Deep Learning

Abstract

Deep learning is a branch of machine learning that tries to model high-level abstractions of data using multiple layers of neurons consisting of complex structures or non-liner transformations. With the increase of the amount of data and the power of computation, neural networks with more complex structures have attracted widespread attention and been applied to various fields. This paper provides an overview of deep learning in neural networks including popular architecture models and training algorithms.

Machine learning in materials science

“…Figure B,C illustrates operations of the convolutional layer and the max pooling layer. Sometimes, other types of layers, such as the drop‐out layer, are used to control the size of the CNN …”

“…Copyright 2016, New Jersey: World Scientific Pub. Co., 2007 . C. A schematic diagram of max a pooling layer.…”

“…C. A schematic diagram of max a pooling layer. The pooling layer divides the input into four regions and outputs the maximum value of each region. Copyright 2016, New Jersey: World Scientific Pub.…”

“…As there are no connections between neurons in the same layer of the CNN model, data flow from the input layer to the hidden layer and, finally, are output by the output layer. Therefore, it is difficult for a CNN to process data that are related . Hence, a RNN should be used to process sequential data.…”

“…Network state S i is often regarded as a memory unit of the hidden layer that stores the output from the previous step and updates the current input of a nonlinear function each time. Therefore, an RNN requires less parameter learning than CNN …”