PROJECT TITLE :

Realization of Spatial Sparseness by Deep ReLU Nets With Massive Data

ABSTRACT:

The tremendous success of Deep Learning presents significant challenges that must be addressed immediately if we are to understand the working mechanism and rationale behind it. It is widely acknowledged that the depth of the data, the structure of the data, and massive amounts of data are three essential components for deep learning. The majority of the most recent theoretical studies for deep learning concentrate on the need for and the benefits provided by deeper and more complex neural network structures. In this piece, we will attempt to provide rigorous verification of the significance of massive data sets in bringing to life the superior performance of deep learning. In particular, we demonstrate that the massiveness of the data is required for the realization of the spatial sparseness, and Deep Neural Networks are essential tools for making full use of massive data in an application like this one. Even though deep Neural Networks and numerous other network structures have been proposed for at least the past 20 years, all of these findings present the reasons why deep learning achieves such great success in the era of Big Data.