11. S. Lazebnik and M. Raginsky: Learning Nearest Neighbor Quantizers from Labeled Data by Information Loss Minimization
This poster presents an information-theoretic clustering
method that produces discriminative quantized representations
of continuous data that is also supplied with class labels.
This method works by learning a set of prototypes in the feature
space such that the index of the nearest prototype of a given
feature vector approximates a sufficient statistic for its class
label. We have applied the method to the application of producing
visual vocabularies for bag-of-features image classification, and
our experiments demonstrate that vocabularies learned by the proposed
method have higher classification performance than standard
vocabularies learned using k-means.