Riccardo De Santi

Timeline

1. 2022

   An ICML Outstanding Paper

   In 2022, he coauthors The Importance of Non-Markovianity in Maximum State Entropy Exploration with Mirco Mutti and Marcello Restelli. The paper shows that non-Markovian deterministic policies suffice for a single-trial entropy objective while Markovian policies suffer non-zero regret, and it receives an Outstanding Paper Award at ICML 2022.

2. December 2023

   PhD at the ETH AI Center

   In December 2023, he starts a PhD at the ETH AI Center, advised by Andreas Krause, Niao He, and Kjell Jorner. As an ETH AI Center Doctoral Fellow in the Learning & Adaptive Systems Group at ETH Zurich, he works on generative optimization and exploration for large-scale scientific discovery.

3. 2024

   Geometric active exploration

   At ICML 2024, he publishes Geometric Active Exploration in Markov Decision Processes: the Benefit of Abstraction with Federico Arangath Joseph, Noah Liniger, Mirco Mutti, and Andreas Krause; it links active exploration with MDP homomorphisms to exploit structures such as permutation invariance in molecular design. He also has a second ICML 2024 paper, Global Reinforcement Learning: Beyond Linear and Convex Rewards via Submodular Semi-gradient Methods.

4. 2025

   Exploration with diffusion models

   In 2025, he publishes Provable Maximum Entropy Manifold Exploration via Diffusion Models at ICML, which casts exploration as entropy maximization over the data manifold defined by a pre-trained diffusion model. Flow Density Control, an algorithm that reduces fine-tuning of flow and diffusion models for objectives beyond average reward to a sequence of simpler fine-tuning tasks, is accepted as a spotlight at NeurIPS 2025.

5. October 2025

   Teaching diffusion models at ETH

   In October 2025, he co-designs with Andreas Krause the new Diffusion Generative Models module of ETH Zurich's Probabilistic AI course, which has about 500 students.

6. 2026

   Six papers on flows for discovery

   In 2026, three of his works on flows for discovery are accepted at ICLR, including Verifier-Constrained Flow Expansion for Discovery Beyond the Data, and three more at ICML, including A Unified Density Operator View of Flow Control and Merging.

7. 2026

   A Rising Star in Data Science

   In 2026, UChicago, Stanford, Harvard, and UCSD select him as a Rising Star in Data Science.

8. As of October 2026

   PhD student at ETH Zurich

   As of October 2026, he is a PhD student at ETH Zurich. After visiting Caltech, he continues working with Yisong Yue's group and Frances H. Arnold's lab to close the loop between generative discovery and chemical wet-lab validation. In October 2026, he introduces LLM-to-Flow Distillation, a method that steers scientific generators with natural-language discovery goals by distilling the knowledge of frontier LLMs.