# Talk by M. Hermann: «Logical Analysis of Data»

**Speaker:**Miki Hermann

**Location:**Online

**Date:**Thu, 18 Feb 2021, 14:30-15:30

**Abstract:** We show how to cope with Big Data by means of Automated Deduction. We generate a Horn or a bijunctive formula from sets of positive and negative tuples T and F, respectively These sets of tuples have been previously shortened to a locally minimal length, provided that the two sets T and F keep an empty intersection. Since the problem to find the minimal length is an NP-optimization problem, we apply different approximation strategies. We also apply two approaches to formula production: (1) the LARGE approach, where only tuples of F falsify the produced formula, and (2) EXACT, where only tuples of T satisfy the formula. In case of the large approach we apply a set cover approximation algorithm to keep the minimal set of clauses falsifying all tuples from the set F. The produced formula can be used to further characterize subsequent sets of tuples and identify their membership among the true or false examples. The whole system, called MCP for Multiple Characterization Problem, has been implemented and can be found at https://github.com/miki-hermann/mcp.