Research map / Architectures and efficiency / Attention and transformers
Pianist Transformer: Towards Expressive Piano Performance Rendering via Scalable Self-Supervised Pre-Training
Hong-Jie You, Jie-Jing Shao, Xiao-Wen Yang, Lin-Han Jia, Lan-Zhe Guo, Yu-Feng Li
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add large-scale self-supervised pre-training (ablation) | Pianist Transformer · ASAP test | overall intersection area: 0.6032 -> 0.8501 | Improved |
| add large-scale self-supervised pre-training (ablation) | Pianist Transformer · ASAP test | overall JS divergence: 0.4103 -> 0.1634 | Improved |
| add pre-training (ablation) | Pianist Transformer · ASAP test (Velocity) | JS divergence: 0.5363 -> 0.1805 | Improved |
| add pre-training (ablation) | Pianist Transformer · ASAP test (Duration) | JS divergence: 0.5399 -> 0.1879 | Improved |
| add pre-training (ablation) | Pianist Transformer · ASAP test (IOI) | JS divergence: 0.2789 -> 0.1740 | Improved |
| add pre-training (ablation) | Pianist Transformer · ASAP test (Pedal) | JS divergence: 0.2860 -> 0.1111 | Improved |
| use asymmetric 10-2 encoder-decoder instead of symmetric 6-6 (ablation) | Pianist Transformer · ASAP test | overall JS divergence: 0.174 -> 0.163 | Improved |
| use asymmetric 10-2 instead of symmetric 6-6 (ablation) | Pianist Transformer · efficiency | CPU inference speed: 1.0x -> 2.1x | Improved |
| use asymmetric 10-2 instead of symmetric 6-6 (ablation) | Pianist Transformer · efficiency | training VRAM: 1.0x -> 0.6x | Improved |
| use symmetric 6-6 instead of 2-layer decoder (ablation) | Pianist Transformer 135M · pre-training validation | loss: 1.260 -> 1.230 | Improved |
| scale pre-training data from 1B to 10B tokens (ablation) | Pianist Transformer 135M · pre-training validation | loss: 1.199 -> 1.195 | No clear effect |
| increase model size from 65M to 135M (ablation) | Pianist Transformer · pre-training validation | loss | No clear effect |
Closest papers by meaning
- Aria-MIDI: A Dataset of Piano MIDI Files for Symbolic Music Modeling (ICLR 2025 · similarity 0.88)
- Rubato: Transcribing Piano Music with Timestamps (NeurIPS 2026 · similarity 0.87)
- MuPT: A Generative Symbolic Music Pretrained Transformer (ICLR 2025 · similarity 0.87)
- MIDI-GPT: A Controllable Generative Model for Computer-Assisted Multitrack Music Composition (AAAI 2025 · similarity 0.87)
- PianoMotion10M: Dataset and Benchmark for Hand Motion Generation in Piano Performance (ICLR 2025 · similarity 0.86)
- MID-FiLD: MIDI Dataset for Fine-Level Dynamics (AAAI 2024 · similarity 0.86)
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.
