VS 2018 - Verification of Systems that Learn

Machine learning is of particular value in areas where developing a precise specification of desired behaviour is outside the scope of our current understanding of the world. For instance machine learning is widely deployed for image classification tasks. In these cases the specification is that the classifier should match the perception ability of a human. This is a difficult property to formally specify. Even when properties can be formally specified, the results of many machine learning systems (e.g. a set of weights in a neural network) are difficult to map onto these or to reason about in appropriate terms. The aim of this symposium is to bring together researchers interested in the question of how systems that learn may be verified. It will take the form of a number of scientific presentations and posters.