Kevin Frans

Papers

Total Citations

119

H-Index

About

Kevin Frans is a leading researcher in meta-learning and hierarchical reinforcement learning, whose work has significantly advanced how AI systems can learn to learn. His most influential contribution, "Meta Learning Shared Hierarchies" (2017, 117 citations), introduced a novel approach for learning hierarchically structured policies that dramatically improve sample efficiency on unseen tasks. By developing shared primitives—policies executed over extended timesteps—Frans created a framework that allows agents to rapidly adapt to new challenges by reusing learned sub-skills. This work has become foundational in the field of hierarchical meta-learning. More recently, in "Population-Based Evolution Optimizes a Meta-Learning Objective" (2021), Frans explored how evolutionary strategies can discover powerful meta-learners without the computational expense of traditional inner-outer loop optimization. His research sits at the intersection of meta-learning, reinforcement learning, and evolutionary computation, addressing fundamental challenges in creating AI systems that can generalize efficiently across diverse tasks. Frans's work continues to inspire new approaches in few-shot learning and adaptive robotics.