regret bound

A regret bound quantifies the maximum expected loss an online learning algorithm incurs compared to the best fixed strategy in hindsight, providing benchmarks for evaluating learning performance.

14 papers