Similarity learning in Internet vision

The biggest challenge in Internet vision stems from the heterogeneity,
complexity, and scale of the data. Internet repositories are usually
created by different people, in different time, at different places,
resulting in great data variability and richness. Trying to model such
data or using standard metrics to measure similarity is usually
impossible. In this talk, I will present a generic
framework for supervised similarity learning based on embedding the
data into a representation metric space. In particular, for binary
similarity, the embedding can be formulates as a binary classification
problem with positive and negative examples, and can be efficiently
learned using boosting algorithms. The utility and efficiency of the
presented approach will be demonstrated on several challenging
Internet vision applications, including video, image, and shape
retrieval.