Title
Toward trustworthy AI - exploring the trade-offs between accuracy, privacy, and explainability
Abstract
The talk will provide a brief overview of trustworthy AI before delving into three specific aspects of trustworthiness, namely accuracy, privacy, and explainability, which are particularly relevant within the health domain. We will discuss how these aspects are interconnected by studying counterfactual explanations for differentially private support vector machines (SVM). Ensuring differential privacy in learning from sensitive data involves introducing uncertainty through noise, thereby compromising the identification of specific individuals in the database. Consequently, this process diminishes both classifier accuracy and explanatory credibility. The talk will introduce robust counterfactual explanations tailored for differentially private SVM classifiers, and will discuss the trade-offs among accuracy, privacy, and explainability using a specific dataset.
Bio
Rami Mochaourab is a research leader at RISE Research Institutes of Sweden, within the domain of applied AI. His primary focus is on trustworthy AI, with particular emphasis on Explainable AI and data privacy. He earned his M.Sc. in Information Systems Engineering and Ph.D. in Electrical Engineering from TU Dresden, Germany, in 2008 and 2012, respectively. Prior to his current role at RISE, he was a researcher at Fraunhofer HHI, Berlin, and held a researcher position at KTH Royal Institute of Technology from 2014 to 2018. Dr. Mochaourab has received three best paper awards and two exemplary reviewer awards from the IEEE.
Dr. Mochaourab currently serves as the Vice Chair of the working group Digitalized Industries at the Digital Futures Centre, Stockholm. In addition, he contributes to standardization of Trustworthy AI through his membership at the Swedish Institutes for Standards (SIS) and his role as a SIS expert at ISO/IEC JTC 1/SC 42 Artificial Intelligence, working group Trustworthiness.
Practical
This seminar is part of the seminar series on Integreat's objective Fair, Explainable and Trustworthy Machine Learning.
The Tuesdays seminars series are devoted to various topics relevant for the Integreat´s research focus. Presenters from the Integreat community and beyond have 40 minutes to present, followed by a group discussion.
Seminars are open for attendance for everybody.
For those not able to attend in person: https://uio.zoom.us/j/62535744013
Point of contacts:
- Organiser and moderator: Professor Robert Jenssen, UiT The Arctic University of Norway
- Integreat Administrative Leader Maria Dikova, University of Oslo