AI Paper Atlas

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

Constraining Representations Yields Models That Know What They Don't Know

Joao Monteiro, Pau Rodriguez, Pierre-Andre Noel, Issam H. Laradji, David Vazquez

ICLR 2023 · Label-efficient and robust learning › Recognition and long-tail · approach: Loss design and classifier theory · 0 citations (OpenAlex)

Read the paper

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
Add TAC to frozen RoBERTa-BASERoBERTa-BASE · HWU64, Banking77, and CLINC150accuracy (%): 92.6 -> 93.9; 93.9 -> 94.2; 97.0 -> 97.4Improved
Use TAC (L1) distances to detect prediction errorsRoBERTa-BASE · HWU64Det. Rate (%): 81.43 -> 83.25Improved
Use TAC (L1) distances to detect prediction errorsRoBERTa-BASE · CLINC150Det. AUROC (%): 91.36 -> 93.25Improved
Use TAC (Cosine) scores to detect prediction errorsViT Base-16x16 · ImageNetDet. AUROC (%): 80.91 -> 89.41Improved
Use TAC (L1) distances to flag examples from unseen classesRoBERTa-BASE · CLINC150 UNKNOWN classesDet. Rate (%): 91.78 -> 92.82Improved
Train TAC from scratch rather than use the base classifier4-layer CNN · MNISTClean accuracy (%): 99.0 -> 98.6Hurt
Train TAC from scratch rather than use the base classifierWideResNet-28-10 · CIFAR-10Det. AUROC (%): 93.7 -> 95.7Improved
Use TAC from scratch rather than the base classifierWideResNet-28-10 · CIFAR-10Clean accuracy (%): 95.6 -> 94.9Hurt
Use a last-layer ECOC ensemble instead of cross-depth TAC (ablation)RoBERTa-BASE · HWU64 and CLINC150accuracy (%): 93.9 -> 73.6; 97.4 -> 83.4Hurt
Add LCE to the last-layer ECOC approach (ablation)RoBERTa-BASE · HWU64 and CLINC150accuracy (%): 93.9 -> 85.2; 97.4 -> 87.8Hurt
Use a last-layer ECOC ensemble instead of cross-depth TAC (ablation)ViT · ImageNetDet. AUROC (%): 89.4 -> 90.06Improved
Use a last-layer ECOC ensemble instead of cross-depth TAC (ablation)ViT · ImageNetaccuracy (%): 80.6 -> 74.4Hurt

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