cv
Curriculum vitae. For contact, please use the social links on the homepage.
Contact Information
| Name | Daniil Tiapkin |
| Professional Title | Research Scientist, Google DeepMind |
| Website | https://d-tiapkin.github.io |
Professional Summary
Research Scientist at Google DeepMind in Paris, working on foundation-model post-training.
I received my PhD in Applied Mathematics & Computer Science from École Polytechnique and Université Paris-Saclay (2025), advised by Éric Moulines and Gilles Stoltz.
My research interests include reinforcement learning, post-training of foundation models, and the connections between amortized sampling and RL.
Experience
-
2026 - Paris, France
Research Scientist
Google DeepMind
Research on post-training of foundation models.
-
2024 - 2024 Paris, France
Student Researcher
Google DeepMind
Research on language-model distillation, supervised by Mathieu Blondel.
- On Teacher Hacking in Language Model Distillation — ICML 2025.
-
2023 - 2026 Paris, France
PhD Candidate
École Polytechnique (CMAP) — Institut Polytechnique de Paris & LMO, Université Paris-Saclay
Advised by Éric Moulines and Gilles Stoltz. Thesis: Sample-Efficient Reinforcement Learning: Exploration, Imitation, and Online Learning.
Education
-
2023 - 2025 Paris, France
PhD in Applied Mathematics & Computer Science
École Polytechnique & Université Paris-Saclay
- Thesis: Sample-Efficient Reinforcement Learning: Exploration, Imitation, and Online Learning
- Advisors: Éric Moulines (CMAP, École Polytechnique) and Gilles Stoltz (LMO, Université Paris-Saclay)
-
2021 - 2023 Moscow, Russia
MSc in Applied Mathematics & Computer Science
HSE University (Faculty of Computer Science)
- Program: Math of Machine Learning
-
2017 - 2021 Moscow, Russia
Awards
-
2024 Oral presentation, AISTATS 2024
AISTATS
Generative Flow Networks as Entropy-Regularized RL selected for oral presentation.
-
2022 Oral presentation, ICML 2022
ICML
From Dirichlet to Rubin: Optimistic Exploration in RL without Bonuses selected for oral presentation.
Selected Publications
For the full and up-to-date list, see the publications page.
-
2026 Beyond Softmax and Entropy: Convergence Rates of Policy Gradients with f-SoftArgmax Parameterization & Coupled Regularization
ICLR 2026
S. Labbi, D. Tiapkin, P. Mangold, É. Moulines.
-
2025 On Teacher Hacking in Language Model Distillation
ICML 2025
D. Tiapkin, D. Calandriello, J. Ferret, S. Perrin, N. Vieillard, A. Ramé, M. Blondel.
-
2024 Demonstration-Regularized RL
ICLR 2024
D. Tiapkin, D. Belomestny, D. Calandriello, É. Moulines, A. Naumov, P. Perrault, M. Valko, P. Ménard.
-
2024 Generative Flow Networks as Entropy-Regularized RL
AISTATS 2024 (Oral)
D. Tiapkin⋆, N. Morozov⋆, A. Naumov, D. Vetrov. (⋆ equal contribution)
-
2022 From Dirichlet to Rubin: Optimistic Exploration in RL without Bonuses
ICML 2022 (Oral)
D. Tiapkin, D. Belomestny, É. Moulines, A. Naumov, S. Samsonov, Y. Tang, M. Valko, P. Ménard.
Skills
Languages
Projects
-
gfnx
Fast and scalable JAX library for Generative Flow Networks.
- GitHub: github.com/d-tiapkin/gfnx
- PyPI: pypi.org/project/gfnx
- Docs: gfnx.readthedocs.io
- Paper: arXiv:2511.16592