Bonn PhD student Paolo Climaco is the winner of the Dataperf Vision Selection Challenge

Bonn, 16.09.2023. Paolo Climaco, PhD student at the Bonn Institute for Numerical Simulation (INS) and member of our BIGS (Bonn International Graduate School of Mathematics), was announced as the winner of the Dataperf Vision Selection Challenge.

The challenge consisted of developing an algorithm to select the most effective training set, with at most one thousand points, from a large pool of more than three million images to maximize the performance of binary classifiers for various visual concepts.

The algorithm, which Paolo developed, is based on farthest point sampling, a fill distance minimization approach with some interesting properties, some of which Paolo and his supervisor Jochen Garcke analyzed in their research paper "Investigating minimizing the training set fill distance in machine learning regression". The research paper was presented at theDMLR workshop as part of the ICML conference last July 2023 in Honolulu, HI.

The challenge was organized by members of the Dataperf community, which consists of researchers from industry and academia focused on data-centric AI. Data-centric AI is an emerging field of research that has gained much attention in the machine learning community, and is characterized by a shift in research focus from improving the learning model design to improving data quality.

See here for the challenge.