Interpretation and representation of learned classifiers

Machines that learn from examples have long been a dream of computer scientists, and new generations of such machines are becoming increasingly versatile. For example, when properly trained, such machines can sometimes classify medical data much faster and with better accuracy than domain experts. However, many researchers applying these techniques need to know what the machines learned or which facts were taken into account when making the decisions. In my research I try to develop techniques and tools to provide this information in a human-comprehendible way. I also investigate new generations of learning machines based on our increasing knowledge of how the brain processes information.