Doina Precup

Sumana Basu

PhD - McGill University

Co-supervisor :

Adriana Romero Soriano

Collaborating Alumni - McGill University

Lynn Cherif

Master's Research - McGill University

Co-supervisor :

PhD - McGill University

Co-supervisor :

PhD - McGill University

Principal supervisor :

David Meger

Jonathan Colaço Carr

Master's Research - McGill University

Principal supervisor :

Prakash Panangaden

Élodie Coté-Gauthier

Collaborating researcher - McGill University

Franco Del Balso

Research Intern - Université de Montréal

Jesse Farebrother

PhD - McGill University

Principal supervisor :

Marc Gendron-Bellemare

PhD - McGill University

Principal supervisor :

PhD - McGill University

Haque Ishfaq

Collaborating Alumni - McGill University

Mohammad Sami Nur Islam Islam

Master's Research - McGill University

Arushi Jain

PhD - McGill University

Postdoctorate - McGill University

Elaine Lau

Master's Research - McGill University

Jonathan Lebensold

Collaborating Alumni - McGill University

Undergraduate - McGill University

Ray Luo

PhD - McGill University

Principal supervisor :

G McCracken

PhD - McGill University

Nazanin Mohammadi Sepahvand

Collaborating Alumni - McGill University

Shahrad Mohammadzadeh

Master's Research - McGill University

Principal supervisor :

Gabriela Moisescu-Pareja

Collaborating researcher - McGill University

Co-supervisor :

Irina Rish

Padideh Nouri

PhD - Université de Montréal

Co-supervisor :

PhD - McGill University

Co-supervisor :

Nate Rahn

PhD - McGill University

Principal supervisor :

Marc Gendron-Bellemare

Sahand Rezaei-Shoshtari

PhD - McGill University

Co-supervisor :

PhD - McGill University

Co-supervisor :

PhD - McGill University

Co-supervisor :

PhD - McGill University

Nishanth Anand Vemgal

PhD - McGill University

PhD - McGill University

Co-supervisor :

Samira Ebrahimi Kahou

Zihan Wang

PhD - McGill University

Skipper: Combining Spatial and Temporal Abstraction for Better Generalization

Guangyuan Wang

Research Intern - McGill University

Steve Wen

Master's Research - McGill University

Co-supervisor :

Gregory Dudek

Zijing Wu

PhD - McGill University

Co-supervisor :

PhD - McGill University

Harry Zhao

Collaborating Alumni - McGill University

Co-supervisor :

Blog Posts

Generic thumbnail for Mila Blog articles.

February 22, 2024

Mingde Harry Zhao

Safa Alver

Harm van Seijen

Romain Laroche

Doina Precup

Yoshua Bengio

Read the article

Publications

Incorporating Spatial Information into Goal-Conditioned Hierarchical Reinforcement Learning via Graph Representations

Shuyuan Zhang

Zihan Wang

Xiao-Wen Chang

The integration of graphs with Goal-conditioned Hierarchical Reinforcement Learning (GCHRL) has recently gained attention, as intermediate g… (see more)oals (subgoals) can be effectively sampled from graphs that naturally represent the overall task structure in most RL tasks. However, existing approaches typically rely on domain-specific knowledge to construct these graphs, limiting their applicability to new tasks. Other graph-based approaches create graphs dynamically during exploration but struggle to fully utilize them, because they have problems passing the information in the graphs to newly visited states. Additionally, current GCHRL methods face challenges such as sample inefficiency and poor subgoal representation. This paper proposes a solution to these issues by developing a graph encoder-decoder to evaluate unseen states. Our proposed method, Graph-Guided sub-Goal representation Generation RL (G4RL), can be incorporated into any existing GCHRL method when operating in environments with primarily symmetric and reversible transitions to enhance performance across this class of problems. We show that the graph encoder-decoder can be effectively implemented using a network trained on the state graph generated during exploration. Empirical results indicate that leveraging high and low-level intrinsic rewards from the graph encoder-decoder significantly enhances the performance of state-of-the-art GCHRL approaches with an extra small computational cost in dense and sparse reward environments.

2025-10-06

TMLR (accepted)

Incorporating Spatial Information into Goal-Conditioned Hierarchical Reinforcement Learning via Graph Representations

Shuyuan Zhang

Zihan Wang

Xiao-Wen Chang

2025-10-06

TMLR (accepted)

Rejecting Hallucinated State Targets during Planning

Romain Laroche

In planning processes of computational decision-making agents, generative or predictive models are often used as "generators" to propose "ta… (see more)rgets" representing sets of expected or desirable states. Unfortunately, learned models inevitably hallucinate infeasible targets that can cause delusional behaviors and safety concerns. We first investigate the kinds of infeasible targets that generators can hallucinate. Then, we devise a strategy to identify and reject infeasible targets by learning a target feasibility evaluator. To ensure that the evaluator is robust and non-delusional, we adopted a design choice combining off-policy compatible learning rule, distributional architecture, and data augmentation based on hindsight relabeling. Attaching to a planning agent, the designed evaluator learns by observing the agent’s interactions with the environment and the targets produced by its generator, without the need to change the agent or its generator. Our controlled experiments show significant reductions in delusional behaviors and performance improvements for various kinds of existing agents.

2025-10-06

Proceedings of the 42nd International Conference on Machine Learning (published)

proceedings.mlr.press

Reward the Reward Designer: Making Reinforcement Learning Useful for Clinical Decision Making

Sumana Basu

Adriana Romero Soriano

2025-09-22

NeurIPS.cc/2025/Workshop/WiML (published)

Unifying Mechanistic Interpretations of Neural Networks Trained on Modular Addition

Gabriela Moisescu-Pareja

Gavin McCracken

Vincent Létourneau

Jonathan Love

2025-09-22

NeurIPS.cc/2025/Workshop/WiML (published)

RL Fine-Tuning Heals OOD Forgetting in SFT

Hangzhan Jin

Sicheng Lyu

Mohammad Hamdaqa

The two-stage fine-tuning paradigm of Supervised Fine-Tuning (SFT) followed by Reinforcement Learning (RL) has empirically shown better reas… (see more)oning performance than one-stage SFT for the post-training of Large Language Models (LLMs). However, the evolution and mechanism behind the synergy of SFT and RL are still under-explored and inconclusive. In our study, we find the well-known claim"SFT memorizes, RL generalizes"is over-simplified, and discover that: (1) OOD performance peaks at the early stage of SFT and then declines (OOD forgetting), the best SFT checkpoint cannot be captured by training/test loss; (2) the subsequent RL stage does not generate fundamentally better OOD capability, instead it plays an \textbf{OOD restoration} role, recovering the lost reasoning ability during SFT; (3) The recovery ability has boundaries, \ie{} \textbf{if SFT trains for too short or too long, RL cannot recover the lost OOD ability;} (4) To uncover the underlying mechanisms behind the forgetting and restoration process, we employ SVD analysis on parameter matrices, manually edit them, and observe their impacts on model performance. Unlike the common belief that the shift of model capacity mainly results from the changes of singular values, we find that they are actually quite stable throughout fine-tuning. Instead, the OOD behavior strongly correlates with the \textbf{rotation of singular vectors}. Our findings re-identify the roles of SFT and RL in the two-stage fine-tuning and discover the rotation of singular vectors as the key mechanism. %reversing the rotations induced by SFT, which shows recovery from forgetting, whereas imposing the SFT parameter directions onto a RL-tuned model results in performance degradation. Code is available at https://github.com/xiaodanguoguo/RL_Heals_SFT

2025-09-08

ArXiv (preprint)

RL Fine-Tuning Heals OOD Forgetting in SFT

Hangzhan Jin

Sicheng Lyu

Mohammad Hamdaqa

2025-09-08

ArXiv (preprint)

RL Fine-Tuning Heals OOD Forgetting in SFT

Hangzhan Jin

Sicheng Lyu

Mohammad Hamdaqa

2025-09-08

ArXiv (preprint)

RL Fine-Tuning Heals OOD Forgetting in SFT

Hangzhan Jin

Sicheng Lyu

Mohammad Hamdaqa

The two-stage fine-tuning paradigm of Supervised Fine-Tuning (SFT) followed by Reinforcement Learning (RL) has empirically shown better reas… (see more)oning performance than one-stage SFT for the post-training of Large Language Models (LLMs). However, the evolution and mechanism behind the synergy of SFT and RL are still under-explored and inconclusive. In our study, we find the well-known claim "SFT memorizes, RL generalizes" is over-simplified, and discover that: (1) OOD performance peaks at the early stage of SFT and then declines (OOD forgetting), the best SFT checkpoint cannot be captured by training/test loss; (2) the subsequent RL stage does not generate fundamentally better OOD capability, instead it plays an \textbf{OOD restoration} role, recovering the lost reasoning ability during SFT; (3) The recovery ability has boundaries, \ie{} \textbf{if SFT trains for too short or too long, RL cannot recover the lost OOD ability;} (4) To uncover the underlying mechanisms behind the forgetting and restoration process, we employ SVD analysis on parameter matrices, manually edit them, and observe their impacts on model performance. Unlike the common belief that the shift of model capacity mainly results from the changes of singular values, we find that they are actually quite stable throughout fine-tuning. Instead, the OOD behavior strongly correlates with the \textbf{rotation of singular vectors}. Our findings re-identify the roles of SFT and RL in the two-stage fine-tuning and discover the rotation of singular vectors as the key mechanism. %reversing the rotations induced by SFT, which shows recovery from forgetting, whereas imposing the SFT parameter directions onto a RL-tuned model results in performance degradation. Code is available at https://github.com/xiaodanguoguo/RL_Heals_SFT

2025-09-01

arXiv (published)

Relative Trajectory Balance is equivalent to Trust-PCL

Recent progress in generative modeling has highlighted the importance of Reinforcement Learning (RL) for fine-tuning, with KL-regularized me… (see more)thods in particular proving to be highly effective for both autoregressive and diffusion models. Complementing this line of work, the Relative Trajectory Balance (RTB) objective was recently introduced in the context of Generative Flow Networks (GFlowNets) to serve the same role of improving fine-tuning in sequential generative models. Building on prior work linking GFlowNets and maximum-entropy RL, we establish in this paper an equivalence between RTB and Trust-PCL, an off-policy RL method with KL regularization. This equivalence situates RTB within the broader theoretical landscape of KL-regularized RL, and clarifies its relationship to earlier methods. Leveraging this insight, we revisit an illustrative example from the RTB paper and show that KL-regularized RL methods achieve comparable performance, offering an alternative perspective to what was previously reported.

2025-09-01

arXiv (published)

Relative Trajectory Balance is equivalent to Trust-PCL

2025-09-01

ArXiv (preprint)