Scholar
Pierre-Luc Bacon
Google Scholar ID: 9H77FYYAAAAJ
University of Montreal
Citations & Impact
Citations
3,174
H-index
23
i10-index
31
Publications
20
Co-authors
16
list available
Contact
No contact links provided.
Publications
Resume (English only)
Academic Achievements
- 2024: 'Neural differential equations for temperature control in buildings under demand response programs' published in Applied Energy
- 2024: 'Do Transformer World Models Give Better Policy Gradients?' presented at ICML
- 2024: 'Maximum entropy GFlowNets with soft Q-learning' presented at AISTATS
- 2024: Multiple papers at ICLR including 'Decoupling regularization from the action space', 'Bridging State and History Representations', 'Course Correcting Koopman Representations', and 'Motif: Intrinsic Motivation from Artificial Intelligence Feedback'
- 2023: Oral presentation at NeurIPS – 'When Do Transformers Shine in RL? Decoupling Memory from Credit Assignment'
- 2023: Poster presentations at NeurIPS – 'Block-State Transformers' and 'Policy Optimization in a Noisy Neighborhood'
- 2023: Spotlight paper at NeurIPS – 'Double Gumbel Q-Learning'
- 2023: ICLR notable top 5% paper – 'Sample-Efficient Reinforcement Learning by Breaking the Replay Ratio Barrier'
- 2022: NeurIPS Datasets and Benchmarks paper – 'Myriad: a real-world testbed to bridge trajectory optimization and deep learning'
- 2022: ICML and RLDM papers – 'The Primacy Bias in Deep Reinforcement Learning' and 'Direct Behavior Specification via Constrained Reinforcement Learning'
- 2022: ICLR paper – 'Continuous-Time Meta-Learning with Forward Mode Differentiation'
- 2021: NeurIPS workshop papers – 'Meta Dynamic Programming' and 'Long-Term Credit Assignment via Model-based Temporal Shortcuts'
Background
- Associate Professor at Université de Montréal's DIRO
- CIFAR AI Chair
- Core member of Mila
- Affiliated with the Institute for Data Valorization (IVADO)
- Research at the intersection of theory and application in reinforcement learning
- Focuses on real-world problems in HVAC systems and molecular modeling
- Works on improving RL through representation learning, neural differential equations, and transformer-based models
- Particularly interested in tackling the curse of horizon in long-term planning
- Recently exploring the use of large language models to address specification challenges in RL for better alignment and sample efficiency
