New benchmark evaluates foundation model transferability for EMG tasks

Researchers have introduced EMG-FM-Bench, a new benchmark designed to evaluate how well foundation models (FMs) can be transferred and adapted for electromyography (EMG) tasks. The benchmark integrates 20 public datasets, encompassing over one million EMG segments, and assesses nine pre-trained FMs. Key findings indicate that while linear probing offers insights into pre-trained representations, full fine-tuning can significantly alter downstream performance. The study also revealed that the benefits of pretraining are not universal across all models and that performance degrades when models are applied to new users, though adaptation techniques can partially mitigate this. AI

IMPACT Provides a framework for assessing the practical applicability of general foundation models to specialized physiological signal processing tasks.

RANK_REASON Academic paper introducing a new benchmark for evaluating foundation models on a specific domain (Electromyography). [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• Electromyography
• EMG-FM-Bench
• foundation model
• Gotit.pub
• Hugging Face
• ScienceCast
• Tianhao Wu

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →