Efficient dual approach to distance metric learning.

Distance metric learning is of fundamental interest in machine learning because the employed distance metric can significantly affect the performance of many learning methods. Quadratic Mahalanobis metric learning is a popular approach to the problem, but typically requires solving a semidefinite programming (SDP) problem, which is computationally expensive. The worst case complexity of solving an SDP problem involving a matrix variable of size D×D with O(D) linear constraints is about O(D(6.

An face-based visual fixation system for prosthetic vision.

Recent studies have shown the success of face recognition using low resolution prosthetic vision, but it requires a zoomed-in and stably-fixated view, which will be challenging for a user with the limited resolution of current prosthetic vision devices. We propose a real-time object detection and tracking system capable of fixating human faces. By integrating both static and temporal information, we are able to improve the robustness of face localization so that it can fixate on faces with large pose variations.

The role of vision processing in prosthetic vision.

Prosthetic vision provides vision which is reduced in resolution and dynamic range compared to normal human vision. This comes about both due to residual damage to the visual system from the condition that caused vision loss, and due to limitations of current technology. However, even with limitations, prosthetic vision may still be able to support functional performance which is sufficient for tasks which are key to restoring independent living and quality of life.

Scalable large-margin Mahalanobis distance metric learning.

For many machine learning algorithms such as k-nearest neighbor ( k-NN) classifiers and k-means clustering, often their success heavily depends on the metric used to calculate distances between different data points. An effective solution for defining such a metric is to learn it from a set of labeled training samples. In this work, we propose a fast and scalable algorithm to learn a Mahalanobis distance metric.