AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

A Capacity-Based Rationale for Multi-Head Attention

Micah Adler

ICML 2026 · Architectures and efficiency › Attention and transformers · approach: Theory of self-attention dynamics · 0 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
split fixed total key dimension across more heads (multi-head instead of single head) (ablation)max-over-heads attention on permutation graphs · permutation graphs m=512 d_model=64 / m=256 d_model=32micro-F1 reaching 0.99Improved
use multiple heads in a full Transformer block with frozen GPT-2 embeddings (ablation)single-layer Transformer, m=6144 · permutation-matching retrievaltest accuracy: h=1 saturates near 0.81, with some seeds collapsing to 0Improved
use many heads (h=8) at very small D_K=64 (d_k=8) (ablation)single-layer Transformer, m=6144 · permutation-matching retrievaltest accuracyHurt
add multiple heads under message retrieval with value channel (d_model=16, h>=4 vs h=1) (ablation)value-channel attention, m=64 · message retrievaltest MSEImproved
relax compression to d_model=32 (multi-head benefit shrinks) (ablation)value-channel attention, m=64 · message retrievaltest MSENo clear effect
add softmax with sqrt(d_k) scaling and summed heads (ablation)idealized attention model · permutation graphsD_K needed for F1 0.99Hurt
increase graph density from permutation (r=1) to r-regular graphs (ablation)max-over-heads attention · r-regular graphs m=128 d_model=32D_K* threshold: D_K* ~ 0.46 x with x = (r m / d_model) log2 mHurt

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