AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Memorizing Long-tail Data Can Help Generalization Through Composition

Mo Zhou, Haoyang Ma, Rong Ge

ICLR 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Decoupled and calibrated long-tail training · 1 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
allow memorization (WD 0 instead of 0.5) (ablation)Sum ResNet-18 · Omniglot-once + MNIST digit-sumtest loss: 2.8744 -> 0.1760Improved
allow memorization (WD 0 instead of 0.5) (ablation)Linear ResNet-18 · Omniglot-once + MNIST digit-sumtest loss: 1.5539 -> 0.1662Improved
allow memorization (WD 0 instead of 0.5) (ablation)2-layer ResNet-18 · Omniglot-once + MNIST digit-sumtest loss: 0.6447 -> 0.1351Improved
allow memorization (WD 0 instead of 0.5) (ablation)Cross-channel ResNet-18 · Omniglot-once + MNIST digit-sumtest loss: 23.6506 -> 22.9671No clear effect
process each digit separately then aggregate, instead of cross-channel ResNet (ablation)ResNet-18 · Heavy-tailed MNIST digit-sum, test with at least one 9test lossImproved
rare-feature frequency alpha=1 (long-tail features appear more often) (ablation)minimum l2-norm linear model · Synthetic linear modelin/out-of-distribution test lossHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy