Seminar by Dr. Emtiyaz KHAN from RIKEN center for Advanced Intelligence Project

Modern data-driven AI systems can work extremely well but also fail miserably for unknown reasons. Fixing such failures are known to be extremely hard and designing systems that are transparent and trustworthy is even harder. I argue that this is possible but we have to fundamentally change the way we currently train our AI systems. I will discuss some of these insights that arise from our work on Bayesian principles, where the key idea is to design training methods that help us understand uncertainties and sensitivities of the AI building blocks. I will keep the discussion highly non-technical but can go into more detailed techniques if the need arises.

Emtiyaz Khan (also known as Emti) is a (tenured) team leader at the RIKEN center for Advanced Intelligence Project (AIP) in Tokyo where he leads the Approximate Bayesian Inference Team. Previously, he was a postdoc and then a scientist at Ecole Polytechnique Fédérale de Lausanne (EPFL), where he also taught two large machine learning courses and received a teaching award. He finished his PhD in machine learning from University of British Columbia in 2012. The main goal of Emti’s research is to understand the principles of learning from data and use them to develop algorithms that can learn like living beings. For more than 10 years, his work has focused on developing Bayesian methods that could lead to such fundamental principles. The approximate Bayesian inference team now continues to use these principles, as well as derive new ones, to solve real-world problems.