Title: Learning representations for relational learning and literature-based discovery
Abstract:
To learn from complex multi-relational databases, text corpora, or networks of interacting data and knowledge nodes, machine learning algorithms often require tabular data, where training instances are represented by fixed-length feature vectors. In practical machine learning applications, learning a suitable data representation is crucial for effective predictive modeling and pattern discovery. This talk presents selected representation learning techniques for automated data transformation into tabular data format, applicable to relational learning and literature-based discovery.
Biography:
Nada Lavrač is Research Councilor at Jožef Stefan Institute (JSI), and former Head of JSI Department of Knowledge Technologies (2004-2020). She is Full Professor at the University of Nova Gorica and the International Postgraduate School Jožef Stefan, where she acts as Head of ICT MSc and PhD programmes. Her research interests are in machine learning, text mining, computational creativity, with applications in bioinformatics, medicine, public health and media analysis. She co-authored 5 research monographs, including "Foundations of Rule Learning" (2013) and "Representation Learning: Propositionalisation and Embeddings" (2021), both published by Springer. She is EurAI Fellow, ELLIS Fellow and Board member, and recipient of several research awards, including the 2021 National Zois Award for outstanding scientific achievements in machine learning.