AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

BEAT: Tokenizing and Generating Symbolic Music by Uniform Temporal Steps

Lekai Qian, Haoyu Gu, Jingwei Zhao, Ziyu Wang

ICML 2026 · Architectures and efficiency › Attention and transformers · approach: State space and recurrent sequence models · 0 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 one-beat step instead of half-beat step (ablation)BEAT · Piano continuation (LMD+MuseScore)JS_GC / FMD: 0.060 -> 0.039 JS_GC; 464.4 -> 436.7 FMDImproved
use one-beat step instead of two-beat step (ablation)BEAT · Piano continuation (LMD+MuseScore)JS_GC / JS_SC: 0.145 -> 0.039 JS_GC; 0.086 -> 0.021 JS_SCImproved
use relative pitch intervals with descending order instead of descending + absolute (ablation)BEAT · Piano continuation (LMD+MuseScore)JS_GC: 0.103 -> 0.039Improved
use descending instead of ascending order (relative) (ablation)BEAT · Piano continuation (LMD+MuseScore)JS_GC: 0.049 -> 0.039Improved
use deterministic ordering instead of random order (ablation)BEAT · Piano continuation (LMD+MuseScore)FMD: 459.5 -> 436.7Improved
use BEAT sparse pattern encoding instead of naive 128-pitch piano-roll grid (ablation)BEAT grid · Piano continuation (LMD+MuseScore)FMD: 582.3 -> 436.7Improved
replace REMI with BEAT tokenization150M Transformer · Stepwise transposition (synthetic)sequence accuracy (%): 28.28 -> 91.92Improved
replace REMI with BEAT tokenization150M Transformer · Time-shift reconstruction, 2 beatsonset precision (%): 78.0 -> 94.2Improved
replace REMI with BEAT tokenization150M Transformer · Beat interleaving Mixedsequence accuracy (%): 7.37 -> 10.31Improved
replace REMI+ with BEAT tokenization150M Transformer · Multi-track continuationJS_SC: .008 -> .009No clear effect
replace REMI with BEAT tokenization150M Transformer · Piano continuationFMD: 550.9 -> 436.7Improved
replace AMT-L with BEAT tokenization150M Transformer · POP909 (out-of-distribution)FMD: 402.5 -> 365.9Improved

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