warm

Definition

Signals that something resumes from an existing state rather than from nothing. A warm restart resets the learning rate while keeping the weights; a warm start initialises training from pretrained weights instead of random ones; a warm cache already holds useful entries. The opposite is cold start — no history to work from, the standard framing for a recommender facing a brand-new user.