April 16: Koray Kavukcuoglu, CIMS
Learning Hierarchies of Features for Object Detection
Object recognition generally involves two phases, feature extraction
and classification. In this talk I will present unsupervised learning
algorithms which are derived from sparse coding methods, have
efficient inference characteristics and that can be used as generic
feature extraction systems. I will also present an efficient
hierarchical object recognition system that combines unsupervised and
supervised training. This is a joint work with Marc'Aurelio Ranzato,
Kevin Jarret, Rob Fergus and Yann LeCun.
|