# Successive Halving

> Source: https://sukruyusufkaya.com/en/glossary/successive-halving
> Updated: 2026-05-13T20:59:24.602Z
> Type: glossary
> Category: makine-ogrenmesi
**TLDR:** An optimization approach that evaluates model candidates in stages, eliminates weak ones, and allocates resources to the strongest candidates.

<p>Successive Halving is used to reduce the cost of full hyperparameter evaluation. Many candidates are started with a small budget, weaker ones are removed, and the remaining strong candidates continue with more resources. This can save significant time for models with long training cycles. However, one must consider carefully whether early-stage performance reliably predicts final performance.</p>