Risk-Sensitive Reinforcement Learning
A branch of reinforcement learning that incorporates risk measures—such as entropic risk and CVaR—into the optimization objective. Rather than solely maximizing expected return, it also controls outcome variance and tail risk. This approach is applicable to high-stakes domains including finance, healthcare, and autonomous driving.
Timeline (last 90 days)
The paper bridges a long-standing exponential gap between the upper and lower bounds of sample complexity in risk-sensitive reinforcement learning.
The new analysis eliminates the previously exponential gap between the upper and lower bounds, leaving only a polynomial gap with respect to the effective horizon.
All Facts (2)
The paper bridges a long-standing exponential gap between the upper and lower bounds of sample complexity in risk-sensitive reinforcement learning.
50%UnverifiedThe new analysis eliminates the previously exponential gap between the upper and lower bounds, leaving only a polynomial gap with respect to the effective horizon.
50%