Felipe Petroski Such
Papers
3
Total Citations
271
H-Index
3
About
Felipe Petroski Such is a leading researcher at the intersection of evolutionary computation and deep reinforcement learning (RL). His most impactful work focuses on dramatically improving exploration in evolution strategies (ES) for deep RL, demonstrating that ES can match the performance of Q-learning and policy gradient methods while being orders of magnitude faster due to superior parallelization. His seminal 2017 and 2018 papers on novelty-seeking populations in ES have accumulated over 260 citations, establishing a new paradigm for scalable, black-box optimization in complex environments. Such's contributions show that simple evolutionary algorithms, when augmented with diversity mechanisms, can efficiently train deep neural networks for challenging control tasks. He has also explored efficient transfer learning and online adaptation using latent variable models for continuous control, addressing the challenge of generalizing learned dynamics across varying physical parameters. His work bridges the gap between evolutionary algorithms and modern deep RL, offering practical, parallelizable solutions for real-world robotics and simulation tasks. Such's research continues to influence how researchers think about exploration, scalability, and sample efficiency in reinforcement learning.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3