AI Paper Atlas

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

Editing a classifier by rewriting its prediction rules

Shibani Santurkar, Dimitris Tsipras, Mahalaxmi Elango, David Bau, Antonio Torralba, Aleksander Madry

NeurIPS 2021 · Label-efficient and robust learning › Recognition and long-tail · approach: Learning from unlabeled and positive-unlabeled data · 4 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
edit with single synthetic snow-to-road exemplar instead of fine-tuningVGG16 · Vehicles on snowy roads (Flickr)error rateImproved
snow-to-road edit side effect on other snow imagesVGG16 · ImageNet test images containing snowaccuracy: 92.27 -> 91.05No clear effect
edit iPod text to blank vs no editCLIP-ResNet-50 · Typographic attacks (6 objects)errors correctedImproved
global fine-tuning instead of editingCLIP-ResNet-50 · Typographic attacksaccuracy on clean iPod imagesHurt
editing vs fine-tuning on non-target classesVGG16 / ResNet · Synthetic concept-transformation benchmark% errors corrected: 3.12 -> 32.89Improved
fine-tuning (local or global) on non-target classesVGG16 · Synthetic concept-transformation benchmark% errors correctedHurt
apply editing constraints to the entire image (-mask) (ablation)Editing · Synthetic benchmark, ImageNet VGG16misclassifications correctedImproved
remove rank-one restriction of the update (-proj) (ablation)Editing · Synthetic benchmarkmisclassifications correctedHurt
edit deeper layers (ablation)Editing · Synthetic benchmark, VGG16misclassifications correctedImproved
fine-tune deeper layers (local) (ablation)Local fine-tuning · Synthetic benchmark, VGG16non-target class errors correctedHurt
increase number of exemplars (3 to 10) (ablation)Editing · Synthetic benchmark, VGG16misclassifications correctedImproved
increase number of exemplars (ablation)Fine-tuning · Synthetic benchmark, VGG16misclassifications correctedMixed

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