AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Escaping the Subspace Trap: The Role of Optimizer Geometry in Model Width Expansion

Jiabei Chen, Haoyu Wang, Yang Yu, Yao Xu, Liangdong Wang, Guang Liu, Shizhu He, Jun Zhao, Kang Liu

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: Layer-wise analysis of language models

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 Muon instead of AdamWQwen3-2.38B symmetric expansion · Downstream averageaverage accuracy: 46.67 -> 47.88Improved
use Muon instead of AdamWQwen3-2.38B diagonal expansion · Downstream averageaverage accuracy: 45.96 -> 47.15Improved
use Muon instead of AdamWSmolLM2-540M symmetric expansion · Downstream averageaverage accuracy: 33.74 -> 34.65Improved
use Muon instead of AdamWSmolLM2-540M diagonal expansion · Downstream averageaverage accuracy: 34.14 -> 34.36No clear effect
use Muon instead of AdamWSmolLM2-540M diagonal expansion · MMLUaccuracy: 26.80 -> 26.54Hurt
use Muon instead of AdamWQwen3-2.38B symmetric expansion · Big Bench Hardaccuracy: 41.05 -> 44.43Improved
use Muon instead of AdamWQwen3-2.38B symmetric expansion · AGIEval-enaccuracy: 26.30 -> 26.27No clear effect
use Muon instead of AdamWQwen3-2.38B · Training from scratchtraining tokens saved to match AdamW loss: 23.0%Improved
use Muon instead of AdamWQwen3-2.38B symmetric expansion · Training efficiencytraining tokens saved to match AdamW loss: 32.7%-36.0%Improved
use Muon instead of AdamWSmolLM2-540M symmetric expansion · Training efficiencytraining tokens saved to match AdamW loss: up to 44.6%Improved
use Muon instead of AdamWdiagonal expansion models · Training efficiencytraining tokens saved to match AdamW loss: 25.1% to 36.6%Improved
use Muon instead of AdamWQwen3-2.38B expanded · Block similarity layer 24 up_projblock similarity: 0.880 -> 0.788Improved

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