Matthias Bethge

Title: Unsupervised representation learning with generative image models.
Abstract: After the great successes of supervised learning in computer vision an important research direction in the field is to improve on unsupervised methods by learning generative models from unlabelled data. A major difficulty in unsupervised learning, however, is the lack of a commonly accepted benchmark and of suitable objective functions to measure and compare the performance of different models against each other. In this tutorial I will give an overview on unsupervised learning with generative image models with particular emphasis on model comparison using both likelihood and alternative criteria.