Katedra za umetno inteligenco vabi na debato z naslovom “Interpretable Features for Local Explanations of Machine Learning Models” v četrtek, 16. 5. 2019, tokrat izjemoma ob 13.15, v P03. Debato bo pripravil Mateusz Staniak.
 
The contents:
Recently, interpretability became a major issue in machine learning due to social implications of using black-box models (for example for recidivism prediction) and legal regulations (GDPR). One of popular approaches to model explanations, first introduced as LIME (Local Interpretable Model-agnostic Explanations, Ribeiro et al. 2016),  is based on fitting an interpretable model around an observation to provide rationale for the decision of the model. We will describe issues that arise when translating the LIME algorithm from the context of images and texts to tabular data.
 
About the speaker: Mateusz Staniak is a PhD candidate in the MI2 Data Lab group a the Warsaw University of Technology, Faculty of Mathematics and Information Science. His main area of research is interpretable machine learning. He is interested in applications of machine learning and statistics to biology and medicine.