Marco Bagatella
PhD student in RL, currently at the Learning and Adaptive Systems group.
bio
I am a PhD student at MPI-IS and ETH Zürich, co-advised by Georg Martius and Andreas Krause. Previously, I was fortunate to spend some time at Meta in Paris, working with Andrea Tirinzoni and Alessandro Lazaric, at the AIT Lab (ETH Zürich), under the supervision of Prof. Otmar Hilliges, and in the Autonomous Learning Group (MPI IS Tübingen). Not so long ago, I obtained a MSc in Computer Science at ETH Zürich, and a few years before that I graduated from Politecnico di Milano (BSc in Engineering of Computing Systems).
research
I am mainly interested in (deep) reinforcement learning, particularly in extracting diverse behavior from offline data and adapting it online. I am very excited about zero-shot methods, and the interplay between multi-task pre-training and test-time adaptation. Outside of RL, I have some experience in representation learning, and I like keeping an eye open towards recent developments with vision/language/action/… models.
news
| Oct 15, 2025 | 🎉 DISCOVER was accepted at NeurIPS 2025. |
|---|---|
| Oct 01, 2025 | ⛰️ I completed a wonderful internship at FAIR (Paris) and I am back in Zurich! |
| May 01, 2025 | 🎉 Active Fine-Tuning of Multi-task Policies and Zero-Shot Offline Imitation Learning via Optimal Transport were accepted at ICML 2025. |
| Aug 01, 2024 | 🎉 Directed Exploration in Reinforcement Learning from Linear Temporal Logic was accepted at EWRL 2024 |
| Jul 22, 2024 | 🪧 Two papers I was involved in were presented at ICML 2024 |
selected publications
-
TD-JEPA: Latent-predictive Representations for Zero-Shot Reinforcement LearningIn arXiv, 2025 -
Optimistic Task Inference for Behavior Foundation ModelsIn arXiv, 2025 -
DISCOVER: Automated Curricula for Sparse-Reward Reinforcement LearningIn Advances in Neural Information Processing Systems, 2025 -
Active Fine-Tuning of Generalist PoliciesIn Forty-second International Conference on Machine Learning, 2025 -
Zero-shot Offline Imitation Learning via Optimal TransportIn Forty-second International Conference on Machine Learning, 2025 -
Goal-conditioned Offline Planning from Curious ExplorationIn Advances in Neural Information Processing Systems, 2023