Learning feature hierarchies with sparse coding

Monday, October 5, 2009 - 2:00pm - 2:30pm
Lind 305
Yann LeCun (New York University)
Keywords: unsupervised learning, object recognition, sparse coding,
convolutional networks

Abstract:Image processing and recognition has traditionally relied on hard-wired
features and trainable classifiers. The next challenge of computer
vision, machine learning, and image processing, is to devise methods
that can automatically learn feature extractors and high-level image
representations from labeled and unlabeled data. The set of methods
collectively known as Deep Learning is an attempt to learn
hierarchies of features with multiple levels of abstraction, and
suitable invariances. I will describe several deep learning methods,
some of which involve new forms of sparse coding. Specific model
architectures for image recognition, based on stacks on non-linear
filter banks, and trained with these methods will be described. A
number of applications to object dectection, object recognition, and
vision-based navigation for mobile robots will be shown.
MSC Code: