Inductive Biases in Reinforcement Learning

Inductive biases encode prior knowledge about the world and play a crucial role in shaping the learning process in reinforcement learning (RL) agents. They allow for incorporating assumptions and steering the learning algorithm toward the most plausible solutions. Embedding proper inductive biases can dramatically improve sample efficiency and generalization.

In the IBRL workshop, we investigate the role of inductive biases in modern RL methods, analyzing their impact from various perspectives. We assess the limitations of current methods and explore novel designs toward more robust, general, and adaptable RL agents. Topics include abstractions and structured policies, generalization, relational biases and representations, learning biases for robotics, and future directions.