AI Paper Atlas

Research map / Optimization / Optimizers

Learning compositional functions via multiplicative weight updates

Jeremy Bernstein, Jiawei Zhao, Markus Meister, Ming-Yu Liu, Anima Anandkumar, Yisong Yue

NeurIPS 2020 · Optimization › Optimizers · approach: Memory-efficient LLM optimizers · 5 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 Madam instead of tuned AdamResNet-18 · CIFAR-10top-1 test error: 6.6\pm0.2 -> 7.8\pm0.2Hurt
Use Madam instead of tuned SGDResNet-18 · CIFAR-100top-1 test error: 29.1\pm0.2 -> 30.2\pm0.1Hurt
Use Madam instead of tuned SGDResNet-50 · ImageNettop-1 test error: 24.1\pm0.1 -> 28.9\pm0.1Hurt
Use Madam instead of tuned AdamBigGAN cGAN · CIFAR-10FID: 23.9\pm0.9 -> 19 \pm 2Improved
Use Madam instead of tuned SGDTransformer · WikiText-2perplexity: 169.6\pm0.6 -> 173.3\pm0.6Hurt
Use 12-bit rather than FP32 MadamMadam with ResNet-18 · CIFAR-10top-1 test error: 7.8\pm0.2 -> 7.0\pm0.1Improved
Use 12-bit rather than FP32 MadamMadam with ResNet-18 · CIFAR-100top-1 test error: 30.2\pm0.1 -> 27.6\pm0.3Improved
Use 12-bit rather than FP32 MadamMadam with ResNet-50 · ImageNettop-1 test error: 28.9\pm0.1 -> 31.1\pm0.1Hurt
Use 12-bit rather than FP32 MadamMadam with BigGAN cGAN · CIFAR-10FID: 19.3\pm0.7 -> 19.8\pm0.8No clear effect
Use 12-bit rather than FP32 MadamMadam with Transformer · WikiText-2perplexity: 173.3\pm0.6 -> 182.3\pm0.6Hurt
Reduce weight precision to 8 bitsMadam · ImageNettop-1 test error: 28.9\pm0.1 -> 50.5\pm0.5Hurt
Reduce weight precision to 8 bitsMadam · WikiText-2perplexity: 173.3\pm0.6 -> 262\pm2Hurt

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