Friday, September 24, morning
Learning upon evolving data has gained in importance and interest during the last years: on the one hand, scholars recognized that models learned upon a static dataset need to be revised as new data are added; on the other hand, the stream paradigm has brought forward a large set of applications where data arrive at high speed and must be processed and then forgotten. In this tutorial, we discuss the theoretical underpinnings of supervised and unsupervised learning on evolving data. We point out the new challenges introduced by evolving data like resource aware learning, change detection, novelty detection, multi-horizons analysis, and reasoning about the learning process. We elaborate on important application areas where data evolution must be taken into account: we discuss the impact of evolution on economical data, and on understanding social networks; we investigate how learning under constraints (time, storage capacity and other resources) is affected by data evolution; we identify applications that require model learning over complex data (as in Customer Relationship Management or social tagging systems) and present appropriate adaptive learning methods.