AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Modelling Attention with Aitchison Geometry: Token Distinguishability and Temperature Scaling

Sam Hilton-Jones, Timothy J. Norman, Zhanxing Zhu

ICML 2026 · Architectures and efficiency › Attention and transformers · approach: Sparse attention for long-context LLMs · 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
use Aitchison distance instead of KL divergence to compare LongLLaMA-3B with OpenLLaMA-3Battention probability monitoring · LongLLaMA-3B vs OpenLLaMA-3B, memory layer 6direction of uniformity changeImproved
scale logits with log L or (log L)^2 inverse temperature instead of sqrt(log L) or none (ablation)Pythia-1B pretraining on RedPajama sample · Pythia-1B, RedPajama-Data-1T-Sample 1Btraining stability / lossHurt
use higher-order inverse-temperature function of context length (ablation)Pythia-1B attention · Pythia-1B, Row 2048 vs Row 8normalised Aitchison distance gap between long and short rowsImproved
use aggressive training (high learning rate, no weight decay) instead of standard hyperparameters (ablation)Pythia-1B with sqrt(log L) scaling · Pythia-1B, RedPajama sampletraining loss / attention saturationHurt
use small learning rate and high weight decay instead of standard hyperparameters (ablation)Pythia-1B with sqrt(log L) scaling · Pythia-1B, RedPajama sampletraining lossHurt

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