Learning Local Invariant Mahalanobis Distances

Ethan Fetaya, Shimon Ullman

For many tasks and data types, there are natural transformations to which the data should be invariant or insensitive. For instance, in visual recognition, natural images should be…