AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

TabSwift: An Efficient Tabular Foundation Model with Row-Wise Attention

Si-Yang Liu, Han-Jia Ye

ICML 2026 · Architectures and efficiency › Attention and transformers · approach: Long-context KV and memory attention · 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 learned exit head with register conditioning instead of entropy stopping at about 14-15 layers (ablation)TabSwift early exit · TALENT classificationaverage accuracy: 0.8370 at 14.06 layers vs 0.8367 at 15.27 layersImproved
use register-conditioned exit head instead of one without registers at tau 0.999 (ablation)TabSwift early exit · TALENT classificationaverage accuracy: 0.8390 -> 0.8398Improved
early exit with threshold 0.5 instead of full depthTabSwift · TALENT classificationaverage accuracy: 0.8398 -> 0.8334No clear effect
early exit with learned head w/ registers tau 0.9 instead of full depthTabSwift · TALENT classificationaverage accuracy: 0.8398 -> 0.8329No clear effect
tighten learned exit threshold to 0.999 with registersTabSwift early exit · TALENT classificationaverage accuracy: 0.8398 -> 0.8398No clear effect
evaluate on datasets with more than 20,000 training rowsTabSwift · TALENT classificationaverage rank: 12.32 vs RealMLP 9.31Hurt
compare with TabPFN v2 on high-dimensional datasets with PCA to 100 dimsTabSwift · High-dimensional datasetsaverage rank: 2.1 vs 1.8Mixed
TabSwift vs TabICL on high-dimensional datasetsTabSwift · High-dimensional datasetsaverage rank: 2.9 -> 2.1Improved

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