O'Reilly logo

Stay ahead with the world's most comprehensive technology and business learning platform.

With Safari, you learn the way you learn best. Get unlimited access to videos, live online training, learning paths, books, tutorials, and more.

Start Free Trial

No credit card required

Fundamentals of Deep Learning

Book Description

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. 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.

Table of Contents

  1. Preface
    1. Prerequisites and Objectives
    2. Conventions Used in This Book
    3. Using Code Examples
    4. Safari® Books Online
    5. How to Contact Us
    6. Acknowledgements
  2. 1. The Neural Network
    1. Building Intelligent Machines
    2. The Limits of Traditional Computer Programs
    3. The Mechanics of Machine Learning
    4. The Neuron
    5. Expressing Linear Perceptrons as Neurons
    6. Feed-Forward Neural Networks
    7. Linear Neurons and Their Limitations
    8. Sigmoid, Tanh, and ReLU Neurons
    9. Softmax Output Layers
    10. Looking Forward
  3. 2. Training Feed-Forward Neural Networks
    1. The Fast-Food Problem
    2. Gradient Descent
    3. The Delta Rule and Learning Rates
    4. Gradient Descent with Sigmoidal Neurons
    5. The Backpropagation Algorithm
    6. Stochastic and Minibatch Gradient Descent
    7. Test Sets, Validation Sets, and Overfitting
    8. Preventing Overfitting in Deep Neural Networks
    9. Summary
  4. 3. Implementing Neural Networks in TensorFlow
    1. What Is TensorFlow?
    2. How Does TensorFlow Compare to Alternatives?
    3. Installing TensorFlow
    4. Creating and Manipulating TensorFlow Variables
    5. TensorFlow Operations
    6. Placeholder Tensors
    7. Sessions in TensorFlow
    8. Navigating Variable Scopes and Sharing Variables
    9. Managing Models over the CPU and GPU
    10. Specifying the Logistic Regression Model in TensorFlow
    11. Logging and Training the Logistic Regression Model
    12. Leveraging TensorBoard to Visualize Computation Graphs and Learning
    13. Building a Multilayer Model for MNIST in TensorFlow
    14. Summary
  5. 4. Beyond Gradient Descent
    1. The Challenges with Gradient Descent
    2. Local Minima in the Error Surfaces of Deep Networks
    3. Model Identifiability
    4. How Pesky Are Spurious Local Minima in Deep Networks?
    5. Flat Regions in the Error Surface
    6. When the Gradient Points in the Wrong Direction
    7. Momentum-Based Optimization
    8. A Brief View of Second-Order Methods
    9. Learning Rate Adaptation
      1. AdaGrad—Accumulating Historical Gradients
      2. RMSProp—Exponentially Weighted Moving Average of Gradients
      3. Adam—Combining Momentum and RMSProp
    10. The Philosophy Behind Optimizer Selection
    11. Summary
  6. 5. Convolutional Neural Networks
    1. Neurons in Human Vision
    2. The Shortcomings of Feature Selection
    3. Vanilla Deep Neural Networks Don’t Scale
    4. Filters and Feature Maps
    5. Full Description of the Convolutional Layer
    6. Max Pooling
    7. Full Architectural Description of Convolution Networks
    8. Closing the Loop on MNIST with Convolutional Networks
    9. Image Preprocessing Pipelines Enable More Robust Models
    10. Accelerating Training with Batch Normalization
    11. Building a Convolutional Network for CIFAR-10
    12. Visualizing Learning in Convolutional Networks
    13. Leveraging Convolutional Filters to Replicate Artistic Styles
    14. Learning Convolutional Filters for Other Problem Domains
    15. Summary
  7. 6. Embedding and Representation Learning
    1. Learning Lower-Dimensional Representations
    2. Principal Component Analysis
    3. Motivating the Autoencoder Architecture
    4. Implementing an Autoencoder in TensorFlow
    5. Denoising to Force Robust Representations
    6. Sparsity in Autoencoders
    7. When Context Is More Informative than the Input Vector
    8. The Word2Vec Framework
    9. Implementing the Skip-Gram Architecture
    10. Summary
  8. 7. Models for Sequence Analysis
    1. Analyzing Variable-Length Inputs
    2. Tackling seq2seq with Neural N-Grams
    3. Implementing a Part-of-Speech Tagger
    4. Dependency Parsing and SyntaxNet
    5. Beam Search and Global Normalization
    6. A Case for Stateful Deep Learning Models
    7. Recurrent Neural Networks
    8. The Challenges with Vanishing Gradients
    9. Long Short-Term Memory (LSTM) Units
    10. TensorFlow Primitives for RNN Models
    11. Implementing a Sentiment Analysis Model
    12. Solving seq2seq Tasks with Recurrent Neural Networks
    13. Augmenting Recurrent Networks with Attention
    14. Dissecting a Neural Translation Network
    15. Summary
  9. 8. Memory Augmented Neural Networks
    1. Neural Turing Machines
    2. Attention-Based Memory Access
    3. NTM Memory Addressing Mechanisms
    4. Differentiable Neural Computers
    5. Interference-Free Writing in DNCs
    6. DNC Memory Reuse
    7. Temporal Linking of DNC Writes
    8. Understanding the DNC Read Head
    9. The DNC Controller Network
    10. Visualizing the DNC in Action
    11. Implementing the DNC in TensorFlow
    12. Teaching a DNC to Read and Comprehend
    13. Summary
  10. 9. Deep Reinforcement Learning
    1. Deep Reinforcement Learning Masters Atari Games
    2. What Is Reinforcement Learning?
    3. Markov Decision Processes (MDP)
      1. Policy
      2. Future Return
      3. Discounted Future Return
    4. Explore Versus Exploit
    5. Policy Versus Value Learning
      1. Policy Learning via Policy Gradients
    6. Pole-Cart with Policy Gradients
      1. OpenAI Gym
      2. Creating an Agent
      3. Building the Model and Optimizer
      4. Sampling Actions
      5. Keeping Track of History
      6. Policy Gradient Main Function
      7. PGAgent Performance on Pole-Cart
    7. Q-Learning and Deep Q-Networks
      1. The Bellman Equation
      2. Issues with Value Iteration
      3. Approximating the Q-Function
      4. Deep Q-Network (DQN)
      5. Training DQN
      6. Learning Stability
      7. Target Q-Network
      8. Experience Replay
      9. From Q-Function to Policy
      10. DQN and the Markov Assumption
      11. DQN’s Solution to the Markov Assumption
      12. Playing Breakout wth DQN
      13. Building Our Architecture
      14. Stacking Frames
      15. Setting Up Training Operations
      16. Updating Our Target Q-Network
      17. Implementing Experience Replay
      18. DQN Main Loop
      19. DQNAgent Results on Breakout
    8. Improving and Moving Beyond DQN
      1. Deep Recurrent Q-Networks (DRQN)
      2. Asynchronous Advantage Actor-Critic Agent (A3C)
      3. UNsupervised REinforcement and Auxiliary Learning (UNREAL)
    9. Summary
  11. Index