PROJECT TITLE :

A General Approach for Achieving Supervised Subspace Learning in Sparse Representation

ABSTRACT:

A vast family of subspace learning algorithms based on dictionary learning has been developed during the last few decades to give alternative solutions for learning subspace features. The majority of them are unsupervised algorithms that are used on data with no labels. It's worth mentioning that label information is accessible in some applications, such as facial recognition, where the dimensionality reduction approaches stated above can't use it to increase their performance. To increase performance in certain labeled cases, an unsupervised subspace learning method must be transformed into the appropriate supervised approach. We offer a technique in this research that may be utilized as a general method for constructing a supervised algorithm based on any unsupervised subspace learning algorithm that uses sparse representation. We also create a new supervised subspace learning algorithm called supervised principal coefficients embedding using the proposed method (SPCE). We demonstrate that SPCE outperforms the state-of-the-art supervised subspace learning algorithm.