第1 章 TensorFlow 簡介 ·············································································.1
1.1 人工智能的編程框架 ................................................................................................. 1
1.1.1 人工智能的發(fā)展 ............................................................................................. 1
1.1.2 人工智能、機(jī)器學(xué)習(xí)和深度學(xué)習(xí)之間的關(guān)系 ............................................. 2
1.2 TensorFlow 與人工智能 ............................................................................................ 3
1.3 TensorFlow 數(shù)據(jù)模型 ................................................................................................ 4
1.4 TensorFlow 計算模型和運(yùn)行模型 ............................................................................ 5
1.5 實驗：矩陣運(yùn)算 ......................................................................................................... 9
1.5.1 實驗?zāi)康?......................................................................................................... 9
1.5.2 實驗要求 ......................................................................................................... 9
1.5.3 實驗原理 ......................................................................................................... 9
1.5.4 實驗步驟 ....................................................................................................... 10
習(xí)題 .................................................................................................................................... 10
第2 章 Python 語言基礎(chǔ) ············································································.11
2.1 Python 語言 ............................................................................................................... 11
2.1.1 Python 語言的發(fā)展 ....................................................................................... 11
2.1.2 Python 安裝 ................................................................................................... 12
2.2 基礎(chǔ)語法 ................................................................................................................... 13
2.2.1 基礎(chǔ)知識 ....................................................................................................... 13
2.2.2 基本程序編寫 ............................................................................................... 15
2.2.3 條件語句 ....................................................................................................... 16
2.2.4 循環(huán)語句 ....................................................................................................... 17
2.3 數(shù)據(jù)結(jié)構(gòu) ................................................................................................................... 18
2.4 面向?qū)ο筇匦?........................................................................................................... 21
2.4.1 類和對象 ....................................................................................................... 21
2.4.2 類的定義 ....................................................................................................... 22
2.4.3 根據(jù)類創(chuàng)建對象 ........................................................................................... 22
2.4.4 構(gòu)造方法與析構(gòu)方法 ................................................................................... 23
2.5 其他高級特性 ........................................................................................................... 24
2.5.1 函數(shù)高級特性 ............................................................................................... 24
2.5.2 閉包 ............................................................................................................... 25
2.6 實驗：Python 基本語法的實現(xiàn) ............................................................................... 26
2.6.1 實驗?zāi)康?....................................................................................................... 26
2.6.2 實驗要求 ....................................................................................................... 26
2.6.3 實驗題目 ....................................................................................................... 26
2.6.4 實驗步驟 ....................................................................................................... 27
習(xí)題 .................................................................................................................................... 28
第3 章 環(huán)境搭建與入門 ·············································································.30
3.1 開發(fā)平臺簡介 ........................................................................................................... 30
3.2 開發(fā)環(huán)境部署 ........................................................................................................... 30
3.2.1 安裝Anaconda .............................................................................................. 30
3.2.2 安裝TensorFlow ........................................................................................... 32
3.2.3 PyCharm 下載與安裝 ................................................................................... 32
3.3 一個簡單的實例 ....................................................................................................... 34
習(xí)題 .................................................................................................................................... 36
第4 章 TensorBoard 可視化 ········································································.37
4.1 什么是TensorBoard.................................................................................................. 37
4.2 基本流程與結(jié)構(gòu) ....................................................................................................... 37
4.3 圖表的可視化 ........................................................................................................... 39
4.3.1 計算圖和會話 ............................................................................................... 39
4.3.2 可視化過程 ................................................................................................... 40
4.4 監(jiān)控指標(biāo)的可視化 ................................................................................................... 41
4.4.1 Scalar ............................................................................................................. 41
4.4.2 Images ........................................................................................................... 41
4.4.3 Histogram ...................................................................................................... 41
4.4.4 Merge_all....................................................................................................... 42
4.5 學(xué)習(xí)過程的可視化 ................................................................................................... 42
4.5.1 數(shù)據(jù)序列化 ................................................................................................... 43
4.5.2 啟動TensorBoard ......................................................................................... 43
4.6 實驗：TensorBoard 可視化實現(xiàn) .............................................................................. 44
4.6.1 實驗?zāi)康?....................................................................................................... 44
4.6.2 實驗要求 ....................................................................................................... 44
4.6.3 實驗原理 ....................................................................................................... 45
4.6.4 實驗步驟 ....................................................................................................... 45
習(xí)題 .................................................................................................................................... 49
第5 章 多層感知機(jī)實現(xiàn) ·············································································.50
5.1 感知機(jī) ....................................................................................................................... 50
5.1.1 感知機(jī)的定義 ............................................................................................... 50
5.1.2 感知機(jī)的神經(jīng)元模型 ................................................................................... 51
5.1.3 感知機(jī)的學(xué)習(xí)算法 ....................................................................................... 51
5.1.4 感知機(jī)的性質(zhì) ............................................................................................... 52
5.2 多層感知機(jī)與前向傳播 ........................................................................................... 53
5.2.1 多層感知機(jī)基本結(jié)構(gòu) ................................................................................... 53
5.2.2 多層感知機(jī)的特點 ....................................................................................... 54
5.3 前向傳播 ................................................................................................................... 55
5.3.1 前向傳播的計算過程 ................................................................................... 55
5.3.2 前向傳播算法 ............................................................................................... 57
5.4 梯度下降 ................................................................................................................... 57
5.4.1 梯度 ............................................................................................................... 57
5.4.2 梯度下降的直觀解釋 ................................................................................... 58
5.4.3 梯度下降法的相關(guān)概念 ............................................................................... 58
5.4.4 梯度下降法的數(shù)學(xué)描述 ............................................................................... 59
5.4.5 梯度下降法的算法調(diào)優(yōu) ............................................................................... 60
5.4.6 常見的梯度下降法 ....................................................................................... 60
5.5 反向傳播 ................................................................................................................... 61
5.5.1 反向傳播算法要解決的問題 ....................................................................... 61
5.5.2 反向傳播算法的基本思路 ........................................................................... 61
5.5.3 反向傳播算法的流程 ................................................................................... 63
5.6 數(shù)據(jù)集 ....................................................................................................................... 64
5.6.1 訓(xùn)練集、測試集和驗證集 ........................................................................... 64
5.6.2 MNIST 數(shù)據(jù)集 ............................................................................................. 64
5.7 多層感知機(jī)的實現(xiàn) ................................................................................................... 66
5.7.1 NumPy 多層感知機(jī)的實現(xiàn) .......................................................................... 66
5.7.2 TensorFlow 多層感知機(jī)的實現(xiàn) ................................................................... 69
5.8 實驗：基于Keras 多層感知機(jī)的MNIST 手寫數(shù)字識別 ...................................... 72
5.8.1 Keras 簡介 ..................................................................................................... 72
5.8.2 實驗?zāi)康?....................................................................................................... 73
5.8.3 實驗要求 ....................................................................................................... 73
5.8.4 實驗步驟 ....................................................................................................... 73
習(xí)題 .................................................................................................................................... 77
第6 章 卷積神經(jīng)網(wǎng)絡(luò)實現(xiàn) ··········································································.78
6.1 CNN 基本原理 .......................................................................................................... 78
6.2 CNN 的卷積操作 ...................................................................................................... 80
6.3 CNN 的池化操作 ...................................................................................................... 82
6.4 使用簡單的CNN 實現(xiàn)手寫字符識別 ..................................................................... 84
6.5 AlexNet ..................................................................................................................... 85
6.6 實驗：基于VGG16 模型的圖像分類實現(xiàn) ............................................................. 87
6.6.1 實驗?zāi)康?....................................................................................................... 87
6.6.2 實驗要求 ....................................................................................................... 87
6.6.3 實驗原理 ....................................................................................................... 88
6.6.4 實驗步驟 ....................................................................................................... 88
習(xí)題 .................................................................................................................................... 93
第7 章 循環(huán)神經(jīng)網(wǎng)絡(luò)實現(xiàn) ··········································································.94
7.1 RNN 簡介 .................................................................................................................. 94
7.1.1 為什么使用RNN.......................................................................................... 94
7.1.2 RNN 的網(wǎng)絡(luò)結(jié)構(gòu)及原理 .............................................................................. 96
7.1.3 RNN 的實現(xiàn) ................................................................................................. 99
7.2 長短時記憶網(wǎng)絡(luò) ..................................................................................................... 100
7.2.1 長期依賴問題 ............................................................................................. 100
7.2.2 長短時記憶網(wǎng)絡(luò) ......................................................................................... 101
7.2.3 LSTM 的實現(xiàn) ............................................................................................. 105
7.3 雙向RNN ................................................................................................................ 106
7.3.1 雙向RNN 的結(jié)構(gòu)及原理 ........................................................................... 106
7.3.2 雙向RNN 的實現(xiàn)....................................................................................... 107
7.4 深層RNN ................................................................................................................ 108
7.5 實驗：基于LSTM 的股票預(yù)測 ............................................................................. 110
7.5.1 實驗?zāi)康?..................................................................................................... 110
7.5.2 實驗要求 ..................................................................................................... 110
7.5.3 實驗原理 ..................................................................................................... 111
7.5.4 實驗步驟 ..................................................................................................... 111
習(xí)題 .................................................................................................................................. 114
第8 章 強(qiáng)化學(xué)習(xí) ····················································································.115
8.1 強(qiáng)化學(xué)習(xí)原理 ......................................................................................................... 115
8.2 馬爾可夫決策過程實現(xiàn) ......................................................................................... 117
8.2.1 馬爾可夫決策過程 ..................................................................................... 117
8.2.2 馬爾可夫決策過程的形式化 ..................................................................... 118
8.3 基于價值的強(qiáng)化學(xué)習(xí)方法 ..................................................................................... 120
8.3.1 基于價值的方法中的策略優(yōu)化 ................................................................. 120
8.3.2 基于價值的方法中的策略評估 ................................................................. 120
8.3.3 Q-Learning .................................................................................................. 122
8.4 Gym 的簡單使用 .................................................................................................... 123
8.5 實驗：基于強(qiáng)化學(xué)習(xí)的小車爬山游戲 ................................................................. 125
8.5.1 實驗?zāi)康?..................................................................................................... 125
8.5.2 實驗要求 ..................................................................................................... 125
8.5.3 實驗原理 ..................................................................................................... 125
8.5.4 實驗步驟 ..................................................................................................... 127
習(xí)題 .................................................................................................................................. 130
第9 章 遷移學(xué)習(xí) ····················································································.131
9.1 遷移學(xué)習(xí)原理 ......................................................................................................... 131
9.1.1 什么是遷移學(xué)習(xí) ......................................................................................... 131
9.1.2 遷移學(xué)習(xí)的基本概念 ................................................................................. 131
9.1.3 遷移學(xué)習(xí)的基本方法 ................................................................................. 133
9.2 基于模型的遷移學(xué)習(xí)方法實現(xiàn) ............................................................................. 134
9.2.1 導(dǎo)入已有的預(yù)訓(xùn)練模型 ............................................................................. 134
9.2.2 模型的復(fù)用 ................................................................................................. 134
9.2.3 基于新模型的預(yù)測 ..................................................................................... 135
9.3 基于VGG-19 的遷移學(xué)習(xí)實現(xiàn) ............................................................................. 135
9.3.1 VGG-19 的原理 .......................................................................................... 135
9.3.2 基于VGG-19 的遷移學(xué)習(xí)的原理及實現(xiàn) ................................................. 136
9.4 實驗：基于Inception V3 的遷移學(xué)習(xí) .................................................................. 138
9.4.1 實驗?zāi)康?..................................................................................................... 138
9.4.2 實驗要求 ..................................................................................................... 138
9.4.3 實驗原理 ..................................................................................................... 139
9.4.4 實驗步驟 ..................................................................................................... 140
習(xí)題 .................................................................................................................................. 143
第10 章 生成對抗網(wǎng)絡(luò) ·············································································.144
10.1 GAN 概述 ............................................................................................................. 144
10.2 GAN 的目標(biāo)函數(shù) ................................................................................................. 144
10.3 GAN 的實現(xiàn) ......................................................................................................... 145
10.4 深度卷積生成對抗網(wǎng)絡(luò) ....................................................................................... 149
10.4.1 DCGAN 結(jié)構(gòu)圖 ........................................................................................ 150
10.4.2 DCGAN 的實現(xiàn) ........................................................................................ 150
10.5 GAN 的衍生模型 ................................................................................................. 153
10.5.1 基于網(wǎng)絡(luò)結(jié)構(gòu)的衍生模型 ....................................................................... 154
10.5.2 基于優(yōu)化方法的衍生模型 ....................................................................... 155
習(xí)題 .................................................................................................................................. 156
第11 章 GPU 并行計算 ············································································.157
11.1 并行計算技術(shù) ....................................................................................................... 157
11.1.1 單機(jī)并行計算 ........................................................................................... 157
11.1.2 分布式并行計算 ....................................................................................... 158
11.1.3 GPU 并行計算技術(shù) .................................................................................. 159
11.1.4 TensorFlow 與GPU .................................................................................. 160
11.2 TensorFlow 加速方法 ........................................................................................... 163
11.3 單GPU 并行加速的實現(xiàn) ..................................................................................... 170
11.4 多GPU 并行加速的實現(xiàn) ..................................................................................... 173
11.5 實驗：基于GPU 的矩陣乘法 ............................................................................. 175
11.5.1 安裝GPU 版本的TensorFlow ................................................................. 175
11.5.2 一個GPU 程序 ......................................................................................... 176
11.5.3 使用GPU 完成矩陣乘法 ......................................................................... 176
習(xí)題 .................................................................................................................................. 177