Diverse data selection via combinatorial quasi-concavity of distance covariance: a polynomial time global minimax algorithm (Q2312659)

This paper develops the systematic approach to one of the central problems in machine learning: features engineering. Namely the following problems have been formulated: (i) \textit{diverse feature selection}, i.e. a subset of features that have the least statistical dependence amongst each other; (ii) \textit{all-relevant feature selection} and (iii) \textit{diverse and relevant feature selection}. This paper reports on the greedy algorithm for diverse variable selection with distance covariance. Namely, the problem is formulated as a quasi-concave set function maximization. The proposed approach is tested on a UCI-ML dataset.