Thomas Jiralerspong

I am a graduate student in computer science co-supervised by Yoshua Bengio and Doina Precup at Mila and Université de Montréal. I previously completed my Bachelor’s in honours computer science at McGill University working with Blake Richards and Doina Precup.

My two main research goals are:

• Building safe and aligned AI systems that have more human-like capabilities such as system 2 thinking, out-of-distribution generalization, long-term planning, and efficient learning.
• Applying AI to projects that have a concrete positive impact on society

Concrete directions I am interested in (and have explored) include:

Safe and aligned AI With Human-Like Capabilities

• LLMs and VLMs (papers on causal graph discovery with LLMs and compositionality in LLMs, internship at Occam AI)
• Input-driven learning (paper on bias-only learning)
• Compositionality (papers on complexity-based theory of compositionality and compositionality in LLMs)
• Discrete Representations (paper on discovering discrete subgoals for RL)
• Causality (paper on causal graph discovery with LLMs)
• Taking inspiration from cognitive science, neuroscience, and psychology (papers on bias-only learning and discovering discrete subgoals for RL)
• Model-based reinforcement learning (paper on temporally extended tree-search planning)

AI for Positive Impact

• Healthcare (paper on RL for Mechanical Ventilation)
• Climate Change (paper on RL for HVAC Control)
• Autonomous Driving (Internship at Waabi)
• Drug Discovery

I am currently:

• Chairman of the Lab representatives at Mila
• Senior advisor for the McGill A.I. Society

I have previously been:

• Research intern at:
  • Waabi - working on variational autoencoders for controllable traffic simulation
  • Vector Institute - working on reinforcement learning for HVAC control
• Software development intern at
  • Amazon
  • Square Enix
  • Expedia
• Undergraduate researcher with
  • Blake Richards - working on discovering important states for reinforcement learning in sparse reward environments
  • Doina Precup - working on model-based reinforcement learning in pixel environments
• Selected to be a part of the MIT Brains Minds and Machines Summer Course
• Selected to attend the 10th Heidelberg Laureate Forum
• Co-leader of McGill’s team for Project X, where we received the highest score out of 25 submitted papers for our work on deep reinforcement learning for mechanical ventilation
• Co-organizer and teaching assistant for the McGill A.I. Society’s Accelerated Introduction to ML Bootcamp
• Teaching assistant for:
  • Software Systems (COMP206) at McGill
  • Representation Learning (IFT6135) at Université de Montréal

In my free time, I enjoy traveling, watching/analyzing good movies, and writing sad songs (check out my (very) amateur music here)!

Reach out at thomas.jiralerspong@mila.quebec if there is anything you want to discuss, I’m always happy to talk!