AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Mitigating Over-smoothing in Transformers via Regularized Nonlocal Functionals

Tam Minh Nguyen, Tan Minh Nguyen, Richard Baraniuk

NeurIPS 2023 · Architectures and efficiency › Attention and transformers · approach: Efficient and robust vision transformers · 6 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 the fidelity regularizer to self-attentionDeiT-Tiny · ImageNetTop-1 accuracy: 72.17 -> 73.01Improved
Add the fidelity regularizer to self-attentionDeiT-Tiny · ADE20KSS MIoU: 35.72 -> 37.24Improved
Add the fidelity regularizer to self-attentionSoftmax Transformer · WikiText-103 validationValid PPL: 33.15 -> 32.60Improved
Combine NeuTRENO with FeatScaleDeiT-Tiny · ImageNetTop-1 accuracy: 72.346 -> 73.23Improved
Set λ̃ to 0.1 (ablation)NeuTRENO DeiT · ADE20KSS MIoU: 35.72 -> 34.94Hurt
Set λ̃ to 0.2 (ablation)NeuTRENO DeiT · ADE20KSS MIoU: 35.72 -> 35.60Hurt
Set λ̃ to 0.4 (ablation)NeuTRENO DeiT · ADE20KSS MIoU: 35.72 -> 37.54Improved
Set λ̃ to 0.5 (ablation)NeuTRENO DeiT · ADE20KSS MIoU: 35.72 -> 37.38Improved
Set λ̃ to 0.6 (ablation)NeuTRENO DeiT · ADE20KSS MIoU: 35.72 -> 37.24Improved
Set λ̃ to 0.8 (ablation)NeuTRENO DeiT · ADE20KSS MIoU: 35.72 -> 36.71Improved
Set λ̃ to 1.0 (ablation)NeuTRENO DeiT · ADE20KSS MIoU: 35.72 -> 36.37Improved
Set λ̃ to 2.0 (ablation)NeuTRENO DeiT · ADE20KSS MIoU: 35.72 -> 26.25Hurt

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