Deep Learning Architectures: A Hierarchy in Convolution Neural Network Technologies

Albeit deep learning has chronicled roots and has been applied to computer vision task since 2000 but a decade ago, neither the expression “Deep learning” nor the methodology was well known. This dormant field regains consciousness when a highly influential paper “Image Net Classification with Deep Convolutional Neural Networks by Krizhevsky, Sutskever and Hinton’in 2012” was published. Now, availability of abundance of data, computational power, and improved algorithms has contributed altogether and brought this technology to forefront in the field of machine learning. In this paper, we focus on growth of various convolution neural network architectures (deep learning architectures), from their predecessors up to recent state-of-the-art deep learning systems. The paper has three sections: (1) Introduction about neural networks along with necessary back ground information. (2) Hierarchy of classical and modern architectures; In this section, the existing methods are explained and their contribution and significance in field of machine learning are highlighted. At last, we point out a set of promising future works and draw our own conclusions.