Publications
Peer-reviewed work on multimodal medical AI, LLM reasoning, and anomaly detection. † denotes the corresponding author.
Process Reward Model as Cross-Task Learner: An Extension of Vision-Language Model to Zero-Shot Medical Reasoning
Jiahua Zhang, Yidong Tian, Jinghao Liang, Dianhan Lin, Yiwen Cai, Yuanqing Liu, Zishan Huang, Jingchun Ni, Jianxing He†
Proceedings of the 34th ACM International Conference on Multimedia (ACM MM) 2026
MedTRACE trains a lightweight process reward model once, then reuses it to steer a frozen vision-language model across unseen CT and whole-slide-image tasks with no gradient updates at deployment.
Can LLM Understand Medical Imaging with Long Context?
Jiahua Zhang, Jinghao Liang, Yiwen Cai, Jingchun Ni, Jianxing He
Proceedings of the IEEE International Conference on Multimedia and Expo (ICME) 2026
Oral at ICME 2026. Frozen CT/WSI encoders are compressed and optimal-transport-aligned into a long-context LLM as pseudo-tokens, giving few-shot diagnosis, staging and response assessment with no fine-tuning on either side.
DiffiT-HSFDA: Diffusion Based Source-Free Domain Adaptation for Histopathology
Jiahua Zhang, Yidong Tian
Advanced Intelligent Computing Technology and Applications (ICIC) 2025
Oral at ICIC 2025. Weakly supervised segmentation of lung-cancer histopathology degrades when staining protocols change; a diffusion-based source-free adaptation pipeline restores accuracy without access to source data or any new annotation.
LayerLog: Log Sequence Anomaly Detection Based on Hierarchical Semantics
Chunkai Zhang, Xinyu Wang, Hongye Zhang, Jiahua Zhang, Hanyu Zhang, Chuanyi Liu, Peiyi Han
Applied Soft Computing 2023
Detects anomalies in system logs by modelling semantics hierarchically, from words to log templates to sequences, rather than treating each log line as an opaque token. Applied Soft Computing, JCR Q1 (IF 7.8).
VESC: A New Variational Autoencoder Based Model for Anomaly Detection
Chunkai Zhang, Xinyu Wang, Jiahua Zhang, Shaocong Li, Hanyu Zhang, Chuanyi Liu, Peiyi Han
International Journal of Machine Learning and Cybernetics 2022
A variational autoencoder variant that reshapes the latent space so reconstruction error separates normal from anomalous samples more cleanly. International Journal of Machine Learning and Cybernetics (IF 2.9).