Zhiyu An @ UC Merced

("Ryan") Zhiyu An @ UC Merced

I am currently a Postdoctoral Scholar in the Department of Computer Science and Engineering (CSE) at University of California, Merced, School of Engineering. I received my Ph.D. in Electrical Engineering and Computer Science at UC Merced in 2025.

My research focuses on developing safe and responsible AI agents that can assist, support, or act on humans’ behalf in energy, agricultural, and business infrastructures. As AI systems grow more powerful, I ask the following question: what properties must they possess in order to be governable, so they can be trusted to be deployed at scale and benefit the greatest number of people? To address this question, I work on Technical AI Governance, where my research spans uncertainty quantification, uncertainty-informed reinforcement learning, LLM alignment, and AI ethics.

My work has been published in top venues across AI, ML, Design Automation, Control, and Cyber-Physical Systems including AAAI, ICLR, DAC, KDD, L4DC, BuildSys, and IEEE IoT-J. I received Best Paper Award Runner-Up at BuildSys 2023 and Best Paper Award at HICSS 2025.

Email: zan7 AT ucmerced.edu

[Google Scholar]

Contact

Department of Computer Science and Engineering

School of Engineering

University of California, Merced

Address: #230V, SE2, 5200 N. Lake Road, Merced, CA 95343

Education

Ph.D. in Electrical Engineering and Computer Science, University of California, Merced

B.A. in Computer Science, University of California, Berkeley

News

(2026.01) One preprint now online! arxiv. Representational Homomorphism Predicts and Improves Compositional Generalization in Transformer Language Models.
(2026.01) One preprint now online! arxiv. Differential Voting: Loss Functions for Axiomatically Diverse Aggregation of Heterogeneous Preferences.
(2026.01) One preprint now online! arxiv. We found a way to reverse-engineering (and audit) a hidden neural incentive mechanism by observing strategic interactions of multiple agents.
(2026.01) One paper on noise-robust exploration for RL accepted to ICLR 2026! arXiv.
(2025.11) One first author paper on moral alignment of LLM agents using reasoning-level reinforcement learning accepted to AAAI 2026!
(2025.09) We have released our newest work on noise-robust exploration for RL on arXiv.
(2025.09) One first author paper on compositional generalization in language models accepted to Proceedings of the NeurIPS 2025 Workshop on Symmetry and Geometry of Neural Representations and the NeurIPS 2025 First Workshop on the Foundations of Reasoning in Language Models.
(2025.09) I'm serving on the Program Committee of AAAI 2026.
(2025.07) Started as Postdoctoral Scholar at UC Merced!
(2025.05) I have successfully defended my Doctoral Dissertation!
(2025.03) One first author paper on uncertainty quantification accepted to L4DC 2025!
(2025.01) Received Best Paper Award at HICSS 2025!
(2025.01) One first author paper now online at IEEE IoT-J.

Still Life with Compass Robot and Teacup of Code

Perhaps this is our last research agenda—after optimization, after scale—where remaining question is not how far machines can go, but which direction we allow them to choose.