Zesheng Ye
University of Melbourne. Melbourne Connect (Building 290), University of Melbourne, VIC 3010, Australia.

Greetings. I am currently a Postdoctoral Research Fellow at Trustworthy Machine Learning and Reasoning Group, School of Computing and Information Systems, The University of Melbourne, working with Dr. Feng Liu.
I research machine learning models that are
- robust against distribution shifts
- efficient during deployment and adaptation for new tasks
- safe in terms of adversarial attack and privacy breach
given the pressing needs for trustworthy machine-learning systems, in contexts of powerful multi-modal Foundation Models emerging nowadays.
I obtained my Ph.D. and Master’s degrees from The University of New South Wales, feeling fortunate to be supervised by Prof. Lina Yao, where I studied uncertainty-aware and sample-efficient human-centric understanding with limited well-annotated data. Our research outcomes spanned multiple real-world application, such as Recommender Systems, Brain-Computer Interface, and Human Movement Predictions. I did my undergraduate study at Southwest Jiaotong University.
news
Feb 14, 2025 | We will present a tutorial on Neural Network Reprogrammability at DASFAA 2025. |
---|---|
Feb 05, 2025 | Our latest finding of Membership Inference Attack for Generative Models, is now in preprint. |
Jan 23, 2025 | We have 1 paper on Model Reprogramming accepted by ICLR 2025. |
Nov 26, 2024 | Our latest finding of Cross-Domain Recommendation, following our SIGIR 2024 paper, is now in preprint. |
Sep 26, 2024 | We have 1 paper on Model Reprogramming accepted by NeurIPS 2024 as Oral presentation. |
selected publications
- Attribute-based Visual Reprogramming for Image Classification with CLIPIn ICLR, 2025
- Bayesian-guided Label Mapping for Visual ReprogrammingIn NeurIPS, 2024
- Sample-specific masks for visual reprogramming-based promptingIn ICML, 2024
- Contrastive conditional neural processesIn CVPR, 2022
- Towards Robust Cross-Domain Recommendation with Joint Identifiability of User PreferencearXiv preprint arXiv:2411.17361, 2024