Dr. Roman Kazinnik, Data Scientist at Brighthouse Financial
Title:Deep learning and mathematical perspectives: historical developments and modern challenges
Abstract: Artificial Intelligence (AI) is often viewed as a massive parallel optimization problem, and has gained its popularity presumably due to the recent wide availability of parallel computing power. On one hand, AI departs from fundamental mathematical conceptions: it doesn’t prescribe any PDE when solving inverse problems, and AI learning algorithms do not care about dense representations in functional space. However, there is a great deal of mathematical principles that are in widely implemented by modern AI.
In this talk I am going to cover some of these techniques and principles that are deployed by AI, and familiar to the applied mathematicians. I will also show how a lack of rigorous underlying model presents, perhaps, one of the major challenges in deploying modern AI methodologies. I consider building such rigorous underlying modeling principles as one of the most interesting modern challenges applied mathematicians can find in AI.