Abstract: Classical machine learning typically assumes that data is independently and identically distributed (IID), while in practice very often that's not the case. In this talk I will discuss challenges and
overview techniques for evaluating machine learned models when data is not IID. I will focus on three main settings: (1) data is autocorrelated spatially, (2) concept drift over time, (3) observations are non-independent phylogenetically. I will discuss challenges and perspectives for knowledge discovery and generalisation from machine learned models with practical examples from macroecology and industrial
process control.

Speaker: Indrė Žliobaitė is an associate professor at the Dept. of Computer science and Dept. of Geosciences and Geography, University of Helsinki, where she leads a research group focusing on data science for understanding evolutionary processes in nature and society. She has contributed popular algorithmic techniques and theory for learning from evolving data, pioneering work in fairness-aware machine learning, and new perspectives to macroevolutionary analyses.

Affiliation: University of Helsinki

Place of Seminar:  Kumpula exactum D122 (in person) & zoom ( Meeting ID: 640 5738 7231 ; Passcode: 825217)