Pierre-Alexandre Kamienny

CV  LinkedIn   Twitter   Scholar

I am a final-year PhD student @{Meta AI, Sorbonne University} working on efficient adaptation of RL agents to unseen tasks. Recently, I have used large language models to regress data manipulating math symbols (symbolic regression) with the hope to make symbolic regressors fast and accurate enough to be applied to model-based RL as intepretable world models.
I am advised by François Charton (Meta AI), Sylvain Lamprier (now at Université d'Angers), and previously Ludovic Denoyer (now at Ubisoft).

My primary research interest is reinforcement learning (RL), particularly enabling RL agents to efficiently adapt to unseen tasks (meta/multi-task RL) by learning "nice" task representations (resp. good coverage policies) in the presence (resp. absence) of training rewards.
As the dynamics of control tasks are commonly governed by physical laws, I embarked upon the quest of developing RL agents that explicitly model dynamics with equations as I believe that it can enable prior knowledge and/or inductive bias injection, sample efficiency gains, better domain randomization (à-la-Sim2Real) and risk-control thanks to interpretability... Practically, this involves using symbolic regression (SR), the search of analytic expressions composed of mathematical operators, e.g. cos, exp, constants and variables. Due to the lack of SR algorithms that infer accurate expressions in reasonable time, I have worked on developing transformer-based models, trained on synthetically-generated datasets, that search with order of magnitudes less time.

Deep Generative Symbolic Regression with Monte-Carlo-Tree-Search and Controllable Neural Symbolic Regression both got accepted to ICML 2023!

• Interned at Facebook AI Research Paris, Nokia Bell Labs and Neoxia.
• Studied CS at the University of Oxford (Keble College) & CentraleSupélec.
• Practice street-lifting, a sport where you need to {muscle-up, pull-up, dips, squat} with as much additional weights as you can. Current PRs are respectively {20, 55, 102.5, 155}.
• Practice urban exploration & rooftoping in Paris!
• Email:   pierrealexandre dot kam at gmail dot com

• Deep Generative Symbolic Regression with Monte-Carlo-Tree-Search
  Pierre-Alexandre Kamienny, Guillaume Lample, Sylvain Lamprier, Marco Virgolin.
  International Conference on Machine Learning (ICML), 2023
  
• Controllable Deep Symbolic Regression
  Tommaso Bendinelli, Luca Biggio, Pierre-Alexandre Kamienny.
  International Conference on Machine Learning (ICML), 2023
• End-to-end symbolic regression with transformers
  Pierre-Alexandre Kamienny*, Stéphane d’Ascoli*, Guillaume Lample, Francois Charton.
  Neural Information Processing Systems (NeurIPS), 2022
  Demo
• Deep symbolic regression for recurrence prediction
  Stéphane d’Ascoli*, Pierre-Alexandre Kamienny*, Guillaume Lample, Francois Charton.
  International Conference on Machine Learning (ICML), 2022
  Demo & Video by Yannic Kilcher
• Direct then Diffuse: Incremental Unsupervised Skill Discovery for State Covering and Goal Reaching
  Pierre-Alexandre Kamienny*, Jean Tarbouriech*, Sylvain Lamprier, Alessandro Lazaric, Ludovic Denoyer.
  International Conference on Learning Representations (ICLR), 2022
• FACMAC: Factored multi-agent centralised policy gradients
  Bei Peng, Tabish Rashid, Christian Schroeder de Witt, Pierre-Alexandre Kamienny, Philip Torr, Wendelin Böhmer, Shimon Whiteson.
  Neural Information Processing Systems (NeurIPS), 2021

• Learning adaptive exploration strategies in dynamic environments through informed policy regularization
  Pierre-Alexandre Kamienny, Matteo Pirotta, Alessandro Lazaric, Thibault Lavril, Nicolas Usunier, Ludovic Denoyer.

• Symbolic Model-Based Reinforcement Learning
  Pierre-Alexandre Kamienny, Sylvain Lamprier.
  Neural Information Processing Systems (NeurIPS), Workshops on "AI for Science" & "Gaussian Processes, Spatiotemporal Modeling, and Decision-making Systems" ”, 2022
• Meta-Reinforcement Learning With Informed Policy Regularization
  Pierre-Alexandre Kamienny, Matteo Pirotta, Alessandro Lazaric, Thibault Lavril, Nicolas Usunier, Ludovic Denoyer.
  International Conference on Machine Learning (ICML), Workshop on "Inductive Biases, Invariances and Generalization in RL", 2020
• Privileged information dropout in reinforcement learning
  Pierre-Alexandre Kamienny, Kai Arulkumaran, Feryal Behbahani, Wendelin Boehmer, Shimon Whiteson.
  International Conference on Learning Representations (ICLR), Workshop on "Beyond tabula rasa in RL", 2020

Website template from here