ASELMAR: Active and semi-supervised learning-based framework to reduce multi-labeling efforts for activity recognition

Aydin Saribudak, Sifan Yuan, Chenyang Gao, Waverly Gestrich-Thompson, Zachary P. Milestone, Randall S. Burd, Ivan Marsic

Abstract: Highlights•ASELMAR, a deep learning framework, reduces labeling time for activity recognition.•Label verification time was 12.1% of the manual annotation effort by domain experts.•The performance improvement was 8.3% by training the model with a few iterations.•ASELMAR’s performance for a speech recognition study shows its broad applicability.•The detection issues due to occlusion and transparent objects are a future direction.