We are interested in designing data transformations (such as anonymization, compression, right-protection, etc.) that do not hinder the "mining capacity" of a dataset. How can we guarantee that mining the original or the transformed dataset would give identical results for machine-learning and data-mining tasks? Are there any commonalities between the various algorithms and transformations? More...

Graphs are one of the most prevalent forms of data representation. We design algorithms that can extract useful knowledge from both static and evolving graph structures. We also consider applications of these techniques in data visualization, data streams and recommendation systems. More...