AI Paper Atlas

Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning

Generalized Logit Adjustment: Calibrating Fine-tuned Models by Removing Label Bias in Foundation Models

Beier Zhu, Kaihua Tang, Qianru Sun, Hanwang Zhang

NeurIPS 2023 · Label-efficient and robust learning › Zero-/few-shot and meta-learning · approach: Robust fine-tuning of zero-shot models · 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
apply GLA on top of E2E/WiSE-FTCLIP ViT-B/16 E2E · ImageNetaccuracy: 81.3 -> 82.8Improved
GLA instead of WiSE-FT ensemblingCLIP ViT-B/16 · ImageNetaccuracy: 81.7 -> 82.8Improved
WiSE-FT ensembling without bias handlingE2E · ImageNet tail classesaccuracy: 77.3 -> 76.4Hurt
GLA on tail classesE2E · ImageNet tail classesaccuracy: 77.3 -> 78.8Improved
remove label bias from zero-shot model (ablation)CLIP ViT-B/16 zero-shot · ImageNetaccuracy: 68.8 -> 69.3Improved
transfer pi_p from ViT-B/32 to ViT-L/14 (ablation)zero-shot CLIP ViT-L/14 · ImageNetaccuracy: 75.6 -> 76.3Improved
GLA vs prompt tuning baselines under shiftprompt tuning · ImageNet-Aaccuracy: 21.3 -> 23.0Improved
prompt tuning baseline vs zero-shot CLIP under shiftCoOp · ImageNet-Sketchaccuracy: 35.5 -> 32.5Hurt
debias zero-shot using estimated pi_p (ablation)CLIP-RN50 zero-shot · 11 datasets averageaccuracy: 59.1 -> 60.7Improved
naive ensembling instead of GLA ensemble (ablation)fine-tuned models · EuroSAT, SUN, Caltech, UCF, AircraftaccuracyHurt
GLA instead of BALLADE2E CLIP-RN50 · ImageNet-LTaccuracy: 67.2 -> 68.2Improved
GLA instead of balanced softmaxprompt tuning · ImageNet-LTaccuracy: 63.5 -> 67.8Improved

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