Propositional Feature Extraction Using Random Forests and Deep Neural Networks
Binary attributes are an important special case of attributes used in machine learning. In this talk we will present our ongoing research in feature extraction for such data. One proposed approach relies on transformation of tree models to propositional formulae and another on imposing propositional structure on deep neural networks by problem specific regularization. By imposing propositional structure on extracted features, we hope to improve interpretability of those features, and yet keep prediction quality. Our methods are primarily intended for analysis of medical healthcare record data in which binary attributes indicate presence or absence of the disease or indicate if a treatment has been performed on the patient. However, they are not limited to that specific domain.
About the lecturer
Mladen Nikolić is an assistant professor of computer science at the Faculty of Mathematics of the University of Belgrade. He got his PhD in computer science from the University of Belgrade in 2013. His interests include machine learning, data mining, and automated reasoning. He worked as a visiting researcher at the Data Analysis and Biomedical Analytics (DABI) Center at Temple University for six months. He participated in several national and international research projects including two projects funded by Swiss National Science Foundation and one funded by DARPA