reweighting

Definition

Assigning new weights to examples or to terms in a loss. It is how you correct a mismatch between the training and target distributions (importance weighting), and how you emphasise rare classes or hard cases without collecting more data. The practical hazard is variance: if a few examples receive very large weights, the effective sample size collapses and training destabilises.

Explainers using this term

• Paper Walkthrough: SA-MRPO — Stop Studying the Subject You've Already AcedLearn What's Left, Not What's Mastered: Saturation Aware Advantage Reweighting for Multi-Reward Policy Optimization