<ul><li>Performance predictors are used in neural architecture search (NAS) to reduce evaluation costs by estimating architecture performance.</li><li>Choice of loss functions heavily influences the effectiveness of predictors.</li><li>Recent approaches have explored ranking-based loss functions in addition to traditional regression loss functions.</li><li>A study categorized loss functions into regression, ranking, and weighted types, showing that combining specific categories can enhance predictor-based NAS.</li></ul>

Loss Functions for Predictor-based Neural Architecture Search

Discover more