Data Mining, 4th Edition

Chapter 10

Deep learning

Abstract

One of the most striking contemporary developments in the field of machine learning is the meteoric ascent of what has been called “deep learning,” and this area is now at the forefront of current research. We discuss key differences between traditional neural network architectures and learning techniques, and those that have become popular in deep learning. A detailed derivation of the backpropagation algorithm in vector-matrix form is provided, and the relationship to computational graphs and deep learning software is discussed. Deep convolutional neural networks are covered, as well as autoencoders, recurrent neural networks, and stochastic approaches based on Boltzmann machines. Key practical aspects of training ...