Dive into Deep LearningDeep learning has revolutionized pattern recognition, introducing tools that power a wide range of technologies in such diverse fields as computer vision, natural language processing, and automatic speech recognition. Applying deep learning requires you to simultaneously understand how to cast a problem, the basic mathematics of modeling, the algorithms for fitting your models to data, and the engineering techniques to implement it all. This book is a comprehensive resource that makes deep learning approachable, while still providing sufficient technical depth to enable engineers, scientists, and students to use deep learning in their own work. No previous background in machine learning or deep learning is required-every concept is explained from scratch and the appendix provides a refresher on the mathematics needed. Runnable code is featured throughout, allowing you to develop your own intuition by putting key ideas into practice. |
Contents
Introduction | 1 |
Preliminaries | 30 |
Linear Neural Networks for Regression | 82 |
Linear Neural Networks for Classification | 126 |
Multilayer Perceptrons | 168 |
Builders Guide | 209 |
Convolutional Neural Networks | 235 |
Modern Convolutional Neural Networks | 270 |
Modern Recurrent Neural Networks | 373 |
Attention Mechanisms and Transformers | 414 |
Tools for Deep Learning | 474 |
Using AWS EC2 Instances | 482 |
Using Google Colab | 490 |
The d21 API Document | 500 |
References | 516 |
| 539 | |
Common terms and phrases
activation function AlexNet algorithms architecture backpropagation batch normalization batch_size block classification CNNs computer vision convolution kernel convolutional layer d21 import torch dataset decoder deep learning deep learning frameworks deep networks def __init__(self def forward(self Defined in Section distribution dot product dropout error estimate example fully connected layer GPUs gradient height and width hidden layer hyperparameters implement input instance label language model loss function machine learning machine translation matrix method minibatch model parameters module multiple neural networks num_heads num_hiddens num_steps numpy operations optimization output channels output layer output shape overfitting performance pixels prediction pretrained probability problem queries random variable ReLU representation ResNet sample save def scalar self-attention self.save_hyperparameters sequence softmax step stochastic gradient descent super().__init__ supervised learning target tensor test set token torch as d21 torch from torch torch import nn training data update valid_lens values vector weights