Invited Speaker: Sergei Kuznetsov

A model of learning from positive and negative examples is naturally described in terms of Formal Concept Analysis (FCA). In these terms, a result of learning consists of two sets of intents (closed subsets of attributes): the first contains intents such that the corresponding extents consists only of positive examples. The second contains intents such that the corresponding extents consists only of negative examples. On the one hand, we show how FCA allows one to realize learning with various data representations, from standard object-attributes to labelled graphs. On the other hand, we use FCA to describe some standard models of Machine Learning such as version spaces and decision trees.

This allows one to compare several Machine Learning approaches, as well as to employ some standard techniques and algorithms of FCA in the domain of Machine Learning. We consider several applications of the FCA-based learning, including analysis of historical documents and Structure-Activity Relationship problem (bioinformatics), as well as competitions in predictive toxicology and spam filtering