I am a Postdoctoral Researcher in the Department of Computer Science and Operations Research (DIRO) at the University of Montreal, and a member of Mila – Quebec Artificial Intelligence Institute under the supervision of Prof. Irina Rish. I study the capacity of artificial deep neural networks to detect drifts in their input data distribution and how they learn, memorize, and forget in sequences of tasks. I am also interested in generative models, state representation learning, and continual learning. I obtained my Ph.D. in Computer Science from Institut Polytechnique de Paris (France) under the supervision of David Filliat. My Ph.D. was focused on replay processes for continual learning. Previously, I obtained my engineering diploma in electronics and robotics from CPE Lyon (France).
I previously worked on state representation learning (SLR). It was mostly about unsupervised learning of latent representation for control applications.
Google Scholar Page: Scholar
News
-
06/20: I successfully defended my PhD thesis at Institut Polytechnique de Paris, “Continual Learning: Tackling Catastrophic Forgetting in Deep Neural Networks with Replay Processes” under the supervision of David Filliat (Ensta-Paris) and Andrei Stoian (Thales)
-
12/19: We presented at Reinforcement Learning workshop at NeurIPS 2019 “DisCoRL: Continual Reinforcement Learning via Policy Distillation” with R Traoré, H Caselles-Dupré, T Sun, N Diaz-Rodriguez and D Filliat
-
10/19: Invited talk at Thales Research and Technology, Palaiseau “Continual Deep Learning with Generative Replay”
-
09/19: We present at ICANN 2019 “Training Discriminative Models to Evaluate Generative Ones” with JF Goudou, A Stoian, D Filliat and “Marginal Replay vs Conditional Replay for Continual Learning” (2018) with A Gepperth, A Stoian, D Filliat
-
06/19: We presented at Multi-Task and Lifelong Reinforcement Learning” workshop ICML 2019 our latest work on a “Continual Reinforcement Learning deployed in Real-life using Policy Distillation and Sim2Real Transfer” with Kalifou René Traoré, Hugo Caselles-Dupré, Te Sun, Natalia Diaz-Rodriguez and David Filliat. We also presented our latest work on “Continual Learning of Generative Models with Maximum Entropy Generative Replay” with Cem Sübakan, Massimo Caccia and Laurent Charlin.
-
04/19: We presented at Structure & Priors in Reinforcement Learning” workshop ICLR 2019 our latest work on a “Decoupling feature extraction from policy learning: assessing benefits of state representation learning in goal based robotics” with Antonin Raffin, Ashley Hill, René Traoré, Natalia Díaz-Rodríguez and David Filliat.
-
14th December 2018: Invited Talks at MILA Department, University of Montréal, “Generative Models from the perspective of Continual Learning”
-
12/18: We presented at Deep Reinforcement Learning” workshop NeurIPS 2018 our latest work on a “Toolbox for state representation learning” and we presented at Continual learning” workshop our latest work on “Generative Models and Continual Learning”
Selected Articles
State representation learning for control: An overview (2018)
T Lesort, N Díaz-Rodríguez, JF Goudou, D Filliat, Neural Networks 108, 379-392 pdf
An overview on strategies to extract latent representation from data in particular for reinforcement learning and control purposes.
Generative Models from the perspective of Continual Learning (2018)
T Lesort, H Caselles-Dupré, M Garcia-Ortiz, A Stoian, D Filliat, IJCNN 2019 arXiv
An empirical study on how various type of generative models may help to learn continually on incremental scenarios.
DisCoRL: Continual Reinforcement Learning via Policy Distillation (2019)
R Traoré, H Caselles-Dupré, T Lesort, T Sun, N Diaz-Rodriguez, D Filliat, NeurIPS 2019 Workshop on “Reinforcement Learning” arXiv
An approach to continual reinforcement learning based on distillation and rehearsal.
Continual Learning for Robotics: Definition, Framework, Learning Strategies, Opportunities and Challenges (2019)
T Lesort, V Lomonaco, A Stoian, D Maltoni, D Filliat, N Díaz-Rodríguez, Information Fusion 58, 52-68
arXiv
An overview on continual learning, existing approaches, evaluation metrics/benchmarks with a proposition of framework for continual approaches and description of how robotics environments need continual learning and may exploit it.
All my publications here or on my Scholar.
Posts
From Machine Learning to Continual Learning
Task Label and Continual learning, what is going on??
Continuum: A Data Loader for Continual Learning