BayCon: Model-agnostic Bayesian Counterfactual Generator

Staff - Faculty of Informatics

Date: 21 June 2022 / 13:00 - 14:00

Online on MS Teams

Click here to join

Speaker: Martin Gjoreski, Università della Svizzera italiana

Abstract: In this talk, I will present BayCon, a novel method for generating counterfactuals. Generating counterfactuals to discover hypothetical predictive scenarios is the de facto standard for explaining machine learning models and their predictions. However, building a counterfactual explainer that is time-efficient, scalable and model-agnostic, in addition to being compatible with continuous and categorical attributes, remains an open challenge. To complicate matters even more, ensuring that the contrastive instances are optimised for feature sparsity, remain close to the explained instance and are not drawn from outside of the data manifold is far from trivial. To address this gap we pro-pose BayCon: a novel counterfactual generator based on probabilistic feature sampling and Bayesian optimisation. Such an approach can combine multiple objectives by employing a surrogate model to guide the counterfactual search. We demonstrate the advantages of our method through a collection of experiments based on six real-life datasets representing three regression and three classification tasks.

Biography: Martin is a postdoc researcher at the Faculty of Informatics, Università della Svizzera italiana (USI), in the research group of prof. Marc Langheinrich. His research field is AI with a focus on the development of standard machine learning and deep learning methods for sensor data. He is particularly interested in applications in fields such as mobile and wearable computing, behavioral analytics, affective computing, and mobile healthcare. From 2014 to 2020, he was part of the Department of Intelligent Systems at the Jozef Stefan Institute, where he did his MSc and PhD studies under the supervision of prof. Matjaz Gams and prof. Mitja Lustrek. In the years 2019 and 2020, was also a teaching assistant at the Faculty of Computer and Information Science, University of Ljubljana. In 2020, Martin was awarded the “Jožef Stefan golden emblem”, signifying an outstanding PhD thesis in Slovenia.

Host: Prof. Silvia Santini