Mathematics & Statistics Graduate Student Seminar - Chris Finlay

This week, Chris Finlay will talk to us about deep learning andadversarial attacks:

"Deep learning is to mathematicians now as mathematical finance was to physicists in the aughts." So said the wise sage to his pupil. In other words, you can make a lot of money doing deep learning, but does it have any rigour?

This talk will attempt to persuade the pizza-going public that yes, indeed, there can be rigour in the over-hyped world of deep learning. Moreover, I'll argue the field is in desperate need of skeptical mathematicians and statisticians. I will illustrate this point with a troubling problem in deep learning, called "adversarial attacks", which is the buzzy way of saying a function is not Lipschitz continuous. This problem manifests in, for example, image classification when a model mis-classifies a perturbed kitten as, say, an octopus. All Kool-Aid adverse mathematicians and statisticians are welcome.

See you all there!

Allgraduate students are invited. As with all talks in thegraduate student seminar, this talk will be accessible to allgraduate students in math and stats.This seminar was made possible by funding from the McGill Mathematics and Statistics Department and PGSS.