AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

B-Cos Alignment for Inherently Interpretable CNNs and Vision Transformers

Moritz Böhle, Navdeeppal Singh, Mario Fritz, Bernt Schiele

TPAMI 2024 · Architectures and efficiency › Attention and transformers · approach: Token merging and masked pretraining · 31 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 MaxOut units to plain B-cos network (ablation)B-cos network (B=2) · CIFAR-10Accuracy (%): 91.5 -> 93.2Improved
increase B from 1.25 to 2.5 (ablation)MaxOut B-cos network · CIFAR-10Accuracy (%): 93.8 -> 92.4Hurt
increase B (alignment pressure) (ablation)B-cos network · CIFAR-10Grid pointing localisation: higher B improves localisationImproved
train B-cos ResNet-152 for 600 epochs with augmentation instead of 90 epochsB-cos ResNet-152 · ImageNet valTop-1 accuracy (%): 76.6 -> 80.2Improved
convert ResNet-152 to B-cos (90 epochs)ResNet-152 · ImageNet val (90 epochs)Top-1 accuracy (%): -1.7Hurt
convert ResNext-50-32x4d to B-cosResNext-50-32x4d · ImageNet val (90 epochs)Top-1 accuracy (%): -2.0Hurt
convert ResNet-50 to B-cosResNet-50 · ImageNet val (90 epochs)Top-1 accuracy (%): -0.2No clear effect
convert ConvNext-Tiny to B-cos (600-epoch protocol)ConvNext-Tiny · ImageNet val (600 epochs, ConvNext-style recipe)Top-1 accuracy (%): -5.0Hurt
remove bias terms from baseline CNNs (as in B-cos) (ablation)ResNet-101 · ImageNet val (90 epochs)Top-1 accuracy (%): -1.1 -> +6.2Improved
convert small ViT-Ti to B-cosViT-Ti · ImageNet val (90 epochs, Simple-ViT protocol)Top-1 accuracy (%): -10.3Hurt
convert large ViT_C-B to B-cosViT_C-B · ImageNet val (90 epochs, Simple-ViT protocol)Top-1 accuracy (%): +0.3No clear effect
add convolutional stem of four layers to B-cos ViT-Ti (ablation)B-cos ViT-Ti · ImageNet val (90 epochs, Simple-ViT protocol)Top-1 accuracy (%): 60.0 -> 67.3Improved

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