AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Predicting Large Model Test Losses with a Noisy Quadratic System

Chuning Li, Chris J. Maddison

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: Neural scaling laws · 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
replace Chinchilla loss model with NQSloss predictor · Pythia + OWT2 holdout IsoFLOPs, x1024 gapHuber loss x1e-5: 9.0 -> 2.5Improved
replace Chinchilla loss model with NQSloss predictor · Pythia + OWT2 holdout B-K plane, x1024 gapHuber loss x1e-5: 9.8 -> 5.6Improved
replace Chinchilla loss model with NQSloss predictor · Llama + LM1B holdout B-K plane, x6 gapHuber loss x1e-5: 8.7 -> 8.2Improved
replace Chinchilla loss model with NQS on training dataloss predictor · Pythia + OWT2 train IsoFLOPsHuber loss x1e-5: 0.8 -> 2.8Hurt
add LayerNorm effective learning rate adjustment (gamma ~ 1/||w||^2) (ablation)NQS · small batch sizes, large modelsloss fitImproved
add LayerNorm adjustment (ablation)NQS · large batch sizesloss fitNo clear effect
increase extrapolation compute gap to x100 and beyondChinchilla · Pythia + OWT2extrapolation errorHurt
increase extrapolation gap up to 1/4000 computeNQS · Pythia + OWT2extrapolation errorNo clear effect
mix critical and non-critical batch sizes in Chinchilla training data (ablation)Chinchilla · Pythia + OWT2prediction accuracyHurt
use NQS to choose N, B, K under compound constraintsNQS · compute plus time/memory/data constraintsoptimality of chosen configImproved

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