AI Paper Atlas

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

Disentangling Sampling and Labeling Bias for Learning in Large-output Spaces

Ankit Singh Rawat, Aditya K Menon, Wittawat Jitkrittum, Sadeep Jayasumana, Felix Yu, Sashank Reddi, Sanjiv Kumar

ICML 2021 · Label-efficient and robust learning › Recognition and long-tail · approach: Learning from unlabeled and positive-unlabeled data · 3 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
Use within-batch negatives with constant weightingsoftmax cross-entropy · CIFAR-100-LTbalanced error on Tail classesImproved
Use within-batch negatives with constant weightingsoftmax cross-entropy · long-tail benchmarksbalanced error on Tail classesImproved
Use within-batch negatives with relative weighting instead of constant weighting (ablation)softmax cross-entropy · long-tail benchmarksbalanced error on Head classesImproved
Use the proposed tail weightinguniform and within-batch negative sampling · long-tail benchmarksbalanced error on Tail classesImproved
Replace the full-label logit-adjustment loss with tail-weighted negative samplingsoftmax cross-entropy · long-tail benchmarksbalanced error on Tail classesHurt
Use within-batch negativessoftmax cross-entropy · ImageNet-LTbalanced error on Tail classesImproved
Use uniform negatives instead of within-batch negatives (ablation)softmax cross-entropy · retrieval benchmarksRecall@k on TailHurt
Use within-batch negatives instead of uniform negatives (ablation)softmax cross-entropy · retrieval benchmarksRecall@k on HeadHurt
Use uniform negatives instead of within-batch negatives (ablation)softmax cross-entropy · retrieval benchmarksRecall@k on HeadImproved
Use negative sampling with contrastive loss instead of the standard baseline (ablation)contrastive loss · long-tail benchmarksbalanced error on Tail classesHurt
Reduce the number of sampled negatives (ablation)softmax cross-entropy · ImageNet-LTbalanced errorHurt

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