Anders is Enhancing AI Performance and Explainability through Data Reduction

The modern push for AI and data driven decision making, coupled with the ease of automated data collection, has resulted in excessive large datasets. The excessive size of modern datasets provides challenges for existing data mining and machine learning techniques. On the other hand, data preprocessing is an integral step in data mining and machine learning as low quality data leads to low quality results and outcomes.

Data reduction algorithms provide a means to address both of these problems, alongside being within the toolbox of explainable artificial intelligence techniques. This figure is a primer to 3 novel optimisation-based wrapper data reduction algorithms, two instance selection (SpFixedIS and SpIS) and one hybrid selection algorithm (SpIFS).

The figure below was created by Anders in collaboration with Graphics et al. as part of a DHCRC visual communication activity.

Click to enlarge