AI Paper Atlas

Research map / Learning theory and algorithms / Kernels and regression theory

Preconditioning Neural Tangent Kernel for Adaptive Optimization

Xiyuan Yang, Wenxuan Bao, Katherine Tieu, Jingrui He

ICML 2026 · Learning theory and algorithms › Kernels and regression theory · approach: Neural tangent kernel analysis

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
add Adam-style preconditioner estimation to NTK (OAK vs NTK)RoBERTa NTK · MNLI/MRPC/QQP avgaccuracy: 47.38 -> 52.91Improved
add Adam-style preconditioner estimation to NTK (OAK vs NTK)Qwen2.5-0.5B NTK · MNLI/MRPC/QQP avgaccuracy: 46.30 -> 48.28Improved
replace SignGD heuristic with estimated preconditionerRoBERTa SignGD-NTK · MNLI/MRPC/QQP avgaccuracy: 52.20 -> 52.91Improved
replace SignGD heuristic with estimated preconditionerQwen2.5-0.5B SignGD-NTK · SST-2/CoLA avgaccuracy: 52.39 -> 54.27Improved
replace NTK with OAKQwen2.5-0.5B · SST-2/CoLA avgaccuracy: 54.13 -> 54.27No clear effect
add variance (AdaGrad) component to momentum-only OAK, i.e. full Adam (ablation)OAK (SGDM) · MNLI n=64accuracy: 35.85 -> 37.66Improved
add momentum component to variance-only OAK, i.e. full Adam (ablation)OAK (AdaGrad) · MNLI n=64accuracy: 37.34 -> 37.66Improved
use AdamW instead of Adam preconditionerOAK · MNLI n=64accuracy: 37.66 -> 37.89No clear effect
use OAK-selected 50% subset instead of full dataRoBERTa finetuning · MNLI n=64accuracy: 35.5 -> 35.3No clear effect
use OAK instead of NTK for data selectionRoBERTa finetuning · MRPC n=64 subsetaccuracy: 71.3 -> 72.8Improved
use OAK in low-data regime instead of finetuningRoBERTa · SST-2 n=8accuracy: 52.28 -> 55.60Improved
increase finetuning samples n (proxy for training steps)OAK vs finetuning gap · SST-2 Qwen2.5-0.5Baccuracy: 55.14 -> 57.69Hurt

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