Author name cluster

Zhisong Pan

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

7 papers

1 author row

EAAI Journal 2026 Journal Article

Multi-task time series forecasting with adaptive graph neural networks based on feature uncertainty

Xiao Han
Zhisong Pan
Yongjie Huang

Multi-task time series forecasting aims to enhance prediction accuracy by leveraging shared knowledge among related tasks, finding widespread applications in critical domains such as financial risk analysis and medical monitoring. However, existing methods often overlook the impact of feature uncertainty on knowledge reliability and fail to dynamically model cross-timestep task relationships. This leads to challenges like negative transfer and the inability of static sharing mechanisms to adapt to temporal dynamics. To address these issues, this paper proposes DPG-Net, a Dynamic Probabilistic Graph Network that utilizes a Bayesian framework to model task features as Gaussian random variables, thereby quantifying their uncertainty. This uncertainty guides a gated attention mechanism to dynamically construct a cross-timestep probabilistic graph, enabling adaptive and reliable knowledge sharing among tasks. Experimental validation on multiple clinical risk prediction datasets demonstrates that DPG-Net achieves superior average performance in terms of AUROC on the MIMIC-III Infection, PhysioNet, MIMIC-III Heart Failure, and MIMIC-III Respiratory Failure datasets compared to state-of-the-art models, with improvements of 11. 48%, 8. 30%, 10. 48%, and 7. 42%, respectively, highlighting its capability to improve prediction accuracy and mitigate negative transfer. Ablation studies further confirm the effectiveness of the probabilistic modeling and dynamic knowledge-sharing mechanisms.

Details DOI

EAAI Journal 2026 Journal Article

Progressive category-aware anti-distillation

Yao Zhang
Yang Li
Zhisong Pan

The widespread use of knowledge distillation has intensified the risk of model theft, revealing the inadequacy of traditional protection techniques against such threats. Anti-distillation has emerged as a promising defense paradigm by disrupting knowledge transfer to prevent unauthorized extraction of dark knowledge while preserving the teacher model’s performance. However, existing anti-distillation methods largely overlook the pivotal role of inter-class relationships in the distillation process. To address this limitation, we propose a progressive category-aware anti-distillation method. Our approach first constructs a relationship matrix between class prototypes to accurately model inter-class relationships, then reconstructs the output distribution to eliminate inter-class correlation information in output. To enhance stability and maintain distributional symmetry, we replace the standard Kullback–Leibler divergence with a symmetric Jensen–Shannon divergence. Moreover, we implement a curriculum learning mechanism to progressively adjust the intensity of inter-class correlation information removal. Extensive experiments on Cifar-100 and ImageNet demonstrate that our approach consistently surpasses existing anti-distillation methods, achieving strong robustness across various architectures — including Convolutional Neural Networks and Transformers — and different distillation settings such as logits-based, feature-based, and data-free distillation, with less than 2. 2% degradation in teacher performance.

Details DOI

IJCAI Conference 2024 Conference Paper

Provable Acceleration of Nesterov’s Accelerated Gradient Method over Heavy Ball Method in Training Over-Parameterized Neural Networks

Xin Liu
Wei Tao
Wei Li
Dazhi Zhan
Jun Wang
Zhisong Pan

Due to its simplicity and efficiency, the first-order gradient method has been extensively employed in training neural networks. Although the optimization problem of the neural network is non-convex, recent research has proved that the first-order method is capable of attaining a global minimum during training over-parameterized neural networks, where the number of parameters is significantly larger than that of training instances. Momentum methods, including the heavy ball (HB) method and Nesterov's accelerated gradient (NAG) method, are the workhorse of first-order gradient methods owning to their accelerated convergence. In practice, NAG often exhibits superior performance than HB. However, current theoretical works fail to distinguish their convergence difference in training neural networks. To fill this gap, we consider the training problem of the two-layer ReLU neural network under over-parameterization and random initialization. Leveraging high-resolution dynamical systems and neural tangent kernel (NTK) theory, our result not only establishes tighter upper bounds of the convergence rate for both HB and NAG, but also provides the first theoretical guarantee for the acceleration of NAG over HB in training neural networks. Finally, we validate our theoretical results on three benchmark datasets.

PDF Details DOI

IJCAI Conference 2022 Conference Paper

Learning Coated Adversarial Camouflages for Object Detectors

Yexin Duan
Jialin Chen
Xingyu Zhou
Junhua Zou
Zhengyun He
Jin Zhang
Wu Zhang
Zhisong Pan

An adversary can fool deep neural network object detectors by generating adversarial noises. Most of the existing works focus on learning local visible noises in an adversarial "patch" fashion. However, the 2D patch attached to a 3D object tends to suffer from an inevitable reduction in attack performance as the viewpoint changes. To remedy this issue, this work proposes the Coated Adversarial Camouflage (CAC) to attack the detectors in arbitrary viewpoints. Unlike the patch trained in the 2D space, our camouflage generated by a conceptually different training framework consists of 3D rendering and dense proposals attack. Specifically, we make the camouflage perform 3D spatial transformations according to the pose changes of the object. Based on the multi-view rendering results, the top-n proposals of the region proposal network are fixed, and all the classifications in the fixed dense proposals are attacked simultaneously to output errors. In addition, we build a virtual 3D scene to fairly and reproducibly evaluate different attacks. Extensive experiments demonstrate the superiority of CAC over the existing attacks, and it shows impressive performance both in the virtual scene and the real world. This poses a potential threat to the security-critical computer vision systems.

PDF Details DOI

AAAI Conference 2022 Conference Paper

Making Adversarial Examples More Transferable and Indistinguishable

Junhua Zou
Yexin Duan
Boyu Li
Wu Zhang
Yu Pan
Zhisong Pan

Fast gradient sign attack series are popular methods that are used to generate adversarial examples. However, most of the approaches based on fast gradient sign attack series cannot balance the indistinguishability and transferability due to the limitations of the basic sign structure. To address this problem, we propose a method, called Adam Iterative Fast Gradient Tanh Method (AI-FGTM), to generate indistinguishable adversarial examples with high transferability. Besides, smaller kernels and dynamic step size are also applied to generate adversarial examples for further increasing the attack success rates. Extensive experiments on an ImageNetcompatible dataset show that our method generates more indistinguishable adversarial examples and achieves higher attack success rates without extra running time and resource. Our best transfer-based attack NI-TI-DI-AITM can fool six classic defense models with an average success rate of 89. 3% and three advanced defense models with an average success rate of 82. 7%, which are higher than the state-of-theart gradient-based attacks. Additionally, our method can also reduce nearly 20% mean perturbation. We expect that our method will serve as a new baseline for generating adversarial examples with better transferability and indistinguishability.

PDF Details

EAAI Journal 2021 Journal Article

A fast X-shaped foreground segmentation network with CompactASPP

Jin Zhang
Shuaihui Wang
Junyang Qiu
Xinran Pan
Junhua Zou
Yexin Duan
Zhisong Pan
Yang Li

Foreground segmentation models are designed to extract moving objects of varying sizes from the background, which can benefit from representations of various scales. As an effective module for capturing multi-scale contexts, Atrous Spatial Pyramid Pooling (ASPP) convolves a final feature representation via multiple parallel atrous convolutions with different dilation rates. However, as the dilation rate increases, ASPP gradually loses its large-scale modeling ability because the sampling of atrous kernel becomes progressively sparse within the receptive field. To solve this problem, we design a CompactASPP module to convolve feature maps compactly. Without significantly increasing the module size, the CompactASPP can encode multi-scale features from all neurons within the receptive field rather than from neurons in several sparsely distributed positions. Furthermore, we leverage CompactASPP modules to enhance our previous X-Net. The proposed Fast X-Net substantially improves the segmentation speed by over 63. 6% and attains new state-of-the-art performances on CDnet2014, SBI2015 and UCSD benchmarks.

Details DOI

AAAI Conference 2021 Conference Paper

Gradient Descent Averaging and Primal-dual Averaging for Strongly Convex Optimization

Wei Tao
Wei Li
Zhisong Pan
Qing Tao

Averaging scheme has attracted extensive attention in deep learning as well as traditional machine learning. It achieves theoretically optimal convergence and also improves the empirical model performance. However, there is still a lack of sufficient convergence analysis for strongly convex optimization. Typically, the convergence about the last iterate of gradient descent methods, which is referred to as individual convergence, fails to attain its optimality due to the existence of logarithmic factor. In order to remove this factor, we first develop gradient descent averaging (GDA), which is a general projection-based dual averaging algorithm in the strongly convex setting. We further present primal-dual averaging for strongly convex cases (SC-PDA), where primal and dual averaging schemes are simultaneously utilized. We prove that GDA yields the optimal convergence rate in terms of output averaging, while SC-PDA derives the optimal individual convergence. Several experiments on SVMs and deep learning models validate the correctness of theoretical analysis and effectiveness of algorithms.

PDF Details