There are two distinct parts to this unit. In the first few lectures I will provide the students with a general overview of connectionism: its origins as an attempt to model the functioning of the brain, and the various classes of algorithms created starting from these foundations. In the second part I will zoom in on the last 10-15 years. I will focus on a general framework for designing machine learning models that deal with complex structured data. I will introduce graphical models and bayesian networks and describe inference and learning algorithms for them. In machine learning there are many details one doesn't want to talk about in public. During the class I will try to address some of these issues for the case of neural networks, i.e. to describe possible strategies for effectively training them in real-world scenarios. Throughout the class and especially towards the end I will show applications of connectionist models to real problems. Together with more classical fields such as image classification and language processing, I will spend some time on applications to biological data, which has been the main focus of my research for the last few years.
