Uncertainty Propagation in Factorised Latent Variable Models

The idea of intelligent machines have fascinated humans for decades. Technological developments have made what was a distant future seem realistic to achieve in our current time. Today we are surrounded by machines many of which are performing tasks and behaving in manners deemed as requiring intelligence for a human to perform.
Humans can be considered intelligent due to our capability of making energy saving decisions in scenarios where information is uncertain. 
When it is not possible to disambiguate the state of the world or the outcome of an action from current knowledge, humans have a fantastic ability to somehow end up making a sensible decision. If we are to achieve machines that are truly intelligent in the manner of a human this behaviour of our brains needs to be translated to computational machines. In this talk I will talk about how uncertainty can be modeled and describe the mechanisms that allow knowledge to be propagated through a reasoning process. I will discuss on an abstract level what intelligence means and why uncertainty plays such a centra part. I will then focus on a specific part of reasoning, namely representation. How can we represent data in such a manner that it can be translated to knowledge such that reasoning can be facilitated by simple relational structures and geometrical relationships.

The Lecture will be held in Swedish.