Hill Sampling for Test-Time Scaling: A Simple and Better Alternative to Repeated Sampling, Evolution, and Training
Surprisingly, when evaluating a method by the best program it generates, it is found that learning model weights via ES is worse than invoking ES with a learning rate set to zero, i.e., using random weight-space perturbations to search for better models.
Jacob Beck, P. Ogren, Ari Kobren
· 0 citations