AI Paper Atlas

Research map / Vision-language and multimodal / Cross-modal learning

Attention Bottlenecks for Multimodal Fusion

Arsha Nagrani, Shan Yang, Anurag Arnab, Aren Jansen, Cordelia Schmid, Chen Sun

NeurIPS 2021 · Vision-language and multimodal › Cross-modal learning · approach: Multimodal fusion with missing modalities · 265 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
Replace late fusion with bottleneck mid-fusionMBT audio-visual classification · Epic-Kitchens-100Top-1 action accuracy: 37.90 -> 43.40Improved
Replace late fusion with bottleneck mid-fusionMBT audio-visual classification · AudioSet minimAP: 41.80 -> 43.92Improved
Replace vanilla cross-attention with attention bottlenecks (ablation)AudioSet fusion model · AudioSet minimAP: over 2 mAPImproved
Use modality-specific encoder weights rather than shared weights (ablation)AudioSet fusion model with early fusion · AudioSet minimAPImproved
Use modality-specific encoder weights rather than shared weights (ablation)AudioSet fusion model with later fusion · AudioSet minimAPNo clear effect
Move fusion from early or late layers to intermediate layers (ablation)AudioSet fusion model · AudioSet minimAPImproved
Vary the number of bottleneck tokens from 4 to other tested sizes (ablation)AudioSet fusion model · AudioSet minimAP: all within 0.5 mAPNo clear effect
Increase input span from 2 seconds to 8 seconds (ablation)Audio-Visual MBT · AudioSet minimAP: 37.95±0.51 -> 42.37±0.44Improved
Use independent MixUp weights for each modality (ablation)Audio-Visual MBT · AudioSetmAP: 42.6 mAP to 43.9 mAPImproved
Sample audio and video windows asynchronously rather than synchronously (ablation)Audio-Visual MBT · AudioSet minimAPNo clear effect
Update bottlenecks sequentially rather than symmetrically (ablation)Audio-Visual MBT · AudioSet minimAP: 43.66±0.26 vs 43.42±0.19; 43.23±0.12No clear effect
Scale backbone from ViT-Base to ViT-Large (ablation)Audio-Visual MBT · AudioSet minimAP: 43.3 -> 42.2Hurt

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