A brief survey of dimension reduction

Dimension reduction problem is a big concern which can reduce the scale of a database and keep the main features of these data simultaneously. This paper aims at reviewing and comparing different dimension reduction algorithms. Mainly, the performances of four basic algorithms (PCA, LDA, LLE and LE), their improved methods and deep learning methods are compared by reviewing the previous work. Their recognition accuracy and running time are carefully analyzed. We conclude that PCA and LDA are used more frequently in related fields. Combined methods usually perform better than original methods. Besides, deep learning method is also an approach developed in recent years, which outperforms existing traditional algorithms, though there are many barriers at present, such as obtaining huge labeled database, the computing and power limitation of different systems etc. Future research should focus on the processing of larger database. Finally, some new applications of dimension reduction are reviewed.