Event Location: NSH 3305

Abstract: One of the goals of the Willow research team is the
cross-pollination of computer vision and machine learning research. I
will present in this talk a brief overview of Willow’s activities,
then focus on two recent examples of this approach. Sparse
coding—that is, modelling data vectors as sparse linear combinations
of dictionary elements—is widely used in machine learning,
neuroscience, signal processing, and statistics. The first part of
this talk addresses the problem of learning the dictionary, adapting
it to specific data and image understanding tasks. In particular, I
will present a fast on-line approach to dictionary learning and more
generally sparse matrix factorization, and demonstrate applications of
sparse coding to pixelwise image classification, edge detection, and
image denoising and demosaicking. In the second part of the talk, I
will formulate the temporal localization of human actions in video
streams as a kernel-based discriminative clustering problem in
weakly-labelled training data. Combining this approach with
script-based annotation of video segments provides a fully automated
method for learning action models and detecting instances of these
models in movies.

Joint work with Francis Bach, Olivier Duchenne, Ivan Laptev, Julien
Mairal, Guillermo Sapiro, Josef Sivic, Oliver Whyte, and Andrew
Zisserman.