Liheng Ma
PhD Candidate · McGill University & Mila
I am a PhD candidate in Electrical & Computer Engineering at McGill University and Mila, advised by Prof. Mark Coates. I received my M.Sc. in Computer Science from the same institutions, supervised by Prof. Reihaneh Rabbany and Dr. Adriana Romero‑Soriano, and was a visiting student with the Torr Vision Group at the University of Oxford, working with Prof. Philip Torr.
My doctoral research focuses on expressive neural network architectures for learning real-world signals with geometric symmetries and temporal dynamics—mainly graphs (permutation symmetry group) and time-series (temporal dynamics). I develop neural networks with strong perception capacity, expressivity, and efficiency.
More recently, I have explored RL post-training for LLM reasoning and efficient world-action models. My current research focuses on developing low-latency world action models. My work has been published at ICML, AISTATS, AAAI, AAMAS, KDD, etc., and has accumulated over 1,000 citations.