IMA Annual Program Year Workshop

Visual Learning and Recognition

Description:

Various mathematical frameworks have been explored for visual learning and visual recognition. Special cases of Bayesian modeling and inference, such as deformable templates and compositional vision, lead to the construction of probability measures on complex structures, such as grammars, graphs and spaces of transformations. Difficult mathematical questions arise in learning these and other representations from training data consisting of labeled examples. Recently, tools from statistical inference and information theory have led to new bounds on generalization error as well as new learning algorithms for classification. Most methods for interpreting complex scenes raise formidable computational challenges, inspiring new strategies for visual search.

Workshop Strategy: The workshop will be somewhat unorthodox in the sense of departing from the usual n-talk format in favor of extensive discussion. Aside from one poster session and one session on "Neural Representation and Computation," which consists of talks from several people from the neuroscience community, the activities are either: i) Strategy sessions, each about an approach or set of tools for visual recognition and involving "advocates" and "critics", and each of duration about two hours; or ii) Discussion sessions about more general topics, each assigned several "panelists" and of duration one hour. The advocates will describe the general strategy, or their variation on it, in the manner of a mini-tutorial, concentrating on mature work and the big picture. The critics will evaluate the approach, both the specific variations represented by the advocates as well as the general potential — strengths and weaknesses, inherent limitations, etc., everything that makes for constructive criticism; audience participation is encouraged.

LIST OF CONFIRMED PARTICIPANTS