AI Paper Atlas

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

Measures of Information Reflect Memorization Patterns

Rachit Bansal, Danish Pruthi, Yonatan Belinkov

NeurIPS 2022 · Label-efficient and robust learning › Recognition and long-tail · approach: Contrastive representation theory · 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
increase heuristic memorization strength alpha (up to 1.0)MLP on Colored MNIST, DistilBERT on Sentiment Adjectives · synthetic spurious correlationneuron entropyHurt
increase heuristic memorization strength alphaMLP and DistilBERT · Colored MNIST, Sentiment Adjectivesneuron-pair mutual informationHurt
train on original raw Bias-in-Bios set instead of balanced or scrubbed setsRoBERTa-base · Bias-in-BiosentropyHurt
de-biasing training sets (balanced, scrubbed)RoBERTa-base · Bias-in-Biosentropy vs compressionNo clear effect
balanced instead of unbalanced common contexts in trainingResNet-18 · NICO++entropy: 35.03% -> 50.08%Improved
unbalanced instead of balanced contexts in trainingResNet-18 · NICO++neuron-pair mutual informationHurt
increase label shuffling fraction beta3-layer MLP on MNIST · Shuffled MNISTentropyImproved
increase label shuffling fraction betaDistilBERT · Shuffled IMDbentropyMixed
increase label shuffling fraction betaMLP and DistilBERT · Shuffled MNIST, Shuffled IMDbneuron-pair mutual informationHurt
increase label shuffling fraction betaMLP and DistilBERT · Shuffled MNIST, Shuffled IMDbvalidation accuracyHurt
use mean entropy for model selection rankinginformation measures · Bias-in-Bios TPR GapKendall tau: 0.20No clear effect
use mean MI for model selection rankinginformation measures · Colored MNISTKendall tau: 1.00Improved

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