AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Gradient Smoothing: Coupling Layer-wise Updates for Improved Optimization

Haoming Meng, Anton Sugolov, Vardan Papyan

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: Representation geometry and collapse · 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
add window smoothing (alpha 0.1, Standard, Proj) to AdamW RLVRGRPO on DeepSeek-R1-Distill-Qwen-1.5B · RLVR math reasoning averagepass@1 accuracy (%): 55.45 -> 57.60Improved
add window smoothing (alpha 0.1, Standard, Proj) to AdamW RLVRGRPO on DeepSeek-R1-Distill-Qwen-1.5B · AIME24pass@1 accuracy (%): 35.00 -> 40.83Improved
add window smoothing (alpha 0.2, Standard, Proj) to AdamW RLVRGRPO on DeepSeek-R1-Distill-Qwen-1.5B · AIME25pass@1 accuracy (%): 24.17 -> 30.00Improved
add window smoothing (alpha 0.2, Standard, Proj) to AdamW RLVRGRPO on DeepSeek-R1-Distill-Qwen-1.5B · AMC23pass@1 accuracy (%): 78.12 -> 77.50Hurt
smooth only projection layers (Proj) instead of all block parameters (Full) at alpha 0.1 (ablation)window smoothing in RLVR · RLVR math reasoning averagepass@1 accuracy (%): 56.80 -> 57.60Improved
increase smoothing strength from alpha 0.1 to 0.2 (Standard, Proj) (ablation)window smoothing in RLVR · RLVR math reasoning averagepass@1 accuracy (%): 57.60 -> 57.38No clear effect
add window smoothing (alpha 0.2, Dir, Proj) to AdamWViT-B · CIFAR-100top-1 accuracy (%): 74.56 -> 75.62Improved
add window smoothing (alpha 0.1, Norm, Proj) to AdamWViT-B · CIFAR-100top-1 accuracy (%): 74.56 -> 75.44Improved
add window smoothing (alpha 0.2, Norm, Proj) to AdamWU-ViT diffusion · CIFAR-10 FID@10kFID: 6.58 -> 5.82Improved
add window smoothing (alpha 0.2, Norm, Proj) to AdamWU-ViT diffusion · CIFAR-10 FID@50kFID: 4.01 -> 3.74Improved
add window smoothing (alpha 0.1) to Adam+NorMuon pretrainingnanochat GPT, depth 24 and 30 · validation loss and COREvalidation loss, COREImproved
increase window smoothing strength alpha from 0.1 to 0.4ViT-B on CIFAR-100 · cosine similarity of layer contributionsmean cosine similarityImproved

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