AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

Rethinking Concept Bottleneck Models: From Pitfalls to Solutions

Merve Tapli, Quentin Bouniot, Wolfgang Stammer, Zeynep Akata, Emre Akbas

CVPR 2026 · Vision-language and multimodal › Human and embodied modeling · approach: Concept bottleneck models · 0 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
add knowledge distillation from linear probe teacher (ablation)non-linear CBM · ImageNet100accuracy: 82.27 -> 85.47Improved
add knowledge distillation from linear probe teacher (ablation)non-linear CBM · Places365accuracy: 33.64 -> 35.44Improved
add knowledge distillation (ablation)linear CBM · ImageNet100accuracy: 87.96 -> 90.37Improved
add knowledge distillation (ablation)non-linear CBM · ImageNet100accuracy: 82.27 -> 85.56Improved
add non-linearity (ReLU) to concept encoder (ablation)CBM · ImageNet100accuracy: 87.96 -> 82.27Hurt
replace relevant concepts with random wordsnon-linear CBM · ImageNet100accuracy: drops by roughly 25 pointsHurt
replace relevant concepts with random wordslinear CBM · ImageNet100accuracyNo clear effect
use relevant instead of random concept setSAIL activations · ImageNet100 task-agnosticentropy: 4.565 -> 3.660Improved
use relevant instead of random concept setSAIL activations · Places365 task-specificentropy: 4.578 -> 2.889Improved
use SAIL instead of CLIP for goodness metricconcept activations · Places365 task-agnosticentropy: 4.482 -> 3.077Improved
use stronger backbone Perception Encoder instead of ResNetCBM-Suite with SigLIP · ImageNet100accuracy: 90.12 -> 96.14Improved
use SigLIP instead of SAIL VLMCBM-Suite with Perception Encoder · ImageNet100accuracy: 93.56 -> 96.14Improved

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