David Yu-Tung Hui

David talking

MSc in Informatique at Université de Montréal

David is a MSc student at Mila, co-advised by Pierre-Luc Bacon and Aaron Courville. Before that, he was a research assistant to Yoshua Bengio and collaborated with Dzmitry Bahdanau. Having previously studied Computer Science with Physics and Mathematics at Trinity College Cambridge, he also has experience in data science and startups, interning at Facebook and Tractable. When not doing research, David enjoys sharing his love for running, assosciation football (soccer), violin and piano with friends and family.

Short CV. Please contact me on LinkedIn.

Research Interests

I aim to create software enabling anybody to train any robot to perform any task. Deep Reinforcement Learning (DRL) offers a promising route to achieve this in theory, but is inconsistent in practice and requires far too many interactions with the environment. My research therefore aims to improve the consistency and sample-efficiency of DRL by stabilising its sensitivity to hyperparameters, random seeds and changes in the environment.

Long Term: the Wozniak test; natural-language instruction-following

Short-Medium Term: adaptive agents; model-free RL; meta-learned optimisers; energy-based models; targeted exploration

Please refer to Google Scholar for a complete list of my preprints, publications and citations.