AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

$\mu$pscaling small models: Principled warm starts and hyperparameter transfer

Yuxin Ma, Nan Chen, Mateo Diaz Diaz, Soufiane Hayou, Dmitriy Kunisky, Soledad Villar

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: Neural scaling laws

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
upscale a trained narrow model with noise injection and muP hyperparameters instead of training the wide model from scratchMLP, ResNet-18, GPT-2 · Forest Cover Type, CIFAR-100, FineWeb (training loss)training lossImproved
upscale instead of training from scratch, measured on validationResNet-18 · CIFAR-100 validationvalidation performanceHurt
upscale instead of training from scratch, measured on validationMLP and GPT-2 · Forest Cover Type, FineWeb validationvalidation performanceImproved
tune noise and learning rate at small width and transfer, instead of tuning at the target width as in Net2Netupscaling procedure · GPT-2 FineWeb, n/n0=10per-run FLOP cost: 48.2x speedupImproved
inject nonzero symmetry-breaking noise into the widened model versus zero noise (ablation)upscaling procedure · all three settingsinitialization and terminal lossImproved
transfer optimal learning rate and noise level across target widths 128 to 2048 (ablation)upscaled GPT-2 and MLP · GPT-2 AdamW, MLP SGD, MLP AdamWoptimal learning rate and noise levelNo clear effect
use zero noise with muP scaling (ablation)widened model · architectures and optimizers in appendixtraining trajectoryNo clear effect

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