Fundamentals of Deep Learning

Tensor Init
Tensor Attributes
Tensor Operations
Gradients in PyTorch
The PyTorch nn Module
PyTorch Datasets and Dataloaders
Building the MNIST Classifier in PyTorch
Summary

6. Beyond Gradient Descent
The Challenges with Gradient Descent
Local Minima in the Error Surfaces of Deep Networks
Model Identifiability
How Pesky Are Spurious Local Minima in Deep Networks?
Flat Regions in the Error Surface
When the Gradient Points in the Wrong Direction
Momentum-Based Optimization
A Brief View of Second-Order Methods
Learning Rate Adaptation
AdaGrad—Accumulating Historical Gradients
RMSProp—Exponentially Weighted Moving Average of Gradients
Adam—Combining Momentum and RMSProp
The Philosophy Behind Optimizer Selection
Summary

7. Convolutional Neural Networks
Neurons in Human Vision
The Shortcomings of Feature Selection
Vanilla Deep Neural Networks Don’t Scale
Filters and Feature Maps
Full Description of the Convolutional Layer
Max Pooling
Full Architectural Description of Convolution Networks
Closing the Loop on MNIST with Convolutional Networks
Image Preprocessing Pipelines Enable More Robust Models
Accelerating Training with Batch Normalization
Group Normalization for Memory Constrained Learning Tasks
Building a Convolutional Network for CIFAR-10
Visualizing Learning in Convolutional Networks
Residual Learning and Skip Connections for Very Deep Networks
Building a Residual Network with Superhuman Vision
Leveraging Convolutional Filters to Replicate Artistic Styles
Learning Convolutional Filters for Other Problem Domains
Summary

8. Embedding and Representation Learning
Learning Lower-Dimensional Representations
Principal Component Analysis
Motivating the Autoencoder Architecture
Implementing an Autoencoder in PyTorch
Denoising to Force Robust Representations
Sparsity in Autoencoders
When Context Is More Informative than the Input Vector
The Word2Vec Framework
Implementing the Skip-Gram Architecture
Summary

9. Models for Sequence Analysis
Analyzing Variable-Length Inputs
Tackling seq2seq with Neural N-Grams
Implementing a Part-of-Speech Tagger
Dependency Parsing and SyntaxNet
Beam Search and Global Normalization
A Case for Stateful Deep Learning Models
Recurrent Neural Networks
The Challenges with Vanishing Gradients
Long Short-Term Memory Units
PyTorch Primitives for RNN Models
Implementing a Sentiment Analysis Model
Solving seq2seq Tasks with Recurrent Neural Networks
Augmenting Recurrent Networks with Attention
Dissecting a Neural Translation Network
Self-Attention and Transformers
Summary

10. Generative Models
Generative Adversarial Networks
Variational Autoencoders
Implementing a VAE
Score-Based Generative Models
Denoising Autoencoders and Score Matching
Summary

11. Methods in Interpretability
Overview
Decision Trees and Tree-Based Algorithms
Linear Regression
Methods for Evaluating Feature Importance
Permutation Feature Importance
Partial Dependence Plots
Extractive Rationalization
LIME
SHAP
Summary

12. Memory Augmented Neural Networks
Neural Turing Machines
Attention-Based Memory Access
NTM Memory Addressing Mechanisms
Differentiable Neural Computers
Interference-Free Writing in DNCs
DNC Memory Reuse
Temporal Linking of DNC Writes
Understanding the DNC Read Head
The DNC Controller Network
Visualizing the DNC in Action
Implementing the DNC in PyTorch
Teaching a DNC to Read and Comprehend
Summary

13. Deep Reinforcement Learning
Deep Reinforcement Learning Masters Atari Games
What Is Reinforcement Learning?
Markov Decision Processes
Policy
Future Return
Discounted Future Return
Explore Versus Exploit
ϵ -Greedy
Annealed ϵ -Greedy
Policy Versus Value Learning
Pole-Cart with Policy Gradients
OpenAI Gym
Creating an Agent
Building the Model and Optimizer
Sampling Actions
Keeping Track of History
Policy Gradient Main Function
PGAgent Performance on Pole-Cart
Trust-Region Policy Optimization
Proximal Policy Optimization
Q-Learning and Deep Q-Networks
The Bellman Equation
Issues with Value Iteration
Approximating the Q-Function
Deep Q-Network
Training DQN
Learning Stability
Target Q-Network
Experience Replay
From Q-Function to Policy
DQN and the Markov Assumption
DQN’s Solution to the Markov Assumption
Playing Breakout with DQN
Building Our Architecture
Stacking Frames
Setting Up Training Operations
Updating Our Target Q-Network
Implementing Experience Replay
DQN Main Loop
DQNAgent Results on Breakout
Improving and Moving Beyond DQN
Deep Recurrent Q-Networks
Asynchronous Advantage Actor-Critic Agent
UNsupervised REinforcement and Auxiliary Learning
Summary

Index
About the Authors