Fundamentals and Learning of Artificial Neural Networks
Nan Zheng, Pinaki Mazumder · 2019
This chapter presents basic concepts of the rate-based artificial neural networks with the emphasis on how learning is conducted. It discusses important concepts and techniques widely used in deep learning. The chapter introduces the operation in the opposite direction, which runs a backward operation for learning. There are mainly three common types of machine learning methods: supervised learning, reinforcement learning, and unsupervised learning. The chapter reviews the concepts of using neural networks for these three types of learning schemes. It also discusses how the loss functions are constructed for different types of learning mechanisms. The chapter provides a concrete case study to illustrate how a neural network can be employed in a reinforcement-learning application. It explains three major types of network topologies. They are: Fully Connected Neural Networks, Convolutional Neural Networks and Recurrent Neural Networks. The chapter also presents several popular datasets and benchmark tests.