상세정보

With the reinvigoration of neural networks in the 2000s, deep learning has become an extremely active area of research, one that's paving the way for modern machine learning. In this practical book, author Nikhil Buduma provides examples and clear explanations to guide you through major concepts of this complicated field.

Companies such as Google, Microsoft, and Facebook are actively growing in-house deep-learning teams. For the rest of us, however, deep learning is still a pretty complex and difficult subject to grasp. If you're familiar with Python, and have a background in calculus, along with a basic understanding of machine learning, this book will get you started.

• Examine the foundations of machine learning and neural networks
• Learn how to train feed-forward neural networks
• Use TensorFlow to implement your first neural network
• Manage problems that arise as you begin to make networks deeper
• Build neural networks that analyze complex images
• Perform effective dimensionality reduction using autoencoders
• Dive deep into sequence analysis to examine language
• Understand the fundamentals of reinforcement learning

정보제공 :

Section	Section Description	Page Number
Preface	p. ix
1	The Neural Network	p. 1
    Building Intelligent Machines	p. 1
    The Limits of Traditional Computer Programs	p. 2
    The Mechanics of Machine Learning	p. 3
    The Neuron	p. 7
    Expressing Linear Perceptrons as Neurons	p. 8
    Feed-Forward Neural Networks	p. 9
    Linear Neurons and Their Limitations	p. 12
    Sigmoid, Tanh, and ReLU Neurons	p. 13
    Softmax Output Layers	p. 15
    Looking Forward	p. 15
2	Training Feed-Forward Neural Networks	p. 17
    The Fast-Food Problem	p. 17
    Gradient Descent	p. 19
    The Delta Rule and Learning Rates	p. 21
    Gradient Descent with Sigmoidal Neurons	p. 22
    The Backpropagation Algorithm	p. 23
    Stochastic and Minibatch Gradient Descent	p. 25
    Test Sets, Validation Sets, and Overfitting	p. 27
    Preventing Overfitting in Deep Neural Networks	p. 34
    Summary	p. 37
3	Implementing Neural Networks in TensorFlow	p. 39
    What Is TensorFlow?	p. 39
    How Does TensorFlow Compare to Alternatives?	p. 40
    Installing TensorFlow	p. 41
    Creating and Manipulating TensorFlow Variables	p. 43
    TensorFlow Operations	p. 45
    Placeholder Tensors	p. 45
    Sessions in TensorFlow	p. 45
    Navigating Variable Scopes and Sharing Variables	p. 48
    Managing Models over the CPU and GPU	p. 51
    Specifying the Logistic Regression Model in TensorFlow	p. 52
    Logging and Training the Logistic Regression Model	p. 55
    Leveraging Tensor Board to Visualize Computation Graphs and Learning	p. 58
    Building a Multilayer Model for MNIST in TensorFlow	p. 59
    Summary	p. 62
4	Beyond Gradient Descent	p. 63
    The Challenges with Gradient Descent	p. 63
    Local Minima in the Error Surfaces of Deep Networks	p. 64
    Model Identifiability	p. 65
    How Pesky Are Spurious Local Minima in Deep Networks?	p. 66
    Flat Regions in the Error Surface	p. 69
    When the Gradient Points in the Wrong Direction	p. 71
    Momentum-Based Optimization	p. 74
    A Brief View of Second-Order Methods	p. 77
    Learning Rate Adaptation	p. 78
    AdaGrad-Accumulating Historical Gradients	p. 79
    RMSProp-Exponentially Weighted Moving Average of Gradients	p. 80
    Adam-Combining Momentum and RMSProp	p. 81
    The Philosophy Behind Optimizer Selection	p. 83
    Summary	p. 83
5	Convolutional Neural Networks	p. 85
    Neurons in Human Vision	p. 85
    The Shortcomings of Feature Selection	p. 86
    Vanilla Deep Neural Networks Don''t Scale	p. 89
    Filters and Feature Maps	p. 90
    Full Description of the Convolutional Layer	p. 95
    Max Pooling	p. 98
    Full Architectural Description of Convolution Networks	p. 99
    Closing the Loop on MNIST with Convolutional Networks	p. 101
    Image Preprocessing Pipelines Enable More Robust Models	p. 103
    Accelerating Training with Batch Normalization	p. 104
    Building a Convolutional Network for CIFAR-10	p. 107
    Visualizing Learning in Convolutional Networks	p. 109
    Leveraging Convolutional Filters to Replicate Artistic Styles	p. 113
    Learning Convolutional Filters for Other Problem Domains	p. 114
    Summary	p. 115
6	Embedding and Representation Learning	p. 117
    Learning Lower-Dimensional Representations	p. 117
    Principal Component Analysis	p. 118
    Motivating the Autoencoder Architecture	p. 120
    Implementing an Autoencoder in TensorFlow	p. 121
    Denoising to Force Robust Representations	p. 134
    Sparsity in Autoencoders	p. 137
    When Context Is More Informative than the Input Vector	p. 140
    The Word2Vec Framework	p. 143
    Implementing the Skip-Gram Architecture	p. 146
    Summary	p. 152
7	Models for Sequence Analysis	p. 153
    Analyzing Variable-Length Inputs	p. 153
    Tackling seq2seq with Neural N-Grams	p. 155
    Implementing a Part-of-Speech Tagger	p. 156
    Dependency Parsing and SyntaxNet	p. 164
    Beam Search and Global Normalization	p. 168
    A Case for Stateful Deep Learning Models	p. 172
    Recurrent Neural Networks	p. 173
    The Challenges with Vanishing Gradients	p. 176
    Long Short-Term Memory (LSTM) Units	p. 178
    TensorFlow Primitives for RNN Models	p. 183
    Implementing a Sentiment Analysis Model	p. 185
    Solving seq2seq Tasks with Recurrent Neural Networks	p. 189
    Augmenting Recurrent Networks with Attention	p. 191
    Dissecting a Neural Translation Network	p. 194
    Summary	p. 217
8	Memory Augmented Neural Networks	p. 219
    Neural Turing Machines	p. 219
    Attention-Based Memory Access	p. 221
    NTM Memory Addressing Mechanisms	p. 223
    Differentiable Neural Computers	p. 226
    Interference-Free Writing in DNCs	p. 229
    DNC Memory Reuse	p. 230
    Temporal Linking of DNC Writes	p. 231
    Understanding the DNC Read Head	p. 232
    The DNC Controller Network	p. 232
    Visualizing the DNC in Action	p. 234
    Implementing the DNC in TensorFlow	p. 237
    Teaching a DNC to Read and Comprehend	p. 242
    Summary	p. 244
9	Deep Reinforcement Learning	p. 245
    Deep Reinforcement Learning Masters Atari Games	p. 245
    What Is Reinforcement Learning?	p. 247
    Markov Decision Processes (MDP)	p. 248
        Policy	p. 249
        Future Return	p. 250
        Discounted Future Return	p. 251
    Explore Versus Exploit	p. 251
    Policy Versus Value Learning	p. 253
        Policy Learning via Policy Gradients	p. 254
    Pole-Cart with Policy Gradients	p. 254
        OpenAI Gym	p. 254
        Creating an Agent	p. 255
        Building the Model and Optimizer	p. 257
        Sampling Actions	p. 257
        Keeping Track of History	p. 257
        Policy Gradient Main Function	p. 258
        PGAgent Performance on Pole-Cart	p. 260
    Q-Learning and Deep Q-Networks	p. 261
        The Bellman Equation	p. 261
        Issues with Value Iteration	p. 262
        Approximating the Q-Function	p. 262
        Deep Q-Network (DQN)	p. 263
        Training DQN	p. 263
        Learning Stability	p. 263
        Target Q-Network	p. 264
        Experience Replay	p. 264
        From Q-Function to Policy	p. 264
        DQN and the Markov Assumption	p. 265
        DQN''s Solution to the Markov Assumption	p. 265
        Playing Breakout wth DQN	p. 265
        Building Our Architecture	p. 268
        Stacking Frames	p. 268
        Setting Lip Training Operations	p. 268
        Updating Our Target Q-Network	p. 269
        Implementing Experience Replay	p. 269
        DQN Main Loop	p. 270
        DQNAgent Results on Breakout	p. 272
    Improving and Moving Beyond DQN	p. 273
        Deep Recurrent Q-Networks (DRQN)	p. 273
        Asynchronous Advantage Actor-Critic Agent (A3C)	p. 274
        Unsupervised REinforcement and Auxiliary Learning (UNREAL)	p. 275
    Summary	p. 276
Index	p. 277