O simbolo perdido

A project-based guide to the basics of deep learning. This concise, project-driven guide to deep learning takes readers through a series of program-writing tasks that introduce them to the use of deep learning in such areas of artificial intelligence as computer vision, natural-language processing, and reinforcement learning. The author, a longtime artificial intelligence researcher specializing in natural-language processing, covers feed-forward neural nets, convolutional neural nets, word embeddings, recurrent neural nets, sequence-to-sequence learning, deep reinforcement learning, unsupervised models, and other fundamental concepts and techniques. Students and practitioners learn the basics of deep learning by working through programs in Tensorflow, an open-source machine learning framework. “I find I learn computer science material best by sitting down and writing programs,” the author writes, and the book reflects this approach. Each chapter includes a programming project, exercises, and references for further reading. An early chapter is devoted to Tensorflow and its interface with Python, the widely used programming language. Familiarity with linear algebra, multivariate calculus, and probability and statistics is required, as is a rudimentary knowledge of programming in Python. The book can be used in both undergraduate and graduate courses; practitioners will find it an essential reference. Contents 8 Preface 12 1. Feed-Forward Neural Nets 14 1.1 Perceptrons 16 1.2 Cross-entropy Loss Functions for Neural Nets 22 1.3 Derivatives and Stochastic Gradient Descent 27 1.4 Writing Our Program 31 1.5 Matrix Representation of Neural Nets 34 1.6 Data Independence 37 1.7 References and Further Readings 38 1.8 Written Exercises 39 2. Tensorflow 42 2.1 Tensorflow Preliminaries 42 2.2 A TF Program 46 2.3 Multilayered NNs 51 2.4 Other Pieces 55 2.4.1 Checkpointing 55 2.4.2 tensordot 56 2.4.3 Initialization of TF Variables 57 2.4.4 Simplifying TF Graph Creation 60 2.5 References and Further Readings 61 2.6 Written Exercises 62 3. Convolutional Neural Networks 64 3.1 Filters, Strides, and Padding 65 3.2 A Simple TF Convolution Example 70 3.3 Multilevel Convolution 74 3.4 Convolution Details 77 3.4.1 Biases 77 3.4.2 Layers with Convolution 78 3.4.3 Pooling 79 3.5 References and Further Readings 80 3.6 Written Exercises 81 4. Word Embeddings and Recurrent NNs 84 4.1 Word Embeddings for Language Models 84 4.2 Building Feed-Forward Language Models 89 4.3 Improving Feed-Forward Language Models 91 4.4 Overfitting 92 4.5 Recurrent Networks 95 4.6 Long Short-Term Memory 101 4.7 References and Further Readings 105 4.8 Written Exercises 105 5. Sequence-to-Sequence Learning 108 5.1 The Seq2Seq Paradigm 109 5.2 Writing a Seq2Seq MT program 112 5.3 Attention in Seq2seq 115 5.4 Multilength Seq2Seq 120 5.5 Programming Exercise 121 5.6 Written Exercises 123 5.7 References and Further Readings 124 6. Deep Reinforcement Learning 126 6.1 Value Iteration 127 6.2 Q-learning 130 6.3 Basic Deep-Q Learning 132 6.4 Policy Gradient Methods 137 6.5 Actor-Critic Methods 143 6.6 Experience Replay 146 6.7 References and Further Readings 147 6.8 Written Exercises 147 7. Unsupervised Neural-Network Models 150 7.1 Basic Autoencoding 150 7.2 Convolutional Autoencoding 153 7.3 Variational Autoencoding 157 7.4 Generative Adversarial Networks 165 7.5 References and Further Readings 170 7.6 Written Exercises 170 A. Answers to Selected Exercises 172 A.1 Chapter 1 172 A.2 Chapter 2 173 A.3 Chapter 3 173 A.4 Chapter 4 174 A.5 Chapter 5 174 A.6 Chapter 6 175 A.7 Chapter 7 175 Bibliography 178 Index 182 "This concise, project-driven guide to deep learning takes readers through a series of program-writing tasks that introduce them to the use of deep learning in such areas of artificial intelligence as computer vision, natural-language processing, and reinforcement learning. The author, a longtime artificial-intelligence researcher specializing in natural-language processing, covers feed-forward neural nets, convolutional neural nets, word embeddings, recurrent neural nets, sequence-to-sequence learning, deep reinforcement learning, unsupervised models, and other fundamental concepts and techniques."--Page 4 de la couverture