AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Compositional Attention: Disentangling Search and Retrieval

Sarthak Mittal, Sharath Chandra Raparthy, Irina Rish, Yoshua Bengio, Guillaume Lajoie

ICLR 2022 · Architectures and efficiency › Attention and transformers · approach: Theory of self-attention dynamics · 1 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 rigid search-retrieval head pairings with dynamically composed retrievalsTransformer attention · Sort-of-CLEVRBinary accuracy: 84.4 ± 5.3 -> 89.8 ± 1.1Improved
Replace rigid search-retrieval head pairings with dynamically composed retrievalsTransformer attention · Sort-of-CLEVRTernary accuracy: 64.9 ± 3.3 -> 67.1 ± 1.5Improved
Replace multi-head attention with compositional attentionUniversal Transformer · Multi-Task Image Classification (CIFAR10)Accuracy: 77.2 ± 0.3 -> 78.0 ± 0.1Improved
Replace multi-head attention with compositional attentionUniversal Transformer · Multi-Task Image Classification (FashionMNIST)Accuracy: 89.4 ± 0.0 -> 90.0 ± 0.0Improved
Replace multi-head attention with compositional attentionUniversal Transformer · Multi-Task Image Classification (SVHN)Accuracy: 85.0 ± 0.1 -> 86.0 ± 0.7Improved
Replace multi-head attention with compositional attentionUniversal Transformer · Multi-Task Image Classification (Equilateral Triangle)Accuracy: 96.8 ± 0.1 -> 97.2 ± 0.3Improved
Replace multi-head attention with compositional attentionTransformer · Equilateral Triangle DetectionTest accuracy: 93.8 ± 0.1 -> 97.0 ± 0.3Improved
Replace multi-head attention with compositional attentionTransformer · WikiText-103 testPerplexity: 39.6 ± 0.3 -> 38.8 ± 0.0Improved
Use 1 retrieval instead of 4 (ablation)Compositional Transformer · Sort-of-CLEVRTernary accuracy: 64.8 ± 4.1 -> 66.4 ± 1.3Improved
Use 4 retrievals instead of 3 (ablation)Compositional Transformer · Sort-of-CLEVRTernary accuracy: 67.1 ± 1.5 -> 64.8 ± 4.1Hurt
Use 1 retrieval instead of 4 (ablation)Compositional Transformer · Multi-Task Image Classification (CIFAR10)Accuracy: 77.2 ± 0.3 -> 77.5 ± 0.5No clear effect
Replace dot-product retrieval selection with a linear MLP (ablation)Compositional attention · Sort-of-CLEVR and Equilateral Triangle DetectionPerformanceMixed

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