[Feb 20th 2019] Explaining the behavior of learning classification systems: a model agnostic approach

Roberto Prevete, Università di Napoli Federico II.

Feb 20, 2019 – 11:00 AM
DIISM, Artificial Intelligence laboratory (room 201), Siena SI

It is difficult to reconstruct and exhaustively explain the decisions/behaviors of current autonomous or semi-autonomous systems based on Machine Learning techniques. This characteristic is due to the fact that they usually do not possess an explicit, declarative representation of knowledge, so it is difficult to provide the required explanatory structures. This feature considerably limits the achievement of their full potential. Importantly, a user who interacts with such systems is interested in reviewing, analyzing their decisions in a way that is comprehensible to human beings. A possible approach to make systems based on machine learning techniques more transparent and understandable to human beings is to consider the system as a black-box (model agnostic approach). In this context, we will
discuss some typical problems and current solutions proposed in literature, and analyze the possibility to exploit the representational power of sparse dictionaries to construct explanations for classification decisions.


Roberto Prevete is researcher in Computer Science at the Department of Electrical Engineering and Information Technology (DIETI) of the University of Naples Federico II. He teaches the course “Machine Learning and applications”. His main research interests include machine learning and statistical methods applied to data analysis, computational neuroscience and computational modeling of biological information processing. He has had research responsibilities in several national and international projects.

 |  Category: Seminars