AI Paper Atlas

Research map / Optimization / Optimizers

Robustness to Unbounded Smoothness of Generalized SignSGD

Michael Crawshaw, Mingrui Liu, Francesco Orabona, Wei Zhang, Zhenxun Zhuang

NeurIPS 2022 · Optimization › Optimizers · approach: SGD momentum and step-size schedules · 4 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
Use generalized SignSGD instead of Adam20-layer ResNet · CIFAR-10test accuracy: 0.8823 ± 0.0041 -> 0.8835 ± 0.0032Improved
Use generalized SignSGD instead of Adam20-layer ResNet · CIFAR-10training loss: 0.1161 ± 0.0111 -> 0.1086 ± 0.0129Improved
Use generalized SignSGD instead of Adam3-layer AWD-LSTM · Penn Treebanktest perplexity: 58.9005 ± 0.3058 -> 58.9661 ± 1.5218No clear effect
Use generalized SignSGD instead of SGDClipGrad3-layer AWD-LSTM · Penn Treebanktraining loss: 3.5774 ± 0.0081 -> 3.7928 ± 0.0425Hurt
Remove gradient clipping from Adam (ablation)16-layer GPT-2 · Wikitext-103training loss: 2.79 vs 2.76No clear effect
Remove gradient clipping from Adam (ablation)16-layer GPT-2 · Wikitext-103perplexity: 27.92 vs 27.97No clear effect
Set beta2 to 0, recovering SignSGD with Momentum (ablation)Generalized SignSGD · Non-convex optimization under coordinate-wise (L0, L1)-smoothnessconvergence bound dependence on gradient normsImproved
Add momentum to SignSGD-type updatesGeneralized SignSGD · Non-convex optimization under unbounded smoothness and stochastic gradientsconvergence guaranteeImproved
Use clipping with Adam rather than Adam without clipping (ablation)16-layer GPT-2 · Wikitext-103training loss: 2.76 vs 2.79No 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