March 3, 2017

9:30 am / 10:30 am

Venue

Clark 314

Talk 1: Up-Scaling Dictionary Learning 9:30 am ? 10:30 am

Sparse approximation and dictionary learning have been applied with great success to various image processing tasks, often leading to state-of-the-art results. Yet, these methods have traditionally been restricted to small dimensions due to the computational constraints that these problems entail.In this talk, I will first give a brief introduction to the topic of dictionary learning, describing the basics of typical algorithms and reviewingsome of their results and limitations. In order to go beyond the small patches in sparsity-based signal and image processing, I will then present a recent work demonstrating how to efficiently handle larger dimensions. Thiswork employs a new cropped wavelets dictionary, which enables a multi-scale decomposition with virtually no border effects. We then employ this dictionary within a double sparsity model while leveraging the scaling properties of online learning. The resulting large trainable atoms, coined trainlets, not only achieve state-of-the-art performance in dictionary learning,but also open the door to new challenges and problems that remained unattainable until now.