Heart of Machine reporting, participation: Li Zannan, Zhang Qian.

The best way to get this physical book is to become a contributor.

Hands-on Deep Learning, a book by Aston Zhang, Li Mu and others, has been released online for free reading. This is an interactive deep learning book for school students, engineers, and researchers.

  • Online book address: zh.diveintodeeplearning.org/index.html

  • GitHub Project: github.com/diveintodee…

This book is an important part of Amazon’s MXNet Zero-based Deep Learning course. Gluon, the front-end tool of Apache MXNet, is recommended for development. It will help you learn how to write production-grade applications using simple and easy-to-read code during the hands-on process.

It is worth mentioning that the book is presented in the form of Jupyter Notepad, where readers can manipulate the code and hyperparameters to get timely feedback and improve learning efficiency.

contributors

Contributors to the book include several scientists working at Amazon:

Mu Li: Chief scientist of Amazon, PhD, Department of Computer Science, Carnegie Mellon University.

Aston Zhang is an Amazon Applied Scientist with a PhD in computer science from the University of Illinois at Urbana-Champaign.

Zachary C. Lipton is an Applied scientist at Amazon, assistant professor at Carnegie Mellon University, and PhD in computer science at the University of California, San Diego.

Alexander J. Smola: Director of Amazon ML, PhD in Computer Science, Technical University of Berlin, Germany.

In addition, the book has over 100 contributors in the open source community. The authors say the online book “project” is still evolving, and contributors will receive exclusive editions and be thanked.

Interactive: Jupyter Notepad + active community support

Each section is a working Jupyter Notepad, and you’re free to modify the code and hyperparameters to get immediate feedback, thus gaining hands-on experience in deep learning.

  • Jupyter notepad download address: zh.diveintodeeplearning.org/d2l-zh.zip

https://v.qq.com/x/page/b13531to52r.html

The book also has active community support, with links at the end of each chapter to discuss learning with thousands of other members of the community.

https://v.qq.com/x/page/s13530z3c2m.html

Structure: formula + diagram + code

This book combines text, formulas, and diagrams to illustrate common models and algorithms in deep learning. It also provides code to demonstrate how to implement them from scratch, and uses real data to provide an interactive learning experience.

These three presentation methods complement each other. Many algorithms can use diagrams to deepen understanding of structures, while algorithms such as LSTM, shown in the figure above, require formulas to understand specific structures. In addition, neither expressions nor legends can contain complete detail, and many details cannot be shown without code.

directory

The introduction

  • preface

  • Introduction to Deep Learning

  • How to Use this book

Preliminary knowledge

  • Get and run this book code

  • Data manipulation

  • Automatic gradient finding

  • Consult the MXNet documentation

Fundamentals of deep learning

  • Linear regression

  • Linear regression is implemented from zero

  • Gluon implementation of linear regression

  • Softmax regression

  • Image Classification Data Set (Fashion MNIST)

  • Softmax regression is implemented from scratch

  • Gluon implementation of Softmax regression

  • Multilayer perceptron

  • Implementation of multilayer perceptron from scratch

  • Gluon implementation of multilayer perceptron

  • Model selection, underfitting and overfitting

  • Weight decay

  • Discarded method

  • Forward propagation, back propagation, and computed graphs

  • Numerical stability and model initialization

  • Real Kaggle competition: Housing price forecast

Deep learning computing

  • Model construction

  • Access, initialization, and sharing of model parameters

  • Deferred initialization of model parameters

  • Custom layer

  • Read and store

  • GPU computing

Convolutional neural network

  • Two dimensional convolution layer

  • Fill and stride

  • Multiple input channels and multiple output channels

  • Pooling layer

  • Convolutional Neural Network (LeNet)

  • Deep Convolutional Neural Network (AlexNet)

  • Networks using Repeating elements (VGG)

  • Network of Networks (NiN)

  • Network with parallel links (GoogLeNet)

  • Batch normalization

  • Residual network (ResNet)

  • Dense Connected Network (DenseNet)

Recurrent neural network

  • Language model

  • Recurrent neural network

  • Language Model dataset (Jay Chou album Lyrics)

  • Implementation of cyclic neural network from scratch

  • Gluon implementation of recurrent neural networks

  • Back propagation through time

  • Gated Cycle Unit (GRU)

  • Long and short-term memory (LSTM)

  • Deep recurrent neural network

  • Bidirectional cyclic neural network

Optimization algorithm

  • Optimization and deep learning

  • Gradient descent and stochastic gradient descent

  • Small batch stochastic gradient descent

  • The momentum method

  • Adagrad

  • RMSProp

  • Adadelta

  • Adam

Computing performance

  • Mixed imperative and symbolic programming

  • Asynchronous computation

  • Automatic parallel computing

  • Many GPU computing

  • Gluon implementation of multi-GPU computing

Computer vision

  • Image augmented

  • fine-tuning

  • Object detection and bounding boxes

  • Anchor box

  • Multiscale target detection

  • Target detection Dataset (Pikachu)

  • Single-shot Multi-Frame Detection (SSD)

  • Regional convolutional Neural Networks (R-CNN) series

  • Semantic segmentation and data sets

  • Full convolutional network (FCN)

  • Style migration

  • Actual Kaggle Competition: Image Classification (CIFAR-10)

  • ImageNet Dogs (ImageNet Dogs)

Natural language processing

  • Word embedding (word2vec)

  • The approximate training

  • The realization of the Word2vec

  • Subword embedding (fastText)

  • Word embedding of global vector

  • Find synonyms and analogies

  • Text emotion Classification: Using recurrent neural networks

  • Text Sentiment Classification: Using Convolutional Neural Network (textCNN)

  • Encoder — Decoder (SEQ2SEQ)

  • Beam search

  • Attentional mechanism

  • Machine translation

The appendix

  • List of major symbols

  • Mathematical basis

  • Use Jupyter notebooks

  • Use AWS to run the code

  • GPU Purchase Guide

  • How can I contribute to the book

  • Gluonbook package index

The book follows a series of videos by Ms. Li and others called ‘Hands-on Deep Learning,’ the first season of the 19-lesson course that ended in February. Dr. Mu Li has organized this series of videos, students who need to learn through the following video.

  • Video: space.bilibili.com/209599371/#…