Mikhail Moshkov
King Abdullah University of Science and Technology, Saudi Arabia
Title: Extensions of dynamic programming for machine learning and knowledge representation
Biography
Biography: Mikhail Moshkov
Abstract
We discuss so called multi-pruning which allows us to construct classifiers (decision trees) that outperform often classifiers constructed by CART. This approach is based on the construction of the set of Pareto optimal points for bi-criteria optimization problem relative to the size of decision trees and the number of misclassifications. The second topic is connected with multi-stage optimization of decision rules relative to the coverage and length. Based on this optimization procedure, we can simulate the work of greedy algorithm for the set cover problem. As a result, for many datasets from UCI ML Repository, we can construct small systems of enough accurate decision rules that cover the most part of objects (rows). The end of the presentation is devoted to the introduction to KAUST