深度学习研究进展
这本书讨论了研究人员最近使用的最先进的深度学习模型。
This book discusses the state-of-the-art deep learning models used by researchers recently.
详细讨论了各种深度架构及其组件。
Various deep architectures and their components are discussed in detail.
应用实例说明了用于训练具有快速收敛的深度体系结构的算法。
Algorithms that are used to train deep architectures with fast convergence rate are illustrated with applications.
讨论了用于优化深度模型的各种微调算法。
Various fine-tuning algorithms are discussed for optimizing the deep models.
这些深度架构不仅能够学习复杂的任务，而且在某些专用的应用程序中甚至可以超越人类。
These deep architectures not only are capable of learning complex tasks but can even outperform humans in some dedicated applications.
尽管在这方面取得了显著的进展，但是训练具有大量超参数的深度结构是一个复杂且不适定的优化问题。
Despite the remarkable advances in this area, training deep architectures with a huge number of hyper-parameters is an intricate and ill-posed optimization problem.
每一章的结尾都概述了各种挑战。
Various challenges are outlined at the end of each chapter.
深度架构的另一个问题是，当大量数据用于训练时，学习是计算密集的。
Another issue with deep architectures is that learning becomes computationally intensive when large volumes of data are used for training.
本书描述了一种迁移学习方法，可以更快地训练深度模型。
The book describes a transfer learning approach for faster training of deep models.
在指纹数据集中演示了这种方法的使用。
The use of this approach is demonstrated in fingerprint datasets.
本书分为八章：
第一章首先介绍机器学习，然后介绍传统机器学习方法的基本局限性。介绍了深度网络，然后简要讨论了为什么要使用深度学习以及深度学习是如何工作的。
The book is organized into eight chapters:
Chapter 1 starts with an introduction to machine learning followed by fundamental limitations of traditional machine learning methods. It introduces deep networks and then briefly discusses why to use deep learning and how deep learning works.
第二章致力于最成功的深度学习技术之一，即卷积神经网络（CNN）。本章的目的是让读者对卷积神经网络结构的各种组成部分进行深入而简单的解释。
Chapter 2 of the book is dedicated to one of the most successful deep learning techniques known as convolutional neural networks (CNNs). The purpose of this chapter is to give its readers an in-depth but easy and uncomplicated explanation of various components of convolutional neural network architectures.
第三章讨论了深度网络的训练和学习过程。本章的目的是为深度学习网络提供一个简单直观的反向传播算法解释，训练过程已经进行了简单明了的解释。
Chapter 3 discusses the training and learning process of deep networks. The aim of this chapter is to provide a simple and intuitive explanation of the backpropagation algorithm for a deep learning network. The training process has been explained step by step with easy and straightforward explanations.
第四章重点介绍了基于CNN的各种深度学习体系结构。它向读者介绍了这些体系结构的框图，讨论了这些深度学习体系结构是如何在解决先前深度学习网络局限性的同时发展起来的。
Chapter 4 focuses on various deep learning architectures that are based on CNN. It introduces a reader to block diagrams of these architectures. It discusses how deep learning architectures have evolved while addressing the limitations of previous deep learning networks.
第五章介绍了各种无监督的深度学习体系结构。概述了属于无监督范畴的体系结构和相关算法的基础知识。
Chapter 5 presents various unsupervised deep learning architectures. The basics of architectures and associated algorithms falling under the unsupervised category are outlined.
第六章讨论了监督式深度学习体系结构在人脸识别问题中的应用。本章比较了有监督深度学习体系结构与传统人脸识别方法的性能。
Chapter 6 discusses the application of supervised deep learning architecture for face recognition problem. A comparison of the performance of supervised deep learning architecture with traditional face recognition methods is provided in this chapter.
第七章重点介绍卷积神经网络（CNN）在指纹识别中的应用。本章详细介绍了CNN的结构和优化提高性能的方法，并对自动指纹识别进行了详细的说明。此外，对深度学习和非深度学习方法进行了比较分析，以说明两者的性能差异。
Chapter 7 focuses on the application of convolutional neural networks (CNNs) for fingerprint recognition. This chapter extensively explains automatic fingerprint recognition with complete details of the CNN architecture and methods used to optimize and enhance the performance. In addition, a comparative analysis of deep learning and non-deep learning methods is presented to show the performance difference.
第八章介绍了如何将无监督深度网络应用于手写数字分类问题。阐述了如何在第一步进行无监督训练，第二步进行监督微调，通过两步构建深度学习模型。
Chapter 8 explains how to apply the unsupervised deep networks to handwritten digit classification problem. It explains how to build a deep learning model in two steps, where unsupervised training is performed during the first step and supervised fine-tuning is carried out during the second step.