AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Scan and Snap: Understanding Training Dynamics and Token Composition in 1-layer Transformer

Yuandong Tian, Yiping Wang, Beidi Chen, Simon Shaolei Du

NeurIPS 2023 · Architectures and efficiency › Attention and transformers · approach: Attention head and block redesign · 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
Attend to distinct contextual tokens rather than common tokens1-layer Transformer self-attention · sequence-class settingattention weight on distinct tokensImproved
Suppress a common contextual token1-layer Transformer self-attention · sequence-class setting with one common tokenattention weight on common tokenHurt
Increase the decoder learning rate eta_Y while holding eta_Z fixed (ablation)1-layer Transformer · Syn-Smallattention sparsityHurt
Increase the self-attention learning rate eta_Z (ablation)1-layer Transformer · Syn-Mediumdistinct-token attention entropyImproved
Increase eta_Y while holding eta_Z fixed (ablation)1-layer Transformer · Syn-Mediumdistinct-token attention entropy: eta_Y/eta_Z\approx 2Mixed
Use Adam instead of SGD (ablation)1-layer Transformer · Syn-Smalldistinct-token selectionMixed
Use a larger decoder learning rate eta_Y (ablation)1-layer Transformer · training dynamicsphase-transition time: t_0 \approx 2K\ln M/\eta_YImproved
Scale up both learning rates eta_Y and eta_Z (ablation)1-layer Transformer · training dynamicsattention sparsityImproved
Decrease eta_Y while holding eta_Z fixed (ablation)1-layer Transformer · after phase transitionattention sparsityImproved
Train on WikiText with SGD1-layer and multi-layer Transformers · WikiTextfirst-layer attention sparsityImproved

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