AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Museformer: Transformer with Fine- and Coarse-Grained Attention for Music Generation

Botao Yu, Peiling Lu, Rui Wang, Wei Hu, Xu Tan, Wei Ye, Shikun Zhang, Tao Qin, Tie-Yan Liu

NeurIPS 2022 · Architectures and efficiency › Attention and transformers · approach: Transformer expressivity and circuits · 20 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 fine- and coarse-grained attention with similarity-selected structure-related barsMuseformer · Lakh MIDI test, first 10240 tokensPPL: 1.64 -> 1.35Improved
Use fine- and coarse-grained attention with similarity-selected structure-related barsMuseformer · Lakh MIDI generated musicSE (%): 1.97 -> 0.95Improved
Use Museformer rather than the strongest comparison modelLakh MIDI human evaluation · Human evaluation, Lakh MIDI samplesOverall: 6.44 ± 2.01 -> 7.12 ± 1.81Improved
Add coarse-grained attention to summarize non-structure-related bars (ablation)Museformer · Lakh MIDI test, first 1024 tokensPPL: 1.65 -> 1.64Improved
Add coarse-grained attention to summarize non-structure-related bars (ablation)Museformer · Lakh MIDI test, first 5120 tokensPPL: 1.42 -> 1.41Improved
Add coarse-grained attention to summarize non-structure-related bars (ablation)Museformer · Lakh MIDI test, first 10240 tokensPPL: 1.38 -> 1.35Improved
Add coarse-grained attention to summarize non-structure-related bars (ablation)Museformer · Lakh MIDI generated musicSE (%): 1.08 -> 0.95Improved
Select fine-grained bars using similarity statistics rather than the most recent 8 bars (ablation)Museformer · Lakh MIDI test, first 1024 tokensPPL: 1.65 -> 1.64Improved
Select fine-grained bars using similarity statistics rather than the most recent 8 bars (ablation)Museformer · Lakh MIDI test, first 5120 tokensPPL: 1.43 -> 1.41Improved
Select fine-grained bars using similarity statistics rather than the most recent 8 bars (ablation)Museformer · Lakh MIDI test, first 10240 tokensPPL: 1.39 -> 1.35Improved
Select fine-grained bars using similarity statistics rather than the most recent 8 bars (ablation)Museformer · Lakh MIDI generated musicSE (%): 6.39 -> 0.95Improved
Disable precise fine-grained attention to previous bars, retaining only summarized previous-bar information (ablation)Museformer · Lakh MIDI music generationcoherenceHurt

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