Haodong Lu

I am Haodong Lu, a Postdoctoral Researcher in the Artificer AI Lab at the University of New South Wales (UNSW), Australia. I completed my PhD at UNSW in 2026 under the supervision of Dr. Dong Gong and Prof. Lina Yao, working closely with CSIRO's Data61 as a scholarship recipient. I am also a recipient of the 2024 Google PhD Fellowship. Before that, I completed both my Honours and BSc. degrees at the University of New South Wales.

My research focuses on AI that keeps learning and improving after deployment. My PhD developed adaptive methods for reliable foundation models under non-stationary conditions: detecting what a model has never seen, and learning continually without forgetting the past. Unified under a framework of modulated learning, this work spans representation and parameter modulation, including prototypical learning, parameter-efficient fine-tuning and mixture of experts, across vision, vision-language and large language models. I now work on self-improving AI: agents and foundation models that keep learning across their lifetime, through test-time learning, agentic memory, continual post-training and reinforcement learning, so that they improve from their own experience and environment feedback.

Research interests: self-improving agents · test-time training · continual and lifelong learning · agentic memory · out-of-distribution detection · parameter-efficient fine-tuning · mixture of experts · vision-language and large language models