Experience replay adapts sEMG gesture models better than EWC
Original title: Cross-User Adaptation for sEMG Gesture Recognition via Continual Learning
On NinaPro DB1, experience replay reaches 83.3% class-incremental accuracy and limits forgetting to a 1.8-5.8% drop.
Gesture models from surface electromyography must adapt to new wearers and classes after deployment. On NinaPro DB1, experience replay beats elastic weight consolidation and naive fine-tuning under domain- and class-incremental drift. Replay holds forgetting to a 1.8-5.8% accuracy drop, versus 16.2-28.5% for the other two. Shuffled, independently windowed segments train stably; ordered streaming windows collapse predictions to one gesture.
- Hand-crafted time-domain features raise per-task accuracy by up to about 40% over raw signals.
- Only replay preserves that feature gain under class-incremental drift.
- Elastic weight consolidation and naive training forget 16.2-28.5% accuracy.
From the original article
1 min read at arxiv.orgProsthetic and assistive interfaces increasingly depend on decoding hand gestures from surface electromyography (sEMG). Because the signal is shaped by person-specific factors, the recurrent neural networks (RNNs) commonly used to model these temporal streams must adapt to new wearers and new gesture classes after deployment.
By Aran Ramakrishnan · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.