AI Paper Atlas

Research map / Optimization / Optimizers

Sign-SGD via Parameter-Free Optimization

Daniil Medyakov, Stanko Sergey, Gleb Molodtsov, Philip Zmushko, Evseev Grigoriy, Egor Petrov, Aleksandr Beznosikov

ICLR 2026 · Optimization › Optimizers · approach: Memory-efficient LLM optimizers

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 tuned cosine-scheduler stepsize with parameter-free ALIAS (no momentum)Sign-SGD · C4 LLaMA-130Mvalidation loss: 2.980 -> 3.006 (with wd)No clear effect
replace tuned AdamW and Prodigy with ALIAS Adam versionALIAS Adam · C4 LLaMA-130Mvalidation loss: 2.929 (AdamW) -> 2.918Improved
use ALIAS Adam version instead of other parameter-free optimizersALIAS Adam · C4 LLaMA-130Mvalidation loss: 2.925 (MoMo) -> 2.918Improved
add momentum to ALIAS (Adam version)ALIAS · C4 LLaMA-130Mvalidation loss: 3.006 -> 2.976Improved
add cosine scheduler on top of ALIAS Adam version (ablation)ALIAS Adam · C4 LLaMA-130Mvalidation loss: 2.976 -> 2.918Improved
use ALIAS Adam version on 350M modelALIAS Adam · C4 LLaMA-350Mvalidation loss: 2.717 (Sign-SGD) / 2.719 (AdamW) -> 2.707Improved
use approximated l_inf norm in memory-efficient ALIAS (ablation)ALIAS · C4 LLaMA-130Mvalidation lossHurt
add constant L_inf term to ALIAS stepsize (0 to 1000) (ablation)ALIAS · C4 LLaMA-130Mvalidation loss: 3.006 -> 3.006 (L_inf 0 vs 1000)No clear effect
change initial distance d0 from 1 to 10^-3 (ablation)ALIAS Adam · C4 LLaMA-130Mvalidation loss: 2.918 -> 2.918No clear effect
reduce batch size from 512 to 64 (more gradient noise) (ablation)ALIAS · C4 LLaMA-130Mvalidation loss: 3.006 -> 3.029 (Sign-SGD 2.980 -> 2.999)Hurt
use ALIAS Adam version for Swin fine-tuning instead of tuned AdamWALIAS Adam · Tiny ImageNet (Swin-T)final accuracy: 77.612 -> 79.161Improved
use ALIAS Adam version instead of AdamW on OGBGALIAS Adam · AlgoPerf OGBGmAP: 0.254 -> 0.242Hurt

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