Deep Learning and Practice with Mindspore

ISBN-13: 9789811622328 / Angielski / Twarda / 2021 / 398 str.

Lei Chen

Spis treści

Chapter 1 Introduction. 1

1.1             AI's Historical Changes 1

1.2             What Is Deep Learning?. 3

1.3             Practical Applications of Deep Learning. 4

1.4             Structure of the Book. 7

1.5             Introduction to MindSpore. 7

Chapter 2 Deep Learning Basics. 18

2.1             Regression Algorithms. 18

2.2             Gradient Descent 21

2.3             Classification Algorithms. 25

2.4             Overfitting and Underfitting. 28

Chapter 3 DNN.. 32

3.1             Feedforward Network. 32

3.2             Backpropagation. 34

3.3             Generalization Ability. 38

3.4             Implementing Simple Neural Networks Using MindSpore. 39

Chapter 4 Training of DNNs. 45

4.1             Main Challenges to Deep Learning Systems 45

4.2             Regularization. 48

4.3             Dropout 51

4.4             Adaptive Learning Rate. 55

4.5             Batch Normalization. 59

4.6             Implementing DNNs Using MindSpore. 61

Chapter 5 Convolutional Neural Network. 66

5.1             Convolution. 66

5.2             Pooling. 69

5.3             Residual Network. 71

5.4             Application: Image Classification. 74

5.5             Implementing Image Classification Based on the
DNN Using MindSpore. 79

Chapter 6 RNN.. 89

6.1             Overview.. 89

6.2             Deep RNN.. 90

6.3             Challenges of Long-Term Dependency. 91

6.4             LSTM Network and GRU.. 93

6.5             Application: Text Prediction. 96

6.6             Implementing Text Prediction Based on LSTM Using MindSpore. 97

Chapter 7 Unsupervised Learning: Word Vector. 101

7.1             Word2Vec. 102

7.2             GloVe. 114

7.3             Transformer 121

7.4             BERT.. 130

7.5             Comparison Between Typical Word Vector Generation Algorithms. 137

7.6             Application: Automatic Question Answering. 139

7.7             Implementing BERT-based Automatic Answering Using MindSpore. 154

Chapter 8 Unsupervised Learning: Graph Vector. 159

8.1             Graph Vector Overview.. 159

8.2             DeepWalk Algorithm... 161

8.3             LINE Algorithm... 166

8.4             Node2Vec Algorithm... 170

8.5             GCN Algorithm... 174

8.6             GAT Algorithm... 179

8.7             Application: Recommendation System.. 183

Chapter 9 Unsupervised Learning: Deep Generative Model 191

9.1             Variational Autoencoder 191

9.2             Generative Adversarial Network. 200

9.3             Application: Data Augmentation. 208

9.4             Implementing GAN-based Data Augmentation Using MindSpore. 221

Chapter 10 Deep Reinforcement Learning. 225

10.1           Basic Concepts of Reinforcement Learning. 225

10.2           Basic Solution Method. 230

10.3           Deep Reinforcement Learning Algorithm... 235

10.4           Latest Applications. 247

10.5           Implementing DQN-based Game Using MindSpore. 253

Chapter 11 Automated Machine Learning. 255

11.1           AutoML Framework. 255

11.2           Existing AutoML Systems. 278

11.3           Meta Learning. 288

11.4           Implementing AutoML Using MindSpore. 294

Chapter 12 Device-Cloud Collaboration. 302

12.1           On-device Inference. 302

12.2           Device-Cloud Transfer Learning. 304

12.3           Device-Cloud Federated Learning. 308

12.4           Device-Cloud Collaboration Framework. 313

Chapter 13 Deep Learning Visualization. 322

13.1           Overview.. 322

13.2           MindSpore Visualization. 337

Chapter 14 Data Preparation for Deep Learning. 354

14.1           Overview of Data Format 354

14.2           Data Format in Deep Learning. 355

14.3           Common Data Formats for Deep Learning. 362

14.4        Training Data Preparation Using the MindSpore Data Format 377

 

O autorze

Chen Lei is a Chair Professor of the Department of Computer Science and Engineering and the Director of the Big Data Institute at Hong Kong University of Science and Technology (HKUST). His research focuses on data-driven AI, human-powered machine learning, knowledge graphs, and data mining on social media. He has published more than 400 papers in world-renowned journals and conference proceedings and won the 2015 SIGMOD Test of Time Award. Currently, he serves as the Editor-in-Chief of the VLDB 2019 Journal, the Associate Editor-in-Chief of the IEEE TKDE Journal, and an executive member of the VLDB Endowment. He is also IEEE Fellow and ACM Distinguished Scientist.

Blurb

This book systematically introduces readers to the theory of deep learning and explores its practical applications based on the MindSpore AI computing framework. Divided into 14 chapters, the book covers deep learning, deep neural networks (DNNs), convolutional neural networks (CNNs), recurrent neural networks (RNNs), unsupervised learning, deep reinforcement learning, automated machine learning, device-cloud collaboration, deep learning visualization, and data preparation for deep learning. To help clarify the complex topics discussed, this book includes numerous examples and links to online resources.