Abstract:

Today’s databases are growing rapidly. For example, YouTube uploads an average of 300 hours of video per minute. In proportion to the data size, processing, storage and transfer costs are increasing. On the contrary, it is known that high-dimensional data content, especially video and image, is largely similar. This type of high-dimensional raw data is vital for low-dimensional reduction, image classification, detection and meaningful information extraction processes. There are many techniques available to reduce the data size. From classical artificial learning techniques; PCA (Basic Components Analysis) and LDA (Natural Separation Analysis), the researchers are interested in providing a mathematical solution ground, while from nonlinear techniques, from deep learning approaches, the Auto-Encoding (Auto-Encoding) is interested in allowing big data to be reduced. In this study, we studied the scale reduction performance of PCA, LDA and Auto-Encoding (AE) methods using real and synthetic data (oral and non-linear). In certain shortspaces (the time spent, the correctness of rebuilding, etc.) The results are comparable.

Keywords: