Unsupervised Transfer Learning For Face Recognition

A major assumption in many machine learning and pattern recognition algorithms is that the training and future data must be in the same feature space and have the same distribution. However, in many realworld applications, this assumption may not hold. In recent years, transfer learning has emerged as a new learning framework to handle this problem.In this paper we examine unsupervised transfer learning methods into two main categories: (a) unsupervised transfer learning based on calculating A,B,V matrices asynchronously; (b) unsupervised transfer learning based on calculating A,B,V matrices synchronously. We study methods without transition matrix or with transition matrix or kernel. We apply four kind of matrix: random matrix, matrix of eigen vectors with one largest eigen value and matrix of eigen vectors with two largest eigen values and matrix of eigen vectors with all largest nonzero eigen values. Also we investigate kernel method. Two kind of inputs (ordinary and Gabor filter) are used. Several properties of the source domain, such as background and the size of subjects, play an important role in determining the final clustering results. Experiments in our different algorithms show that accuracy improve when we use kernel method withtransition matrix of eigen vectors with two largest eigen values and use Gabor filter to extract features of input data.

Keywords:

Unsupervised Transfer Learning , Face recognition , image representation