We further validate its utility on computerized adaptive testing, where HiLLM-CD improves AUC by 11.4% on average. Our code is available at: https://github.com/xyq20011011/HiLLM.
- Open Access
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 6164-6175
SocraticChem: Physics-Grounded Socratic Inquiry for Safety-Critical Experimental Science
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 6164-6175
Open AccessSocraticChem: Physics-Grounded Socratic Inquiry for Safety-Critical Experimental Science
- 1 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Medical Imaging
- Vol. 44 (7) , 3025-3037
Balancing Multi-Target Semi-Supervised Medical Image Segmentation With Collaborative Generalist and Specialists
Despite the promising performance achieved by current semi-supervised models in segmenting individual medical targets, many of these models suffer a notable decrease in performance when tasked with the simultaneous segmentation of multiple targets. A vital factor could be attributed to the imbalanced scales among different targets: during simultaneously segmenting multiple targets, large targets dominate the loss, leading to small targets being misclassified as larger ones. To this end, we propose a novel method, which consists of a Collaborative Generalist and several Specialists, termed CGS. It is centered around the idea of employing a specialist for each target class, thus avoiding the dominance of larger targets. The generalist performs conventional multi-target segmentation, while each specialist is dedicated to distinguishing a specific target class from the remaining target classes and the background. Based on a theoretical insight, we demonstrate that CGS can achieve a more balanced training. Moreover, we develop cross-consistency losses to foster collaborative learning between the generalist and the specialists. Lastly, regarding their intrinsic relation that the target class of any specialized head should belong to the remaining classes of the other heads, we introduce an inter-head error detection module to further enhance the quality of pseudo-labels. Experimental results on three popular benchmarks showcase its superior performance compared to state-of-the-art methods. Our code is available at https://github.com/wangyou0804/CGSpinkhttps://github.com/wangyou0804/CGS.
- 1 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Medical Imaging
- Vol. 44 (7) , 3025-3037
Balancing Multi-Target Semi-Supervised Medical Image Segmentation With Collaborative Generalist and Specialists
Despite the promising performance achieved by current semi-supervised models in segmenting individual medical targets, many of these models suffer a notable decrease in performance when tasked with the simultaneous segmentation of multiple targets. A vital factor could be attributed to the imbalanced scales among different targets: during simultaneously segmenting multiple targets, large targets dominate the loss, leading to small targets being misclassified as larger ones. To this end, we propose a novel method, which consists of a Collaborative Generalist and several Specialists, termed CGS. It is centered around the idea of employing a specialist for each target class, thus avoiding the dominance of larger targets. The generalist performs conventional multi-target segmentation, while each specialist is dedicated to distinguishing a specific target class from the remaining target classes and the background. Based on a theoretical insight, we demonstrate that CGS can achieve a more balanced training. Moreover, we develop cross-consistency losses to foster collaborative learning between the generalist and the specialists. Lastly, regarding their intrinsic relation that the target class of any specialized head should belong to the remaining classes of the other heads, we introduce an inter-head error detection module to further enhance the quality of pseudo-labels. Experimental results on three popular benchmarks showcase its superior performance compared to state-of-the-art methods. Our code is available at https://github.com/wangyou0804/CGSpinkhttps://github.com/wangyou0804/CGS.
- 22 February 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Knowledge and Data Engineering
- Vol. 36 (8) , 4163-4177
Open-Domain Semi-Supervised Learning via Glocal Cluster Structure Exploitation
Semi-supervised learning (SSL) aims to reduce the heavy reliance of current deep models on costly manual annotation by leveraging a large amount of unlabeled data in combination with a much smaller set of labeled data. However, most existing SSL methods assume that all labeled and unlabeled data are drawn from the same feature distribution, which can be impractical in real-world applications. In this study, we take the initial step to systematically investigate the open-domain semi-supervised learning setting, where a feature distribution mismatch exists between labeled and unlabeled data. In pursuit of an effective solution for open-domain SSL, we propose a novel framework called GlocalMatch , which aims to exploit both glo bal and lo cal ( i.e. , glocal) cluster structure of open-domain unlabeled data. The glocal cluster structure is utilized in two complementary ways. Firstly, GlocalMatch optimizes a Glocal Cluster Compacting (GCC) objective, that encourages feature representations of the same class, whether with in the same domain or across different domains, to become closer to each other. Secondly, GlocalMatch incorporates a Glocal Semantic Aggregation (GSA) strategy to produce more reliable pseudo-labels by aggregating predictions from neighboring clusters. Extensive experiments demonstrate that GlocalMatch outperforms the state-of-the-art SSL methods significantly, achieving superior performance for both in-domain and out-of-domain generalization. The code is released in https://github.com/nukezil/GlocalMatch .
- 22 February 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Knowledge and Data Engineering
- Vol. 36 (8) , 4163-4177
Open-Domain Semi-Supervised Learning via Glocal Cluster Structure Exploitation
Semi-supervised learning (SSL) aims to reduce the heavy reliance of current deep models on costly manual annotation by leveraging a large amount of unlabeled data in combination with a much smaller set of labeled data. However, most existing SSL methods assume that all labeled and unlabeled data are drawn from the same feature distribution, which can be impractical in real-world applications. In this study, we take the initial step to systematically investigate the open-domain semi-supervised learning setting, where a feature distribution mismatch exists between labeled and unlabeled data. In pursuit of an effective solution for open-domain SSL, we propose a novel framework called GlocalMatch , which aims to exploit both glo bal and lo cal ( i.e. , glocal) cluster structure of open-domain unlabeled data. The glocal cluster structure is utilized in two complementary ways. Firstly, GlocalMatch optimizes a Glocal Cluster Compacting (GCC) objective, that encourages feature representations of the same class, whether with in the same domain or across different domains, to become closer to each other. Secondly, GlocalMatch incorporates a Glocal Semantic Aggregation (GSA) strategy to produce more reliable pseudo-labels by aggregating predictions from neighboring clusters. Extensive experiments demonstrate that GlocalMatch outperforms the state-of-the-art SSL methods significantly, achieving superior performance for both in-domain and out-of-domain generalization. The code is released in https://github.com/nukezil/GlocalMatch .
- Open Access
- 1 July 2022
- Published by International Joint Conferences on Artificial Intelligence
- Vol. 19 (1) , 373-383
Open Relation Extraction with Non-existent and Multi-span Relationships
Open relation extraction (ORE) aims to assign semantic relationships among arguments, essential to the automatic construction of knowledge graphs (KG). The previous ORE methods and some benchmark datasets consider a relation between two arguments as definitely existing and in a simple single-span form, neglecting possible non-existent relationships and flexible, expressive multi-span relations. However, detecting non-existent relations is necessary for a pipelined information extraction system (first performing named entity recognition then relation extraction), and multi-span relationships contribute to the diversity of connections in KGs. To fulfill the practical demands of ORE, we design a novel Query-based Multi-head Open Relation Extractor (QuORE) to extract single/multi-span relations and detect non-existent relationships effectively. Moreover, we re-construct some public datasets covering English and Chinese to derive augmented and multi-span relation tuples. Extensive experiment results show that our method outperforms the state-of-the-art ORE model LOREM in the extraction of existing single/multi-span relations and the overall performances on four datasets with non-existent relationships.
- 1 July 2022
- Published by International Joint Conferences on Artificial Intelligence
- Vol. 19 (1) , 373-383
Open AccessOpen Relation Extraction with Non-existent and Multi-span Relationships
Open relation extraction (ORE) aims to assign semantic relationships among arguments, essential to the automatic construction of knowledge graphs (KG). The previous ORE methods and some benchmark datasets consider a relation between two arguments as definitely existing and in a simple single-span form, neglecting possible non-existent relationships and flexible, expressive multi-span relations. However, detecting non-existent relations is necessary for a pipelined information extraction system (first performing named entity recognition then relation extraction), and multi-span relationships contribute to the diversity of connections in KGs. To fulfill the practical demands of ORE, we design a novel Query-based Multi-head Open Relation Extractor (QuORE) to extract single/multi-span relations and detect non-existent relationships effectively. Moreover, we re-construct some public datasets covering English and Chinese to derive augmented and multi-span relation tuples. Extensive experiment results show that our method outperforms the state-of-the-art ORE model LOREM in the extraction of existing single/multi-span relations and the overall performances on four datasets with non-existent relationships.