AI Paper Atlas

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

Robust Concept Erasure via Kernelized Rate-Distortion Maximization

Somnath Basu Roy Chowdhury, Nicholas Monath, Kumar Avinava Dubey, Amr Ahmed, Snigdha Chaturvedi

NeurIPS 2023 · Label-efficient and robust learning › Recognition and long-tail · approach: Multilingual language representation pretraining

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
Apply KRaM concept erasureBERT representations · DIAL race probingAcc. (a): 87.7 -> 54.0Improved
Apply KRaM concept erasureGloVe embeddings · gender probingAcc. (a): 100.0 -> 52.6Improved
Apply KRaM concept erasureBERT representations · DIAL sentiment taskAcc. (y): 75.5 -> 72.4Hurt
Apply KRaM concept erasureUCI Crimes representations · UCI Crimesdelta GDP: 0.058 -> 0.001Improved
Use a linear erasure function instead of non-linear (ablation)KRaM · Syntheticalignment Ak: 0.67 -> 0.75Improved
Use a linear erasure function instead of non-linear (ablation)KRaM · GloVe genderAcc. (a): 52.6 -> 67.0Hurt
Swap kernel from Gaussian to Laplace (ablation)KRaM · SyntheticMSE (a): 0.109 -> 0.083Hurt
Swap kernel from Gaussian to Cauchy (ablation)KRaM · SyntheticMSE (a): 0.109 -> 0.092Hurt
Increase lambda penalty on |R(Z)-b| (ablation)KRaM · GloVe gendergender accuracyHurt
Increase scaling factor r in R(Z)=rb (ablation)KRaM · GloVe genderAk and gender accuracyMixed
Vary Gaussian kernel sigma (ablation)KRaM · SyntheticMSE (a)Mixed
Replace rate-distortion objective with MI estimators (InfoNCE, MINE, CLUB, KNIFE) (ablation)KRaM · GloVe genderAk: 0.65 -> 0.53Hurt

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