Adaptive Learn-then-Test: Statistically Valid and Efficient Hyperparameter Selection
This addresses the need for efficient and statistically valid hyperparameter selection in AI, particularly in costly or safety-critical testing scenarios, though it is incremental as it builds on the learn-then-test technique.
The paper tackles the problem of hyperparameter selection by introducing adaptive learn-then-test (aLTT), which reduces testing rounds while maintaining statistical guarantees, achieving the same performance as existing methods with only a fraction of the rounds in applications like offline reinforcement learning and prompt engineering.
We introduce adaptive learn-then-test (aLTT), an efficient hyperparameter selection procedure that provides finite-sample statistical guarantees on the population risk of AI models. Unlike the existing learn-then-test (LTT) technique, which relies on conventional p-value-based multiple hypothesis testing (MHT), aLTT implements sequential data-dependent MHT with early termination by leveraging e-processes. As a result, aLTT can reduce the number of testing rounds, making it particularly well-suited for scenarios in which testing is costly or presents safety risks. Apart from maintaining statistical validity, in applications such as online policy selection for offline reinforcement learning and prompt engineering, aLTT is shown to achieve the same performance as LTT while requiring only a fraction of the testing rounds.