Michal Valko

Roles & Affiliations: Michal Valko is the Chief Models Officer at a stealth startup, a tenured researcher at Inria (Inria Lille - Nord Europe), and a lecturer in the MVA Master program at École Normale Supérieure Paris-Saclay. He previously held roles including Principal Llama Engineer at Meta and Senior Staff Research Scientist at Google DeepMind.

Education: Ph.D. in Machine Learning from the University of Pittsburgh (2011), Habilitation à Diriger des Recherches (HdR) from École Normale Supérieure Paris-Saclay (2016).

Research Interests: Michal focuses on minimizing human supervision in AI systems, including deep reinforcement learning, bandit algorithms, self-supervised learning, and large language models (LLMs). His work emphasizes scalable test-time inference, fine-tuning, and alignment of LLMs. He also explores theoretical foundations of reinforcement learning and representation learning.

Awards: Recipient of the Outstanding Paper Award at ICML 2023, Inria's Prime d'Excellence Scientifique (2014–2017 and 2018–2021), Homer Warner Best Paper Award (2010), and Google Best Paper Award (2010).

Advising & Grants: Advised over 20 PhD and postdoctoral students, including notable alumni like Daniele Calandriello and Jean Tarbouriech. His research has been supported by grants from ANR (Extra-Learn, BOLD), Intel, and collaborative projects like EduBand with Carnegie Mellon.

Labs & Teams: Co-founder of the SequeL team at Inria and a key contributor to the DeepMind Paris and Meta GenAI research groups. Active in organizing conferences like ICML, NeurIPS, and workshops on Graphs in Machine Learning.