Machine learning - probabilistic perspective

Abstract

The talk aims to describe and explain how variational autoencoders work. How to derive the loss function to perform training? Understanding of those concepts involves prior knowledge about probability and statistics. During the talk, I will introduce those concepts based on easier algorithms. I will begin from a very simple example of parameter estimation of Gaussian distribution. Then I will be introducing more and more complex theoretical principles relying on algorithms such as linear regression, logistic regression, and probabilistic PCA. This incremental gain of knowledge will finally lead to a better understanding of variational autoencoders as well as the use of probability theory in machine learning.