Neural Network Introduction

To enable computers to master knowledge like human beings, a multi-layer connection network needs to be constructed to define a complex object. After iterative computing and training of the network, it can extract object features. Generally, this method is called deep learning. After a series of development, deep learning has displayed its tremendous application value and is receiving increasing attentions from the industry and academia. Deep learning has achieved remarkable progresses in image, voice, natural language processing, big data feature extraction, and ad click-through rate estimation. As a result, multiple infrastructures, such as Caffe, MXNet, and TensorFlow, have been developed to promote deep learning across fields.

Deep neural network research fuels rapid development of neural network models, enabling them to complete more and more complex processing tasks in a wider range of fields. With the rapid development of semiconductor chips and computer technologies for decades, ever fast and energy-efficient computing resources have been provided for neural network models and data, such as CPUs, GPUs, TPUs, and the latest Ascend AI Processor launched by Huawei.

Artificial neural network (ANN) may also be referred to as neural network (NN) for short, which is an important branch of machine learning (ML). Scientists perform mathematical modeling on the most basic neurons and build artificial neural networks based on the certain hierarchical relationship of neurons, enabling artificial neural networks to learn knowledge and adjust their internal structures through learning and training and thereby achieve various complex computations.