Conceptsampling regret
Sampling Regret
A performance measure for the tilted sampling problem, defined as the temperature-normalized cumulative KL divergence between the learner's distribution and the Gibbs target distribution. Its uncertainty structure is dominated by the squared confidence width, and regret guarantees are attained at the logarithmic level.
Timeline (last 90 days)
Oct 16
Sampling regret is defined as the temperature-normalized cumulative KL divergence between the learner's distribution and the Gibbs target distribution, corresponding to the suboptimality gap in entropy-regularized reward maximization.
Unverified50%