AI Paper Atlas

Research map / Learning theory and algorithms / Prediction and decision losses

A Holistic Approach to Unifying Automatic Concept Extraction and Concept Importance Estimation

Thomas FEL, Victor Boutin, Louis Béthune, Remi Cadene, Mazda Moayeri, Léo Andéol, Mathieu Chalvidal, Thomas Serre

NeurIPS 2023 · Learning theory and algorithms › Prediction and decision losses · approach: Provable post-hoc explanations · 9 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
Use PCA rather than KMeans for concept extractionConcept extraction on the penultimate layer · ImageNet, 10 classesRelative ℓ2: 0.60 / 0.54 / 0.73 vs 0.72 / 0.66 / 0.84Improved
Use KMeans rather than PCA for concept extractionConcept extraction on the penultimate layer · ImageNet, 10 classesSparsity: 0.95 / 0.95 / 0.95 vs 0.00 / 0.00 / 0.0Improved
Use KMeans rather than PCA for concept extractionConcept extraction on the penultimate layer · ImageNet, 10 classesStability: 0.07 / 0.08 / 0.04 vs 0.41 / 0.38 / 0.43Improved
Use NMF rather than PCA for concept extractionConcept extraction on the penultimate layer · ImageNet, 10 classesFID: 0.38 / 0.21 / 0.24 vs 0.47 / 0.17 / 0.24Improved
Use NMF rather than KMeans for concept extractionConcept extraction on the penultimate layer · ImageNet, 10 classesOOD: 1.98 / 0.29 / 0.15 vs 1.76 / 0.29 / 0.15Mixed
Decompose concepts in the last layer rather than earlier blocksConcept extraction and concept importance estimation · ImageNet, ResNet50 blocks B2–B5C-Deletion, C-Insertion, and C-μFidelityImproved
Use Gradient Input, Integrated Gradients, Occlusion, or RISE for concept importance in the penultimate layerConcept attribution · Penultimate-layer decompositionC-Deletion, C-Insertion, and C-μFidelityImproved

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