Many problems in large scale social networks, data mining, and machine learning can be solved efficiently by methods drawn from optimization (including convex optimization) or using methods based on stochastic search. This group's research focus is to analyze the convergence behavior of many different algorithms to solve machine learning problems, and to compare the theoretical predictions with the observed behavior from computational experiments.
A PhD thesis from this group, "Computer Go and Monte Carlo Tree Search: Opening Book and Parallel Solutions," was published during 2018. The group continues to work improving the statistical accuracy of the thesis work and examining the relation between the perceived difficulty of the game and the number of nodes (trees) at which winning rates no longer measurably improve. The perceived difficulty of the game was measured by observing the concentration of answers provided by the trees on a move by move basis. If the concentration was low (many different answers) the game was rated as more difficult than if the concentration was high. Continuing work will go deeper into this correlation by testing the effects of increasing the number of nodes on selected problems which have known best answers rather than entire games and the winning rates over those games.
A joint project with the U of M Institute of Virology explores the genomics of bacteria and archaea, specifically the incidence of self-targeting spacers in CRISPR arrays. Normally spacers are key to the organisms defense against phage infections, but self-targeting is a sign of a potentially destructive auto-immune response. The researchers use rapid scans of genetic sequence data to find instances.