Learning Local Invariant Mahalanobis Distances
This addresses the need for transformation-invariant metrics in machine learning applications like visual recognition, offering a new approach to enhance robustness.
The paper tackles the problem of learning invariant metrics for data transformations, such as rotation and translation in images, by proposing a novel method to learn a local Mahalanobis metric per datum, which improves performance with computational efficiency.
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 insensitive to rotation and translation. This requirement and its implications have been important in many machine learning applications, and tolerance for image transformations was primarily achieved by using robust feature vectors. In this paper we propose a novel and computationally efficient way to learn a local Mahalanobis metric per datum, and show how we can learn a local invariant metric to any transformation in order to improve performance.