A Probabilistic Framework for Discriminative Dictionary Learning

A Probabilistic Framework for Discriminative Dictionary Learning

​Bernard Ghanem and Narendra Ahuja
"A Probabilistic Framework for Discriminative Dictionary Learning"
(arXiv e-print 2011  arXiv:1109.2389v1 [cs.CV])
Bernard Ghanem and Narendra Ahuja
Pobabilistic Framework, Dictionary Learning
2011
​In this paper, we address the problem of discriminative dictionary learning (DDL),  where sparse linear representation and classification are combined in a probabilistic  framework. As such, a single discriminative dictionary and linear binary classifiers  are learned jointly. By encoding sparse representation and discriminative  classification models in a MAP setting, we propose a general optimization framework  that allows for a data-driven tradeoff between faithful representation and  accurate classification. As opposed to previous work, our learning methodology  is capable of incorporating a diverse family of classification cost functions (including  those used in popular boosting methods), while avoiding the need for involved  optimization techniques. We show that DDL can be solved by a sequence of updates  that make use of well-known and well-studied sparse coding and dictionary  learning algorithms from the literature. To validate our DDL framework, we apply  it to digit classification and face recognition and test it on standard benchmarks.