From Bits to Information: Theory and Applications of Learning Machines

Learning is becoming the central problem in trying to understand the brain and in trying to develop intelligent machines. In this talk I will give an up-to-date outline of some of our recent efforts in developing machines that learn. I will sketch our work on statistical learning theory and in particular a theoretical framework for classification and function approximation which connects regularization theory and Support Vector Machines. Our main application focus is classification (and regression) in various domains -- such as sound, text, video and bioinformatics. In particular, I will describe the evolution of trainable, hierarchical classifiers for problems in visual object detection. I will also speculate on the implications of this research for how the brain works and review some recent models and neuroscience data which provide a glimpse of how the visual cortex learns to identify and categorize visual objects.