AI Paper Atlas

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

Learning Bottleneck Concepts in Image Classification

Bowen Wang, Liangzhi Li, Yuta Nakashima, Hajime Nagahara

CVPR 2023 · Label-efficient and robust learning › Recognition and long-tail · approach: Self-explaining and visual explanation networks · 62 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
Use contrastive rather than reconstruction loss for concept self-supervision (ablation)BotCL · CUB200accuracy: 0.693 -> 0.740Improved
Use contrastive rather than reconstruction loss for concept self-supervision (ablation)BotCL · ImageNet (first 200 classes)accuracy: 0.720 -> 0.795Improved
Use reconstruction rather than contrastive loss for concept self-supervision (ablation)BotCL · MNISTaccuracy: 0.980 -> 0.983Improved
Use contrastive rather than reconstruction loss for concept self-supervision (ablation)BotCL · Syntheticaccuracy: 0.785 -> 0.998Improved
Increase the number of concepts (ablation)BotCL · SyntheticaccuracyImproved
Increase the quantization-loss weight (ablation)BotCL · CUB200, ImageNet, and SyntheticaccuracyHurt
Add individual-consistency and mutual-distinctiveness regularizers (ablation)BotCL · CUB200 and ImageNetaccuracyNo clear effect
Increase individual-consistency and mutual-distinctiveness loss weights substantially (ablation)BotCL · All evaluated datasetsaccuracyHurt
Use a larger number of concepts (ablation)BotCL · All datasetsaccuracyMixed

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