The Black-Box Optimization Lab develops methods for optimizing systems whose inner workings are unknown, using only the outcomes of trying them, along two pillars: evolutionary computation and reinforcement learning.
Evolutionary computation (CMA-ES): design principles and theory / improving and extending CMA-ES / worst-case, risk-aware, and bilevel optimization
Reinforcement learning: bilevel reinforcement learning (designing incentives and environments) / learning from human preferences and alignment / reliable reinforcement learning
New to the field?: Introduce to the evolutionary computation and reinforcement learning topics we focus on, along with examples
Research findings: Our recent research findings, categorized by topic
Learning materials: Learning materials for Evolution Strategies, one of our research topics, focusing on CMA-ES
Software: Code for our research, open source and mostly in Python