A data-driven active learning approach to reusing ML solutions in scientific applications

Hamideh Hajiabadi, Christopher Gerking, Lennart Hilbert, Anne Koziolek

Abstract: Highlights•Active learning framework unlocks potential of image segmentation algorithms without annotations.•Proposing unsupervised metric for initial algorithm choice and settings in segmentation.•Creating an object-to-parameter map for segmentation tasks.•Framework yields satisfying, efficient results in biological case study with minimal interactions.