MLG 2006: Proceedings of the International Workshop on Mining and Learning with Graphs : in conjunction with ECML/PKDD 2006

Abstract

At a time where the amount of data collected day by day far exceeds the human capabilities to extract the knowledge hidden in it, it becomes more and more important to automate the process of learning. Typical data collections have two things in common: They are huge and the information stored in them is highly structured. Graphs are one of the most popular data representations in mathematics, computer science, engineering disciplines, and other natural sciences. This workshop on Mining and Learning with Graphs (MLG) thus concentrated on learning from graphs and its subclasses such as but not limited to trees, sequences (GTS). The primary goal of MLG was to bring together researchers working on various aspects of mining and learning with graphs. It is hence in the tradition of previous ECML/PKDD workshops on Mining Graphs, Trees, and Sequences (MGTS) but extends its scope to include other areas of machine learning and data mining also concerned with graphs and their subclasses such as: Algorithmic aspects of Theoretical aspects of Open problems in Novel applications of Evaluative studies of the following non-exclusive list of topics Kernels and Distances for graphs. GTS-structured output spaces. Frequent GTS mining. Learning with generative GTS models and compact (e.g., intensional) representations like GTS transformations, grammars, or matchings. Theoretical aspects of learning from GTS. Probabilistic modelling of GTS. GTS-based approaches to transductive and semi-supervised learning. Compared to previous workshops on MGTS, MLG was able to attract a record number of submissions (28 full and 6 short papers). For time and space restrictions, we could only accept 9 full papers and 15 short papers (most of which were originally submitted as full papers). For the first time, the workshop received the support of the PASCAL Network of Excellence that sponsored the invited talk and the best paper award. We most sincerely thank the PASCAL network for this sponsoring; the program committee and additional reviewers for their reviews; as well as the authors for their high quality submissions.