Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
From Pose to Muscle: Multimodal Learning for Piano Hand Muscle Electromyography
Authors: RUOFAN LIU, YICHEN PENG, Takanori Oku, Chen-Chieh Liao, Erwin Wu, Shinichi Furuya, Hideki Koike
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | A comprehensive evaluation compares our EMG inference approach with several baselines in piano playing and analyzes its generalization and adaptation across users and tasks. |
| Researcher Affiliation | Collaboration | 1Institute of Science Tokyo, 2Sony Computer Science Laboratories |
| Pseudocode | No | The paper describes the methodology and network architecture in detail but does not present any explicitly labeled pseudocode or algorithm blocks. The processes are described in paragraph form. |
| Open Source Code | Yes | Instructions regarding accessing and using our Piano KPM dataset and network are provided at https://github.com/ruofanliu0129/Piano KPMNet.git. |
| Open Datasets | Yes | To the best of our knowledge, this is the largest publicly available EMG dataset of professional piano performance, comprising data from 20 expert pianists performing 7 distinct musical tasks, with 12.64 hours of high-quality recordings. Instructions regarding accessing and using our Piano KPM dataset and network are provided at https://github.com/ruofanliu0129/Piano KPMNet.git. |
| Dataset Splits | Yes | Table 3: Dataset split and results for held-out evaluations. We separate two test sets to measure generalization to new users (Cross-User) and different performance tasks (Cross-Task). ... We partition the dataset into training, validation, and test sets with an approximate ratio of 70% : 10% : 20%. |
| Hardware Specification | Yes | Model training is performed on a high-performance computing system with an AMD EPYC 9654 96-core/192-thread processor, 768 Gi B DDR5-4800 RAM, and NVIDIA H100 SXM5 GPUs, and the entire process takes approximately 14 hours. Model inferring is conducted on a more accessible setup with an Intel Core i9-10900X CPU, 128 GB RAM, and an NVIDIA Ge Force RTX 4090 with 24 GB GPU memory. |
| Software Dependencies | No | The paper mentions 'implemented on the popular framework Py Torch' but does not specify the version number of PyTorch or any other software dependencies with their versions. |
| Experiment Setup | Yes | We train the model for 200 epochs to ensure sufficient convergence. The batch size is set to 64 to balance computational efficiency and training stability. EMG and keystroke signals sampled at 1000 Hz use a window length of 1024 for both training and inference. ... The model employs an Adam W optimizer with a learning rate of 0.0001, and a Step LR scheduler with a step size of 20 and a decay factor (gamma) of 0.5. The loss function is a weighted combination of MSE and OT losses, where the weights λmse and λot are both set to 1. |