ConceptLong-tail Recognition / 长尾分布识别
Long-Tailed Recognition
A task paradigm in machine learning for scenarios with highly imbalanced class distributions, where minority classes have very few samples and majority classes are sample-rich. The core challenge is learning discriminative features and maintaining overall performance despite limited samples in rare classes.
Timeline (last 90 days)
Nov 2
Long-tailed recognition typically adopts a two-stage training strategy: the first stage learns a shared feature encoder on the original imbalanced data, and the second stage freezes the encoder and balances the classifier through resampling, reweighting, or classifier correction (such as logit adjustment and τ-normalization).
Unverified50%