Pierre-Luc Bacon

Pierre-Luc Bacon is a reinforcement learning researcher whose option-critic architecture learns temporally extended actions from data. After a PhD at McGill University with Doina Precup in 2018 and a postdoc at Stanford with Emma Brunskill, he joined Université de Montréal in December 2019 as a Facebook CIFAR AI Chair. With Aaron Courville and others, he coauthored The Primacy Bias in Deep Reinforcement Learning in 2022, and in February 2026 he and Adrien Goldszal described RefGen, an AI system that searches for new refrigerant molecules. He is an associate professor at Université de Montréal, a core academic member of Mila and a Canada CIFAR AI Chair renewed in 2025.

Now at: Université de Montréal and Mila

Timeline

2016–2017

The option-critic architecture

In 2016, he posts The Option-Critic Architecture with Jean Harb and Doina Precup, accepted at AAAI in 2017. The paper derives policy gradient theorems for options (temporally extended actions) and learns their internal policies and termination conditions together with the policy over options. In September 2017, he coauthors When Waiting is not an Option with Jean Harb, Martin Klissarov and Doina Precup, which defines good options through a deliberation cost.
2017–2018

Awards and a McGill PhD

In 2017, he receives the Outstanding Student Paper Award from the Association for the Advancement of AI and a best paper award at the Hierarchical Reinforcement Learning Workshop at NeurIPS. He earns his PhD at McGill University with Doina Precup in 2018, with the thesis Temporal Representation Learning.
By 2019

Postdoc at Stanford

He works as a postdoc at Stanford with Emma Brunskill, at the Stanford AI for Human Impact Lab, studying long-horizon learning, off-policy methods and temporal abstraction. In 2019, he coauthors Goal-Directed Learning as a Bi-level Optimization Problem with Dilip Arumugam and Brunskill for RLDM.
December 2019

Joining Université de Montréal

In December 2019, he joins the Department of Computer Science and Operations Research (DIRO) at Université de Montréal as an assistant professor, his first faculty position in Canada. CIFAR names him one of its two Facebook CIFAR AI Chairs, alongside Siva Reddy, with Mila and Université de Montréal as his affiliations.
2022

The primacy bias in deep RL

In 2022, he coauthors The Primacy Bias in Deep Reinforcement Learning with Evgenii Nikishin, Max Schwarzer, Pierluca D'Oro and Aaron Courville, published at ICML. The paper finds that deep RL agents overfit to early interactions and proposes periodically resetting part of the agent, which it applies to Atari 100k and the DeepMind Control Suite.
2025

A renewed CIFAR AI Chair

In 2025, CIFAR renews his Canada CIFAR AI Chair. Also in 2025, he coauthors Network Sparsity Unlocks the Scaling Potential of Deep Reinforcement Learning with Guozheng Ma, Lu Li, Zilin Wang, Li Shen and Dacheng Tao for ICML, where one-shot random pruning of network weights before training lets deep RL networks scale beyond their dense counterparts.
February 2026

RefGen for new refrigerants

In February 2026, he and Adrien Goldszal describe RefGen, a physics-informed AI system that fine-tunes a language model on molecular structures and uses reinforcement learning with thermodynamic models to generate refrigerant molecules. After generating over 1 million molecules and filtering them with expert thresholds, RefGen identifies more than 800 viable candidates.
As of October 2026

Associate professor at Université de Montréal

He is an associate professor in DIRO at Université de Montréal, a core academic member of Mila and a Canada CIFAR AI Chair. His research covers reinforcement learning and optimal control, and his course book Building Up RL starts from dynamics, trajectory optimization and model predictive control and builds up to reinforcement learning.