Curious instance selection

Michal Moran, Tom Cohen, Yuval Ben-Zion, Goren Gordon

Abstract: Highlights•Instance selection is aimed at reducing data while minimizing performance loss.•Curious Instance Selection (CIS) is based on reinforcement learning and artificial curiosity.•CIS learns the value of instance clusters with respect to the specific data science task.•CIS provides control over data reduction and model quality, supporting decision making.