Date of Publication :17th October 2017

Abstract: Deep learning refers to artificial neural networks comprising of multiple layers. The deep learning algorithms automate the representation of abstract features by composing simple representations from raw data at one level to complex representations at the higher. These algorithms have improved the current state of art results and bought new insights to the current data. In this paper, the deep neural architectures and their application on NLP have been discussed. The paper also evaluates the various approaches to train the data. Various classifications of deep neural nets are also an integral part of the paper. On the basis of architecture and transfer of information from input to output layers via hidden layers, deep neural nets have been broadly classified and elaborated. Brief comparisons of the various techniques used in deep neural networks on various parameters are evaluated and have been presented.

Reference :