DIKU Bits: Image Analysis: Deep Learning is not the only path! – Københavns Universitet

DIKU Bits: Image Analysis: Deep Learning is not the only path!

Speaker

Francois Bernard Lauze, associate professor in the Image Analysis, Computational Modelling and Geometry Section (IMAGE).

Abstract

In the past few years, Deep Learning has invaded the field of Image Analysis and is so prevalent that for most of us, it seems that there is no salvation out of the Deep Learning realm. But what to do if you don't have 50 millions examples of a task available for training? In this presentation, I will discuss how inference, and especially Bayesian inference can be used to design optimization algorithms for series of task in Image Analysis. I will discuss it through a series of examples, ranging from video inpainting to tomographic reconstruction. I won't forget Deep Learning, and will briefly discuss interesting problems which bridge the two worlds.

Zooming in on Francois Bernard Lauze

Which courses do you teach? (BSc and MSc)
I am DIKU coordinator for Linear Algebra for Computer Scientists (B.Sc) and I teach several M.Sc courses: Vision and Image Processing, Numerical Optimization and Introduction to Data Science.

Which technology/research/projects/startup are you excited to see the evolution of?
I am interested in Hyperloop technologies. And in general fascinated by these new means of transportation which can change the geography / geometry of a country, region. At a very different scale, the Kibble balance is a wonderfully complex measurement device which will soon provide us with a new definition of the kilogram, based on Planck's constant.

What is your favorite sketch from the DIKUrevy?
DOS BOOT and other related sketches!