AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Mixtape: Breaking the Softmax Bottleneck Efficiently

Zhilin Yang, Thang Luong, Ruslan Salakhutdinov, Quoc V Le

NeurIPS 2019 · Architectures and efficiency › RNNs and normalization · approach: Differentiable optimization layers · 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
add gate sharing for infrequent tokens (ablation)Mixtape · Penn Treebankperplexity: 56.33 -> 56.37No clear effect
add gate sharing for infrequent tokens (ablation)Mixtape · One Billion Wordperplexity: 36.77 -> 36.52No clear effect
add gate sharing for infrequent tokens (ablation)Mixtape · Penn Treebanktraining time per instance, all layers, same memory (s): 10.44 -> 3.66Improved
add gate sharing for infrequent tokens (ablation)Mixtape · One Billion Wordtraining time per instance, all layers, same memory (s): 48.6 -> 11.3Improved
replace MoS-15 with Mixtape output layerPenn Treebank LSTM · Penn Treebankperplexity: 56.14 -> 56.37No clear effect
replace MoS-7 with Mixtape output layerOne Billion Word LSTM · One Billion Wordperplexity: 37.10 -> 36.52Improved
replace softmax with Mixtape output layerLSTM · Penn Treebankperplexity: 59.19 -> 56.37Improved
replace softmax with Mixtape output layerLSTM · One Billion Wordperplexity: 42.77 -> 36.52Improved
replace softmax with Mixtape output layerTransformer · WMT14 En-DeBLEU: 29.0 -> 29.3Improved
replace softmax with Mixtape output layerTransformer · WMT14 En-FrBLEU: 43.0 -> 43.9Improved
replace MoS-9 with Mixtape output layerTransformer · WMT14 En-DeBLEU: 29.5 -> 29.3Hurt
replace MoS-9 with Mixtape output layerTransformer · WMT14 En-FrBLEU: 42.1 -> 43.9Improved

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