Particle Value Functions
This work addresses risk-sensitive decision-making in reinforcement learning for agents needing to emphasize extreme returns, though it appears incremental with a focus on specific scenarios.
The paper tackles the problem of policy gradients reacting slowly to rare rewards in reinforcement learning by introducing particle value functions, which bound risk-sensitive value functions and show benefits in Cliffworld experiments.
The policy gradients of the expected return objective can react slowly to rare rewards. Yet, in some cases agents may wish to emphasize the low or high returns regardless of their probability. Borrowing from the economics and control literature, we review the risk-sensitive value function that arises from an exponential utility and illustrate its effects on an example. This risk-sensitive value function is not always applicable to reinforcement learning problems, so we introduce the particle value function defined by a particle filter over the distributions of an agent's experience, which bounds the risk-sensitive one. We illustrate the benefit of the policy gradients of this objective in Cliffworld.