Zhiqiang Liu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Zhiqiang Liu has a total of 44 co-authors. Here are the top co-authors by number of shared publications:
- Jie Gao
Tianjin University
7 shared publications - Mei Yu
School of Computer Science and Technology, Tianjin University, Tianjin, China
6 shared publications - Ruiguo Yu
College of Intelligence and Computing, Tianjin University, Tianjin 300350, P. R. China
6 shared publications - Xuewei Li
Tianjin University
6 shared publications Xuewei Li
5 shared publicationsJie Gao
3 shared publications
Latest Publications
- 7 October 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 334
RLDAL: A reinforcement learning-driven active learning framework for region-level sampling in medical image segmentation
- 7 October 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 334
RLDAL: A reinforcement learning-driven active learning framework for region-level sampling in medical image segmentation
- 2 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. PP (10709908) , 1-5
DVCL: Dual-level View Consistency Learning for Semi-supervised Medical Image Segmentation
Semi-supervised medical image segmentation reduces annotation cost by leveraging abundant unlabeled data, but its performance is often limited by unreliable pseudo-labels in anatomically ambiguous regions. To improve pseudo-label quality, existing dual-network semi-supervised learning (SSL) methods typically use prediction agreement between branches as an indicator of pseudo-label reliability. However, agreement alone may arise from correlated errors and does not reflect intra-model semantic stability. We therefore formulate reliable pseudo-supervision in dual-network learning as a dual-level problem involving both inter-model reliability and intra-model reliability. To address this issue, we propose Dual-level View Consistency Learning (DVCL), which improves pseudo-label reliability from both perspectives. Specifically, an Inter-model View Regularization module reduces branch coupling to promote more trustworthy consensus, while an Intra-model View Consistency module exploits cross-scale semantic agreement to identify and suppress unstable pseudo-supervision within each network. Experiments on the LA, ACDC, and PTU datasets show that DVCL consistently outperforms representative SSL methods under limited-label settings on both overlap and boundary metrics.
- 2 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. PP (10709908) , 1-5
DVCL: Dual-level View Consistency Learning for Semi-supervised Medical Image Segmentation
Semi-supervised medical image segmentation reduces annotation cost by leveraging abundant unlabeled data, but its performance is often limited by unreliable pseudo-labels in anatomically ambiguous regions. To improve pseudo-label quality, existing dual-network semi-supervised learning (SSL) methods typically use prediction agreement between branches as an indicator of pseudo-label reliability. However, agreement alone may arise from correlated errors and does not reflect intra-model semantic stability. We therefore formulate reliable pseudo-supervision in dual-network learning as a dual-level problem involving both inter-model reliability and intra-model reliability. To address this issue, we propose Dual-level View Consistency Learning (DVCL), which improves pseudo-label reliability from both perspectives. Specifically, an Inter-model View Regularization module reduces branch coupling to promote more trustworthy consensus, while an Intra-model View Consistency module exploits cross-scale semantic agreement to identify and suppress unstable pseudo-supervision within each network. Experiments on the LA, ACDC, and PTU datasets show that DVCL consistently outperforms representative SSL methods under limited-label settings on both overlap and boundary metrics.
- 1 June 2026
- Published by Elsevier in Knowledge-Based Systems
- Vol. 347
STLIB: An information bottleneck-guided LLM-spatiotemporal framework for traffic forecasting
- 1 June 2026
- Published by Elsevier in Knowledge-Based Systems
- Vol. 347
STLIB: An information bottleneck-guided LLM-spatiotemporal framework for traffic forecasting
- Open Access
Focusing Image Generation to Mitigate Spurious Correlations
Instance features in images exhibit spurious correlations with background features, affecting the training process of deep neural classifiers. This leads to insufficient attention to instance features by the classifier, resulting in erroneous classification outcomes. In this paper, we propose a data augmentation method called Spurious Correlations Guided Synthesis (SCGS) that mitigates spurious correlations through image generation model. This approach does not require expensive spurious attribute (group) labels for the training data and can be widely applied to other debiasing methods. Specifically, SCGS first identifies the incorrect attention regions of a pre-trained classifier on the training images, and then uses an image generation model to generate new training data based on these incorrect attended regions. SCGS increases the diversity and scale of the dataset to reduce the impact of spurious correlations on classifiers. Changes in the classifier's attention regions and experimental results on three different domain datasets demonstrate that this method is effective in reducing the classifier's reliance on spurious correlations.
Open AccessFocusing Image Generation to Mitigate Spurious Correlations
Instance features in images exhibit spurious correlations with background features, affecting the training process of deep neural classifiers. This leads to insufficient attention to instance features by the classifier, resulting in erroneous classification outcomes. In this paper, we propose a data augmentation method called Spurious Correlations Guided Synthesis (SCGS) that mitigates spurious correlations through image generation model. This approach does not require expensive spurious attribute (group) labels for the training data and can be widely applied to other debiasing methods. Specifically, SCGS first identifies the incorrect attention regions of a pre-trained classifier on the training images, and then uses an image generation model to generate new training data based on these incorrect attended regions. SCGS increases the diversity and scale of the dataset to reduce the impact of spurious correlations on classifiers. Changes in the classifier's attention regions and experimental results on three different domain datasets demonstrate that this method is effective in reducing the classifier's reliance on spurious correlations.
- 10 November 2023
- Published by Springer Nature
- p. 286-297
StyleDisentangle: Disentangled Image Editing Based on StyleGAN2
- 10 November 2023
- Published by Springer Nature
- p. 286-297
StyleDisentangle: Disentangled Image Editing Based on StyleGAN2
- 31 July 2023
- Published by Springer Nature
- p. 181-192
An Industrial Defect Detection Network with Fine-Grained Supervision and Adaptive Contrast Enhancement
- 31 July 2023
- Published by Springer Nature
- p. 181-192
An Industrial Defect Detection Network with Fine-Grained Supervision and Adaptive Contrast Enhancement
- 30 July 2023
- Published by Springer Nature
- p. 16-27
A Prior-Guided Generative Adversarial Net for Semantically Strict Ultrasound Images Augmentation
- 30 July 2023
- Published by Springer Nature
- p. 16-27
A Prior-Guided Generative Adversarial Net for Semantically Strict Ultrasound Images Augmentation
- 18 November 2022
- Published by Springer Nature in International Journal of Machine Learning and Cybernetics
- Vol. 15 (1) , 51-63
Spatio-temporal graph cross-correlation auto-encoding network for wind power prediction
- 18 November 2022
- Published by Springer Nature in International Journal of Machine Learning and Cybernetics
- Vol. 15 (1) , 51-63
Spatio-temporal graph cross-correlation auto-encoding network for wind power prediction
- 5 September 2022
- Published by Springer Nature in Applied Intelligence
- Vol. 53 (10) , 11403-11419
Time series cross-correlation network for wind power prediction
- 5 September 2022
- Published by Springer Nature in Applied Intelligence
- Vol. 53 (10) , 11403-11419
Time series cross-correlation network for wind power prediction