Jingchao Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Jingchao Wang has a total of 41 co-authors. Here are the top co-authors by number of shared publications:
- Guoheng Huang
Guangdong University of Technology
6 shared publications - Tongxu Lin
Beijing University of Posts and Telecommunications, Beijing, China
5 shared publications - Yang Shi
Guangdong University of Technology, Guangzhou, China
5 shared publications - Chi-Man Pun
University of Macau
4 shared publications - Liangsi Lu
Guangdong University of Technology, Guangzhou, China
4 shared publications - Xiaochen Yuan
Macao Polytechnic University
4 shared publications
Latest Publications
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 8887-8895
Diffusion Image Editing via Asynchronous Token Decoding
Text-guided diffusion image editing aims to modify semantic attributes of an image while preserving its identity, layout, and background. However, na\''ively switching the text condition during sampling often causes global drift, as denoising dynamics propagate changes across tokens and can disrupt unedited regions. To address this issue, we propose Asynchronous Token Decoding Edit (ATDEdit), an inference-time framework that views each sampler step as a parallel update of a globally coupled token matrix and enables token-indexed condition switching with differentiated update policies. Instead of applying synchronous target-conditioned updates to all tokens, ATDEdit estimates editable locations using token-wise conditional surprisal and applies target-conditioned corrections to the selected token set. It supplies source key/value memory at keep-token positions and projects selected keep-token latent rows back to their source values; these operations promote background preservation but do not constitute a pixel-level invariance guarantee. This approach combines local editing and background preservation without external or user-provided spatial masks and without model fine-tuning. On PIE-Bench, ATDEdit achieves the strongest reported preservation metrics, including 27.44~dB PSNR and 0.055 LPIPS, while retaining competitive semantic alignment.
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 8887-8895
Open AccessDiffusion Image Editing via Asynchronous Token Decoding
Text-guided diffusion image editing aims to modify semantic attributes of an image while preserving its identity, layout, and background. However, na\''ively switching the text condition during sampling often causes global drift, as denoising dynamics propagate changes across tokens and can disrupt unedited regions. To address this issue, we propose Asynchronous Token Decoding Edit (ATDEdit), an inference-time framework that views each sampler step as a parallel update of a globally coupled token matrix and enables token-indexed condition switching with differentiated update policies. Instead of applying synchronous target-conditioned updates to all tokens, ATDEdit estimates editable locations using token-wise conditional surprisal and applies target-conditioned corrections to the selected token set. It supplies source key/value memory at keep-token positions and projects selected keep-token latent rows back to their source values; these operations promote background preservation but do not constitute a pixel-level invariance guarantee. This approach combines local editing and background preservation without external or user-provided spatial masks and without model fine-tuning. On PIE-Bench, ATDEdit achieves the strongest reported preservation metrics, including 27.44~dB PSNR and 0.055 LPIPS, while retaining competitive semantic alignment.
- Open Access
- 5 September 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Knowledge Discovery From Data
- Vol. 20 (8) , 1-30
Enhancing Robustness of Constant Curvature Graph Convolutional Network with Lipschitz Regularization
Non-Euclidean spaces inherently enable high-fidelity embeddings for hierarchical and cyclical data due to their geometric properties. Existing approaches unify hyperbolic and spherical embeddings within the framework of constant curvature spaces. However, current methods for Lipschitz regularization remain limited to non-positive curvature geometries, such as hyperbolic and Euclidean spaces, and cannot be naturally extended to the general constant curvature setting. In this article, we present a rigorous Lipschitz analysis for constant curvature graph convolutional networks ( \(\kappa\) -GCNs) and enhance their robustness through Lipschitz regularization. We derive upper bounds for the Lipschitz constants across constant curvature spaces, thereby standardizing the Lipschitz limits of the \(\kappa\) -stereographic model. Furthermore, we incorporate these bounds into a regularization framework for \(\kappa\) -GCNs to improve stability and robustness. Experimental results demonstrate that the proposed regularization method often strengthens the robustness of \(\kappa\) -GCNs across various curvature regimes, particularly under Gaussian feature noise.- 5 September 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Knowledge Discovery From Data
- Vol. 20 (8) , 1-30
Open AccessEnhancing Robustness of Constant Curvature Graph Convolutional Network with Lipschitz Regularization
Non-Euclidean spaces inherently enable high-fidelity embeddings for hierarchical and cyclical data due to their geometric properties. Existing approaches unify hyperbolic and spherical embeddings within the framework of constant curvature spaces. However, current methods for Lipschitz regularization remain limited to non-positive curvature geometries, such as hyperbolic and Euclidean spaces, and cannot be naturally extended to the general constant curvature setting. In this article, we present a rigorous Lipschitz analysis for constant curvature graph convolutional networks ( \(\kappa\) -GCNs) and enhance their robustness through Lipschitz regularization. We derive upper bounds for the Lipschitz constants across constant curvature spaces, thereby standardizing the Lipschitz limits of the \(\kappa\) -stereographic model. Furthermore, we incorporate these bounds into a regularization framework for \(\kappa\) -GCNs to improve stability and robustness. Experimental results demonstrate that the proposed regularization method often strengthens the robustness of \(\kappa\) -GCNs across various curvature regimes, particularly under Gaussian feature noise. - Open Access
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 4264-4275
Over-squashing as Transport Congestion: A Sandpile Dynamics Perspective
Message-passing graph neural networks (MP-GNNs) are widely used for learning on relational data. However, their performance drops on tasks requiring long-range interactions due to over-squashing, where exponential information compression overwhelms fixed-width embeddings. While existing analyses often attribute this to geometric bottlenecks under linear diffusion assumptions, thresholded nonlinearities in GNNs motivate a load-release view akin to Abelian sandpiles. Using the discrete sandpile model as a structural proxy, we show that graph bottlenecks force large stabilization cost, effectively creating zones of high transport congestion. We characterize stabilization-invariant equivalence classes induced by the reduced Laplacian and derive cut-based lower bounds linking bottlenecks to unavoidable stabilization effort. The resulting theory is discrete, whereas our implementation is a continuous vector-valued surrogate. The theory identifies the relevant design factors, namely capacity and cut size. Guided by these insights, we propose a differentiable Sandpile Stabilization Layer (SSL) and congestion-aware objectives designed to redistribute excess load and manage stabilization costs. Experiments on long-range benchmarks, together with congestion and collision diagnostics, show that targeting sandpile-identified bottlenecks mitigates representation collapse and improves over standard baselines. Project Page: https://sandpile-gnn.github.io/
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 4264-4275
Open AccessOver-squashing as Transport Congestion: A Sandpile Dynamics Perspective
Message-passing graph neural networks (MP-GNNs) are widely used for learning on relational data. However, their performance drops on tasks requiring long-range interactions due to over-squashing, where exponential information compression overwhelms fixed-width embeddings. While existing analyses often attribute this to geometric bottlenecks under linear diffusion assumptions, thresholded nonlinearities in GNNs motivate a load-release view akin to Abelian sandpiles. Using the discrete sandpile model as a structural proxy, we show that graph bottlenecks force large stabilization cost, effectively creating zones of high transport congestion. We characterize stabilization-invariant equivalence classes induced by the reduced Laplacian and derive cut-based lower bounds linking bottlenecks to unavoidable stabilization effort. The resulting theory is discrete, whereas our implementation is a continuous vector-valued surrogate. The theory identifies the relevant design factors, namely capacity and cut size. Guided by these insights, we propose a differentiable Sandpile Stabilization Layer (SSL) and congestion-aware objectives designed to redistribute excess load and manage stabilization costs. Experiments on long-range benchmarks, together with congestion and collision diagnostics, show that targeting sandpile-identified bottlenecks mitigates representation collapse and improves over standard baselines. Project Page: https://sandpile-gnn.github.io/
- Open Access
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 463-474
Riemannian Liquid Spatio-Temporal Graph Network
Liquid Time-Constant networks (LTCs), a type of continuous-time graph neural network, excel at modeling irregularly-sampled dynamics but are fundamentally confined to Euclidean space. This limitation introduces significant geometric distortion when representing real-world graphs with inherent non-Euclidean structures (e.g., hierarchies and cycles), degrading representation quality. To overcome this limitation, we introduce the Riemannian Liquid Spatio-Temporal Graph Network (RLSTG), a framework that unifies continuous-time liquid dynamics with the geometric inductive biases of Riemannian manifolds. RLSTG models graph evolution through an Ordinary Differential Equation (ODE) formulated directly on a curved manifold, enabling it to faithfully capture the intrinsic geometry of both structurally static and dynamic spatio-temporal graphs. Moreover, we provide rigorous theoretical guarantees for RLSTG, extending stability theorems of LTCs to the Riemannian domain and quantifying its expressive power via state trajectory analysis. Extensive experiments on real-world benchmarks demonstrate that, by combining advanced temporal dynamics with a Riemannian spatial representation, RLSTG achieves superior performance on graphs with complex structures. Project Page: https://rlstg.github.io
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 463-474
Open AccessRiemannian Liquid Spatio-Temporal Graph Network
Liquid Time-Constant networks (LTCs), a type of continuous-time graph neural network, excel at modeling irregularly-sampled dynamics but are fundamentally confined to Euclidean space. This limitation introduces significant geometric distortion when representing real-world graphs with inherent non-Euclidean structures (e.g., hierarchies and cycles), degrading representation quality. To overcome this limitation, we introduce the Riemannian Liquid Spatio-Temporal Graph Network (RLSTG), a framework that unifies continuous-time liquid dynamics with the geometric inductive biases of Riemannian manifolds. RLSTG models graph evolution through an Ordinary Differential Equation (ODE) formulated directly on a curved manifold, enabling it to faithfully capture the intrinsic geometry of both structurally static and dynamic spatio-temporal graphs. Moreover, we provide rigorous theoretical guarantees for RLSTG, extending stability theorems of LTCs to the Riemannian domain and quantifying its expressive power via state trajectory analysis. Extensive experiments on real-world benchmarks demonstrate that, by combining advanced temporal dynamics with a Riemannian spatial representation, RLSTG achieves superior performance on graphs with complex structures. Project Page: https://rlstg.github.io
- Open Access
- 21 February 2026
- Published by Association for Computing Machinery (ACM)
- p. 541-551
PR-CapsNet: Pseudo-Riemannian Capsule Network with Adaptive Curvature Routing for Graph Learning
- 21 February 2026
- Published by Association for Computing Machinery (ACM)
- p. 541-551
Open AccessPR-CapsNet: Pseudo-Riemannian Capsule Network with Adaptive Curvature Routing for Graph Learning
- 16 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 30 (5) , 4175-4185
CAM-Interacted Vision GNN for Multi-Label Medical Images
Vision Graph Neural Network (ViG) is designed to recognize different objects through graph-level processing. However, ViG constructs graphs with appearance-level neighbors and neglects the category semantic. The oversight results in the unintentional connection of patches that belong to different objects, thus affecting the distinctiveness of categories in multi-label medical image learning. Since the pixel-level annotations for images are not easily available, category-aware graphs can not be directly built. To solve this problem, we consider localizing category-specific regions using Class Activation Maps (CAMs), an effective way to highlight regions belonging to each category without requiring manual annotations. Specifically, we propose a CAM-interacted Vision GNN (CiV-GNN), in which category-aware graphs are formed to perform intra-category graph processing. CIV-GNN includes a Class-activated Patch Division (CAPD) module, which introduces CAMs as guidance for category-aware graph building. Furthermore, we develop a Multi-graph Interactive Processing (MIP) module to model the relations between category-aware graphs, promoting inter-category interaction learning. Experimental results show that CiV-GNN performs well in surgical tool localization and multi-label medical image classification. Specifically, for m2cai16-localization, CiV-GNN exhibits a 1.43% and 7.02% improvement in mAP50 and mAP50-95, respectively, compared to YOLOv8.
- 16 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 30 (5) , 4175-4185
CAM-Interacted Vision GNN for Multi-Label Medical Images
Vision Graph Neural Network (ViG) is designed to recognize different objects through graph-level processing. However, ViG constructs graphs with appearance-level neighbors and neglects the category semantic. The oversight results in the unintentional connection of patches that belong to different objects, thus affecting the distinctiveness of categories in multi-label medical image learning. Since the pixel-level annotations for images are not easily available, category-aware graphs can not be directly built. To solve this problem, we consider localizing category-specific regions using Class Activation Maps (CAMs), an effective way to highlight regions belonging to each category without requiring manual annotations. Specifically, we propose a CAM-interacted Vision GNN (CiV-GNN), in which category-aware graphs are formed to perform intra-category graph processing. CIV-GNN includes a Class-activated Patch Division (CAPD) module, which introduces CAMs as guidance for category-aware graph building. Furthermore, we develop a Multi-graph Interactive Processing (MIP) module to model the relations between category-aware graphs, promoting inter-category interaction learning. Experimental results show that CiV-GNN performs well in surgical tool localization and multi-label medical image classification. Specifically, for m2cai16-localization, CiV-GNN exhibits a 1.43% and 7.02% improvement in mAP50 and mAP50-95, respectively, compared to YOLOv8.
- Open Access
- 20 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 47-58
Code Retrieval with Mixture of Experts Prototype Learning Based on Classification
- 20 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 47-58
Open AccessCode Retrieval with Mixture of Experts Prototype Learning Based on Classification
- Open Access
- 28 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 8527-8535
Beyond Direct Relationships: Exploring Multi-Order Label Pair Dependencies for Knowledge Distillation
Multi-label image classification is crucial for a wide range of multimedia applications. To address the resource limitation issue, various knowledge distillation (KD) methods have been developed to transfer knowledge from a large network (referred to as the "teacher") to a small network (referred to as the "student"). However, existing KD methods do not explicitly distill the dependencies between labels, which limits the model ability to capture multi-label correlation. Furthermore, although existing methods for multi-label image classification have utilized the second-order label pair dependency (direct dependency between two labels), the high-order label pair dependency, which captures the indirect dependency between two labels, remains unexplored. In this paper, we propose a Multi-Order Label Pair Dependencies Knowledge Distillation (MDKD) framework. MDKD explicitly distills the knowledge to capture multi-order dependencies between labels, including the label pair dependencies from second-order and high-order, thus transferring the insight of label correlations from different perspectives. Extensive experiments on Pascal VOC2007, MSCOCO2014, and NUS-WIDE demonstrate the superior performances of MDKD.
- 28 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 8527-8535
Open AccessBeyond Direct Relationships: Exploring Multi-Order Label Pair Dependencies for Knowledge Distillation
Multi-label image classification is crucial for a wide range of multimedia applications. To address the resource limitation issue, various knowledge distillation (KD) methods have been developed to transfer knowledge from a large network (referred to as the "teacher") to a small network (referred to as the "student"). However, existing KD methods do not explicitly distill the dependencies between labels, which limits the model ability to capture multi-label correlation. Furthermore, although existing methods for multi-label image classification have utilized the second-order label pair dependency (direct dependency between two labels), the high-order label pair dependency, which captures the indirect dependency between two labels, remains unexplored. In this paper, we propose a Multi-Order Label Pair Dependencies Knowledge Distillation (MDKD) framework. MDKD explicitly distills the knowledge to capture multi-order dependencies between labels, including the label pair dependencies from second-order and high-order, thus transferring the insight of label correlations from different perspectives. Extensive experiments on Pascal VOC2007, MSCOCO2014, and NUS-WIDE demonstrate the superior performances of MDKD.
- 5 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Emerging Topics in Computational Intelligence
- Vol. 9 (2) , 1673-1687
A Novel Hypercomplex Graph Convolution Refining Mechanism
Hypercomplex graph convolutions with higher hypercomplex dimensions can extract more complex features in graphs and features with varying levels of complexity are suited for different situation. However, existing hypercomplex graph neural networks have a constraint that they can only carry out hypercomplex graph convolutions in a predetermined and unchangeable dimension. To address this limitation, this paper presents a solution to overcome this limitation by introducing the FFT-based Adaptive Fourier hypercomplex graph convolution filtering mechanism (FAF mechanism), which can adaptively select hypercomplex graph convolutions with the most appropriate dimensions for different situations by projecting the outputs from all candidate hypercomplex graph convolutions to the frequency domain and selecting the one with the highest energy via the FFT-based Adaptive Fourier Decomposition. Meanwhile, we apply the FAF mechanism to our proposed hypercomplex high-order interaction graph neural network (HHG-Net), which performs high-order interaction and strengthens interaction features through quantum graph hierarchical attention module and feature interaction gated graph convolution. During convolution filtering, the FAF mechanism projects the outputs from different candidate hypercomplex graph convolutions to the frequency domain, extracts their energy, and selects the convolution that outputs the largest energy. After that, the model with selected hypercomplex graph convolutions is trained again. Our method outperforms many benchmarks, including the model with hypercomplex graph convolutions selected by DARTS, in node classification, graph classification, and text classification. This showcases the versatility of our approach, which can be effectively applied to both graph and text data.
- 5 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Emerging Topics in Computational Intelligence
- Vol. 9 (2) , 1673-1687
A Novel Hypercomplex Graph Convolution Refining Mechanism
Hypercomplex graph convolutions with higher hypercomplex dimensions can extract more complex features in graphs and features with varying levels of complexity are suited for different situation. However, existing hypercomplex graph neural networks have a constraint that they can only carry out hypercomplex graph convolutions in a predetermined and unchangeable dimension. To address this limitation, this paper presents a solution to overcome this limitation by introducing the FFT-based Adaptive Fourier hypercomplex graph convolution filtering mechanism (FAF mechanism), which can adaptively select hypercomplex graph convolutions with the most appropriate dimensions for different situations by projecting the outputs from all candidate hypercomplex graph convolutions to the frequency domain and selecting the one with the highest energy via the FFT-based Adaptive Fourier Decomposition. Meanwhile, we apply the FAF mechanism to our proposed hypercomplex high-order interaction graph neural network (HHG-Net), which performs high-order interaction and strengthens interaction features through quantum graph hierarchical attention module and feature interaction gated graph convolution. During convolution filtering, the FAF mechanism projects the outputs from different candidate hypercomplex graph convolutions to the frequency domain, extracts their energy, and selects the convolution that outputs the largest energy. After that, the model with selected hypercomplex graph convolutions is trained again. Our method outperforms many benchmarks, including the model with hypercomplex graph convolutions selected by DARTS, in node classification, graph classification, and text classification. This showcases the versatility of our approach, which can be effectively applied to both graph and text data.
- 30 August 2024
- Published by Elsevier in Expert Systems with Applications
- Vol. 259
The Structure-sharing Hypergraph Reasoning Attention Module for CNNs
- 30 August 2024
- Published by Elsevier in Expert Systems with Applications
- Vol. 259
The Structure-sharing Hypergraph Reasoning Attention Module for CNNs