Theoretical needs for pattern miningArnaud Soulet (Univ. de Tours)
Pattern mining is a subfield of Knowledge Discovery based on enumeration algorithms. This seminar will present works concerning pattern mining opening up research avenues related to complexity and combinatorics. First, a major challenge is to assess the quality of mining methods without resorting to experimental studies. We will present a preliminary work showing the importance of complexity to characterize mining methods. Second, a new pattern mining approach, called pattern sampling, has recently been introduced. It consists in drawing a sample of patterns with a probability proportional to a degree of interest. This instant discovery technique is unfortunately limited to a few configurations because of the counting problems. We will illustrate this problem for sequential pattern sampling.