[KongchangAI]
Conceptreward-proportional sampling

Reward-Proportional Sampling

A sampling objective requiring that the probability of sampling an outcome be proportional to its reward value, thereby preserving the relative weights of all high-reward solutions rather than selecting only the solution with the maximum reward.

Timeline (last 90 days)

Nov 4

The goal of DARS is to achieve reward-proportional sampling in a continuous outcome space, i.e., sampling probability proportional to reward p(x) ∝ r(x), without modifying the standard policy gradient framework.

Unverified50%

All Facts (1)

Unverified

The goal of DARS is to achieve reward-proportional sampling in a continuous outcome space, i.e., sampling probability proportional to reward p(x) ∝ r(x), without modifying the standard policy gradient framework.

50%

Source Articles

Density-Aware Reward Scaling: Making Reinforcement Learning Sampling More BalancedNov 4