AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Root Mean Square Layer Normalization

Biao Zhang, Rico Sennrich

NeurIPS 2019 · Architectures and efficiency › RNNs and normalization · approach: Normalization layer theory · 106 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
Replace LayerNorm with RMSNormRNNSearch · WMT14 English-German, TensorFlow newstest2014SacreBLEU: 22.6 -> 22.4No clear effect
Replace LayerNorm with RMSNormRNNSearch · WMT14 English-German, TensorFlow newstest2017SacreBLEU: 23.6 -> 23.7No clear effect
Replace LayerNorm with RMSNormRNNSearch · WMT14 English-German, TensorFlowseconds per 1k training steps: 665±32.5s -> 501±11.8s ({24.7%})Improved
Replace RMS normalization with L2 normalization (ablation)RNNSearch · WMT14 English-German, TensorFlow newstest2014SacreBLEU: 22.4 -> 20.7Hurt
Estimate RMS from 6.25% of summed inputs (ablation)RNNSearch · WMT14 English-German, TensorFlow newstest2014SacreBLEU: 22.4 -> 22.6No clear effect
Estimate RMS from 6.25% of summed inputs (ablation)RNNSearch · WMT14 English-German, TensorFlow newstest2017SacreBLEU: 23.7 -> 23.1No clear effect
Remove normalization (ablation)Transformer · WMT14 English-GermanSacreBLEUHurt
Replace LayerNorm with RMSNormTransformer · WMT14 English-German, newstest2014SacreBLEU: 26.6 -> 26.8No clear effect
Replace LayerNorm with RMSNormTransformer · WMT14 English-German, newstest2017SacreBLEU: 27.7 -> 27.7No clear effect
Replace LayerNorm with RMSNormTransformer · WMT14 English-Germanseconds per 1k training steps: 248±1.31s -> 231±0.04s ({6.9%})Improved
Replace LayerNorm with RMSNormOrder-embedding model · Microsoft COCO image-caption retrievalCaption Retrieval R@1: 47.9 -> 48.7Improved
Estimate RMS from a subset of summed inputs (ablation)Order-embedding model · Microsoft COCO image-caption retrievalCaption Retrieval R@1: 48.7 -> 46.8Mixed

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