[Seminar] Data stream mining: going beyond training then testing

Abstract: Processing and analyzing potentially unbounded data sequences, the so-called data streams, is key in multiple applications. However, assuming such streams as stationary is bound to fail. In this talk, we will discuss what data stream mining are and what are the fundamental challenges they bring to the table, including learning over time, computational restrictions, and concept drift. Finally, I will bring forward some of the ongoing research we have in our research group including green data stream mining, concept drift adaptation in image classification, and textual data streams.

Bio:  Jean Paul Barddal received his Ph.D. in Informatics from the Graduate Program in Informatics (PPGIA), in 2018, where he is currently a professor. He is also an executive member of Center for Artificial Intelligence Solutions (CISIA) and the Artificial Intelligence Institute (AI2). His research topics include machine learning from data streams, including classification, regression, clustering, and recommender systems. He has published over 60 papers including developments to algorithms in data stream mining or their application to real-world problems.