本書采用“理論 +實(shí)踐”的方式，全面系統(tǒng)地講授了深度學(xué)習(xí)的基本原理以及使用 TensorFlow實(shí)現(xiàn)各類深度學(xué)習(xí)網(wǎng)絡(luò)的方法。全書共 10章，第 1～3章主要介紹深度學(xué)習(xí)的基礎(chǔ)知識，包括深度學(xué)習(xí)的概念和應(yīng)用、深層神經(jīng)網(wǎng)絡(luò)的訓(xùn)練和優(yōu)化、 TensorFlow的內(nèi)涵和特點(diǎn)等內(nèi)容；第 4～5章主要介紹 TensorFlow的安裝，以及計(jì)算模型、數(shù)據(jù)模型、運(yùn)行模型等 TensorFlow編程的基礎(chǔ)知識；第 6～10章主要圍繞 TensorFlow介紹各類深度學(xué)習(xí)網(wǎng)絡(luò)，包括單個神經(jīng)元、多層神經(jīng)網(wǎng)絡(luò)、卷積神經(jīng)網(wǎng)絡(luò)、循環(huán)神經(jīng)網(wǎng)絡(luò)、深度學(xué)習(xí)網(wǎng)絡(luò)進(jìn)階等。全書在各個章節(jié)設(shè)置有大量的實(shí)驗(yàn)和實(shí)操案例，兼具知識性和實(shí)用性。

閉應(yīng)洲，南寧師范大學(xué)教授，主要研究方向?yàn)橹悄苡?jì)算、智能醫(yī)學(xué)圖像處理及社會計(jì)算。主持和參與了10多項(xiàng)科研項(xiàng)目的研究工作，發(fā)布論文50多篇。2012年2月至2013年2月在美國亞利桑那州立大學(xué)訪學(xué)，重點(diǎn)研究從海量數(shù)據(jù)中獲取知識所必需的理論和技術(shù)。

目 錄

第 1章引言····················································································································1
1.1 人工智能簡介······································································································1
1.2 機(jī)器學(xué)習(xí)簡介······································································································2
1.2.1 機(jī)器學(xué)習(xí)的概念·····························································································2
1.2.2 機(jī)器學(xué)習(xí)的本質(zhì)·····························································································2
1.2.3 機(jī)器學(xué)習(xí)的步驟·····························································································3
1.2.4 機(jī)器學(xué)習(xí)的關(guān)鍵點(diǎn)··························································································5
1.2.5 機(jī)器學(xué)習(xí)的實(shí)戰(zhàn)·····························································································6
1.2.6 機(jī)器學(xué)習(xí)的教材·····························································································7
1.3 機(jī)器學(xué)習(xí)的分類 ··································································································8
1.3.1 有監(jiān)督學(xué)習(xí)···································································································8
1.3.2 無監(jiān)督學(xué)習(xí)···································································································9
1.3.3 半監(jiān)督學(xué)習(xí)································································································.10
1.3.4 強(qiáng)化學(xué)習(xí)···································································································.11
1.4 本章小結(jié)··········································································································.12
第 2章深度學(xué)習(xí)的原理 ·······························································································.13
2.1 深度學(xué)習(xí)簡介···································································································.13
2.1.1 深度學(xué)習(xí)的概念··························································································.13
2.1.2 深度學(xué)習(xí)的特點(diǎn)··························································································.13
2.2 深度學(xué)習(xí)的現(xiàn)實(shí)意義 ························································································.14
2.2.1 多層神經(jīng)網(wǎng)絡(luò)的模型結(jié)構(gòu) ··············································································.14
2.2.2 非線性處理能力··························································································.14
2.2.3 特征自動提取和轉(zhuǎn)換····················································································.16
2.3 深度學(xué)習(xí)的應(yīng)用領(lǐng)域 ························································································.16
2.3.1 計(jì)算機(jī)視覺································································································.17
2.3.2 自然語言處理·····························································································.20
2.3.3 語音識別···································································································.21
2.4 深層神經(jīng)網(wǎng)絡(luò)簡介····························································································.22
2.4.1 神經(jīng)元模型································································································.22
2.4.2 單層神經(jīng)網(wǎng)絡(luò)·····························································································.23
2.4.3 深層神經(jīng)網(wǎng)絡(luò)·····························································································.24
2.4.4 深層神經(jīng)網(wǎng)絡(luò)節(jié)點(diǎn)·······················································································.24
2.4.5 深層神經(jīng)網(wǎng)絡(luò)參數(shù)·······················································································.25
2.4.6 節(jié)點(diǎn)輸出值計(jì)算··························································································.25
2.5 深層神經(jīng)網(wǎng)絡(luò)的訓(xùn)練與優(yōu)化 ··············································································.26
2.5.1 深層神經(jīng)網(wǎng)絡(luò)的訓(xùn)練····················································································.26
2.5.2 深層神經(jīng)網(wǎng)絡(luò)的優(yōu)化····················································································.32
2.6 本章小結(jié)··········································································································.35
第 3章深度學(xué)習(xí)框架簡介 ····························································································.37
3.1 TensorFlow簡介 ·······························································································.37
3.2 TensorFlow的特點(diǎn)····························································································.38
3.3 其他深度學(xué)習(xí)框架····························································································.38
3.4 本章小結(jié)··········································································································.41
第 4章 TensorFlow的安裝···························································································.42
4.1 安裝準(zhǔn)備··········································································································.42
4.1.1 硬件檢查···································································································.42
4.1.2 處理器推薦—GPU····················································································.44
4.1.3 系統(tǒng)選擇—Linux ·····················································································.53
4.1.4 配合 Python語言使用···················································································.53
4.1.5 Anaconda的安裝·························································································.54
4.2 TensorFlow的主要依賴包 ·················································································.55
4.2.1 Protocol Buffer····························································································.56
4.2.2 Bazel········································································································.57
4.3 Python安裝 TensorFlow·····················································································.59
4.3.1 使用 pip安裝 ·····························································································.59
4.3.2 從源代碼編譯并安裝····················································································.59
4.4 TensorFlow的使用····························································································.60
4.4.1 向量求和···································································································.60
4.4.2 加載過程的問題··························································································.61
4.5 推薦使用 IDE ···································································································.61
4.6 本章小結(jié)··········································································································.62
第 5章 TensorFlow編程基礎(chǔ) ·······················································································.63
5.1 計(jì)算圖與張量···································································································.63
5.1.1 初識計(jì)算圖與張量·······················································································.63
5.1.2 TensorFlow的計(jì)算模型—計(jì)算圖··································································.63
5.1.3 TensorFlow的數(shù)據(jù)模型—張量·····································································.66
5.2 TensorFlow的運(yùn)行模型 —會話 ·······································································.68
5.2.1 TensorFlow的系統(tǒng)結(jié)構(gòu) ················································································.68
5.2.2 會話的使用································································································.69
5.2.3 使用 with/as進(jìn)行上下文管理 ·········································································.70
5.2.4 會話的配置································································································.71
5.2.5 占位符······································································································.72
5.3 TensorFlow變量 ·······························································································.73
5.3.1 變量的創(chuàng)建································································································.73
5.3.2 變量與張量································································································.76
·VI.

5.3.3 管理變量空間·····························································································.77
5.4 實(shí)驗(yàn)：識別圖中模糊的手寫數(shù)字 ·······································································.82
5.5 本章小結(jié)··········································································································.88
第 6章單個神經(jīng)元 ······································································································.89
6.1 神經(jīng)元擬合原理 ·······························································································.89
6.1.1 正向傳播···································································································.90
6.1.2 反向傳播···································································································.90
6.2 激活函數(shù)··········································································································.91
6.2.1 Sigmoid函數(shù)······························································································.91
6.2.2 Tanh函數(shù)··································································································.92
6.2.3 ReLU函數(shù) ································································································.93
6.2.4 Swish函數(shù) ································································································.96
6.3 Softmax算法與損失函數(shù) ···················································································.96
6.3.1 Softmax算法······························································································.97
6.3.2 損失函數(shù)···································································································.98
6.3.3 綜合應(yīng)用實(shí)驗(yàn)·····························································································101
6.4 梯度下降··········································································································104
6.4.1 梯度下降方法·····························································································105
6.4.2 梯度下降函數(shù)·····························································································105
6.4.3 退化學(xué)習(xí)率································································································106
6.5 學(xué)習(xí)參數(shù)初始化 ·······························································································108
6.6 使用 Maxout網(wǎng)絡(luò)擴(kuò)展單個神經(jīng)元 ·····································································109
6.6.1 Maxout簡介 ······························································································109
6.6.2 使用 Maxout網(wǎng)絡(luò)實(shí)現(xiàn) MNIST分類 ·································································110
6.7 本章小結(jié)··········································································································111
第 7章多層神經(jīng)網(wǎng)絡(luò) ···································································································112
7.1 線性問題與非線性問題 ·····················································································112
7.1.1 用線性邏輯回歸處理二分類問題 ·····································································112
7.1.2 用線性邏輯回歸處理多分類問題 ·····································································116
7.1.3 非線性問題淺析··························································································121
7.2 解決非線性問題 ·······························································································121
7.2.1 使用帶隱藏層的神經(jīng)網(wǎng)絡(luò)擬合異或操作 ····························································121
7.2.2 非線性網(wǎng)絡(luò)的可視化····················································································123
7.3 利用全連接神經(jīng)網(wǎng)絡(luò)將圖片進(jìn)行分類 ································································125
7.4 全連接神經(jīng)網(wǎng)絡(luò)模型的優(yōu)化方法 ·······································································127
7.4.1 利用異或數(shù)據(jù)集演示過擬合問題 ·····································································127
7.4.2 通過正則化改善過擬合情況 ···········································································132
7.4.3 通過增大數(shù)據(jù)集改善過擬合 ···········································································134
7.4.4 基于 Dropout技術(shù)來擬合異或數(shù)據(jù)集 ·······························································135
7.4.5 全連接神經(jīng)網(wǎng)絡(luò)的深淺關(guān)系 ···········································································138
7.5 本章小結(jié)··········································································································139
第 8章卷積神經(jīng)網(wǎng)絡(luò) ···································································································140
8.1 認(rèn)識卷積神經(jīng)網(wǎng)絡(luò)····························································································140
8.1.1 全連接神經(jīng)網(wǎng)絡(luò)的局限性 ··············································································140
8.1.2 卷積神經(jīng)網(wǎng)絡(luò)簡介·······················································································140
8.2 卷積神經(jīng)網(wǎng)絡(luò)的結(jié)構(gòu) ························································································141
8.2.1 網(wǎng)絡(luò)結(jié)構(gòu)簡介·····························································································141
8.2.2 卷積層······································································································144
8.2.3 池化層······································································································147
8.3 卷積神經(jīng)網(wǎng)絡(luò)的相關(guān)函數(shù) ·················································································147
8.3.1 卷積函數(shù) tf.nn.conv2d···················································································147
8.3.2 池化函數(shù) tf.nn.max_pool和 tf.nn.avg_pool··························································154
8.4 使用卷積神經(jīng)網(wǎng)絡(luò)對圖片分類 ··········································································157
8.4.1 CIFAR數(shù)據(jù)集介紹及使用 ·············································································157
8.4.2 CIFAR數(shù)據(jù)集的處理 ···················································································160
8.4.3 建立一個卷積神經(jīng)網(wǎng)絡(luò) ·················································································166
8.5 反卷積神經(jīng)網(wǎng)絡(luò) ·······························································································168
8.5.1 反卷積計(jì)算································································································169
8.5.2 反池化計(jì)算································································································171
8.5.3 反卷積神經(jīng)網(wǎng)絡(luò)的應(yīng)用 ·················································································171
8.6 卷積神經(jīng)網(wǎng)絡(luò)進(jìn)階····························································································171
8.6.1 函數(shù)封裝庫的使用·······················································································172
8.6.2 深度學(xué)習(xí)的模型訓(xùn)練技巧 ··············································································174
8.7 本章小結(jié)··········································································································182
第 9章循環(huán)神經(jīng)網(wǎng)絡(luò) ···································································································183
9.1 循環(huán)神經(jīng)網(wǎng)絡(luò)的原理 ························································································183
9.1.1 循環(huán)神經(jīng)網(wǎng)絡(luò)的基本結(jié)構(gòu) ··············································································183
9.1.2 RNN的反向傳播過程 ···················································································184
9.1.3 搭建簡單 RNN····························································································186
9.2 改進(jìn)的 RNN ·····································································································192
9.2.1 LSTM·······································································································193
9.2.2 改進(jìn)的 LSTM ·····························································································196
9.2.3 Bi-RNN ····································································································198
9.2.4 CTC·········································································································200
9.3 RNN實(shí)戰(zhàn)·········································································································200
9.3.1 cell類 ······································································································200
9.3.2 構(gòu)建 RNN··································································································201
9.3.3 使用 RNN對 MNIST數(shù)據(jù)集分類 ····································································207
9.3.4 RNN的初始化 ····························································································213
9.3.5 RNN的優(yōu)化 ·······························································································213
·VIII.

9.3.6 利用 BiRNN實(shí)現(xiàn)語音識別 ············································································214
9.4 本章小結(jié)··········································································································228
第 10章深度學(xué)習(xí)網(wǎng)絡(luò)進(jìn)階 ··························································································229
10.1深層神經(jīng)網(wǎng)絡(luò) ·································································································229
10.1.1 深層神經(jīng)網(wǎng)絡(luò)介紹 ·····················································································229
10.1.2 GoogLeNet模型 ························································································230
10.1.3 ResNet模型 ·····························································································234
10.1.4 Inception-ResNet-v2模型 ·············································································235
10.1.5 TensorFlow中圖片分類模型庫 —slim ···························································235
10.1.6 slim深度網(wǎng)絡(luò)模型實(shí)戰(zhàn)圖像識別 ···································································241
10.1.7 實(shí)物檢測模型庫 ························································································244
10.1.8 實(shí)物檢測領(lǐng)域的相關(guān)模型 ············································································245
10.1.9 NASNet控制器 ·························································································246
10.2生成對抗神經(jīng)網(wǎng)絡(luò) ··························································································247
10.2.1 什么是 GAN ·····························································································247
10.2.2 各種不同的 GAN ·······················································································248
10.2.3 GAN實(shí)踐································································································253
10.2.4 GAN網(wǎng)絡(luò)的高級接口 TFGAN ······································································263
10.3本章小結(jié) ········································································································264