邱錫鵬教授姊妹篇《神經網(wǎng)絡與深度學習：案例與實踐》重磅來襲（贈書）
內容簡介
掃碼購買
讀者對象
高等院校人工智能、數(shù)據(jù)科學、計算機等相關專業(yè)學生
深度學習入門者
工業(yè)界從事人工智能應用的專業(yè)人員
本書亮點
緊密配套蒲公英書：章節(jié)設計一一對應，以模型解讀+案例實踐的形式進行介紹。 更適合深度學習的入門者使用：實踐案例使用飛槳框架編寫，代碼簡潔，從零開始一步步進行深度學習的實踐，搭建一個輕量級的機器學習框架以及相應的算子庫來完成實際任務。 術語翻譯更加規(guī)范：機器學習領域的很多名詞存在難翻譯和亂翻譯的現(xiàn)象，邱錫鵬教授與周志華老師、李航老師、李沐、阿斯頓·張一起討論和確定了機器學習相關術語的翻譯問題，本書中采用了相關術語的最新譯法。 全方位深度學習入門及提高解決方案：提供免費的視頻課程、豐富題庫和實訓環(huán)境，邱錫鵬教授和百度飛槳研發(fā)團隊親自講解示范。
目錄
序
前言
第1章實踐基礎1
1.1如何運行本書的代碼...................................2
1.1.1本地運行.....................................2
1.1.2代碼下載與使用方法..............................3
1.1.3在線運行.....................................4
1.2張量............................................6
1.2.1創(chuàng)建張量.....................................6
1.2.2張量的屬性....................................9
1.2.3張量與Numpy數(shù)組轉換............................13
1.2.4張量的訪問....................................13
1.2.5張量的運算....................................16
1.3算子............................................20
1.3.1算子定義.....................................21
1.3.2自動微分機制..................................25
1.3.3預定義的算子..................................27
1.3.4本書中實現(xiàn)的算子................................27
1.3.5本書中實現(xiàn)的優(yōu)化器..............................29
1.4本書中使用的數(shù)據(jù)集和實現(xiàn)的Dataset類........................29
1.4.1數(shù)據(jù)集......................................29
1.4.2Dataset類....................................31
1.5本書中實現(xiàn)的Runner類.................................31
1.6小結............................................32
第2章機器學習概述33
2.1機器學習實踐五要素...................................34
2.1.1數(shù)據(jù)........................................35
2.1.2模型........................................36
2.1.3學習準則.....................................36
2.1.4優(yōu)化算法.....................................37
2.1.5評價指標.....................................37
2.2實現(xiàn)一個簡單的線性回歸模型..............................38
2.2.1數(shù)據(jù)集構建....................................38
2.2.2模型構建.....................................40
2.2.3損失函數(shù).....................................42
2.2.4優(yōu)化器......................................43
2.2.5模型訓練.....................................45
2.2.6模型評價.....................................45
2.3多項式回歸........................................46
2.3.1數(shù)據(jù)集構建：ToySin25..............................46
2.3.2模型構建.....................................48
2.3.3模型訓練.....................................49
2.3.4模型評價.....................................50
2.3.5通過引入正則化項來緩解過擬合........................52
2.4構建Runner類......................................53
2.5實踐：基于線性回歸的波士頓房價預測.........................55
2.5.1數(shù)據(jù)處理.....................................55
2.5.2模型構建.....................................62
2.5.3完善Runner類：RunnerV1...........................62
2.5.4模型訓練.....................................63
2.5.5模型評價.....................................64
2.5.6模型預測.....................................64
2.6小結............................................65
第3章線性分類67
3.1基于Logistic回歸的二分類任務.............................68
3.1.1數(shù)據(jù)集構建....................................69
3.1.2模型構建.....................................71
3.1.3損失函數(shù).....................................73
3.1.4模型優(yōu)化.....................................74
3.1.5評價指標.....................................77
3.1.6完善Runner類：RunnerV2...........................77
3.1.7模型訓練.....................................80
3.1.8模型評價.....................................82
3.2基于Softmax回歸的多分類任務............................82
3.2.1數(shù)據(jù)集構建....................................83
3.2.2模型構建.....................................86
3.2.3損失函數(shù).....................................88
第3章線性分類67
3.1基于Logistic回歸的二分類任務.............................68
3.1.1數(shù)據(jù)集構建....................................69
3.1.2模型構建.....................................71
3.1.3損失函數(shù).....................................73
3.1.4模型優(yōu)化.....................................74
3.1.5評價指標.....................................77
3.1.6完善Runner類：RunnerV2...........................77
3.1.7模型訓練.....................................80
3.1.8模型評價.....................................82
3.2基于Softmax回歸的多分類任務............................82
3.2.1數(shù)據(jù)集構建....................................83
3.2.2模型構建.....................................86
3.2.3損失函數(shù).....................................88
3.2.4模型優(yōu)化.....................................89
3.2.5模型訓練.....................................91
3.2.6模型評價.....................................92
3.3實踐：基于Softmax回歸完成鳶尾花分類任務.....................92
3.3.1數(shù)據(jù)處理.....................................93
3.3.2模型構建.....................................95
3.3.3模型訓練.....................................96
3.3.4模型評價.....................................96
3.3.5模型預測.....................................97
3.4小結............................................97
第4章前饋神經網(wǎng)絡994.1神經元...........................................99
4.1.1凈活性值.....................................100
4.1.2激活函數(shù).....................................101
4.2基于前饋神經網(wǎng)絡的二分類任務............................104
4.2.1數(shù)據(jù)集構建....................................105
4.2.2模型構建.....................................105
4.2.3損失函數(shù).....................................108
4.2.4模型優(yōu)化.....................................109
4.2.5完善Runner類：RunnerV2_1..........................115
4.2.6模型訓練.....................................117
4.2.7模型評價.....................................118
4.3自動梯度計算和預定義算子...............................119
4.3.1利用預定義算子重新實現(xiàn)前饋神經網(wǎng)絡....................119
4.3.2完善Runner類：RunnerV2_2..........................120
4.3.3模型訓練.....................................122
4.3.4模型評價.....................................124
4.4優(yōu)化問題.........................................124
4.4.1參數(shù)初始化....................................124
4.4.2梯度消失問題..................................126
4.4.3死亡ReLU問題.................................129
4.5實踐：基于前饋神經網(wǎng)絡完成鳶尾花分類任務.....................130
4.5.1小批量梯度下降法................................130
4.5.2數(shù)據(jù)處理.....................................132
4.5.3模型構建.....................................133
4.5.4完善Runner類：RunnerV3...........................135
4.5.5模型訓練.....................................140
4.5.6模型評價.....................................142
4.5.7模型預測.....................................142
第5章卷積神經網(wǎng)絡145
5.1卷積............................................146
5.1.1二維卷積運算..................................146
5.1.2二維卷積算子..................................147
5.1.3卷積的變種....................................148
5.1.4帶步長和零填充的二維卷積算子........................149
5.1.5使用卷積運算完成圖像邊緣檢測任務.....................151
5.2卷積神經網(wǎng)絡的基礎算子................................152
5.2.1卷積層算子....................................152
5.2.2匯聚層算子....................................156
5.3基于LeNet實現(xiàn)手寫體數(shù)字識別任務..........................157
5.3.1數(shù)據(jù)集構建....................................158
5.3.2模型構建.....................................161
5.3.3模型訓練.....................................164
5.3.4模型評價.....................................165
5.3.5模型預測.....................................165
5.4基于殘差網(wǎng)絡的手寫體數(shù)字識別............................166
5.4.1模型構建.....................................167
5.4.2沒有殘差連接的ResNet18...........................173
5.4.3帶殘差連接的ResNet18.............................174
5.5實踐：基于ResNet18網(wǎng)絡完成圖像分類任務......................175
5.5.1數(shù)據(jù)處理.....................................176
5.5.2模型構建.....................................179
5.5.3模型訓練.....................................179
5.5.4模型評價.....................................181
5.5.5模型預測.....................................181
5.6小結............................................182
第6章循環(huán)神經網(wǎng)絡183
6.1循環(huán)神經網(wǎng)絡的記憶能力實驗..............................184
6.1.1數(shù)據(jù)集構建....................................185
6.1.2模型構建.....................................189
6.1.3模型訓練.....................................194
6.1.4模型評價.....................................196
6.2梯度爆炸實驗.......................................196
6.2.1梯度打印函數(shù)..................................197
6.2.2復現(xiàn)梯度爆炸問題................................197
6.2.3使用梯度截斷解決梯度爆炸問題........................199
6.3LSTM的記憶能力實驗..................................200
6.3.1模型構建.....................................202
6.3.2模型訓練.....................................204
6.3.3模型評價.....................................206
6.4實踐：基于雙向LSTM模型完成文本分類任務.....................207
6.4.1數(shù)據(jù)處理.....................................207
6.4.2模型構建.....................................212
6.4.3模型訓練.....................................214
6.4.4模型評價.....................................215
6.4.5模型預測.....................................216
6.5小結............................................216
第7章網(wǎng)絡優(yōu)化與正則化217
7.1小批量梯度下降法....................................218
7.2批大小的調整實驗....................................218
7.3不同優(yōu)化算法的比較分析................................221
7.3.1優(yōu)化算法的實驗設定..............................222
7.3.2學習率調整....................................229
7.3.3梯度估計修正..................................235
7.3.4不同優(yōu)化器的3D可視化對比..........................240
7.4參數(shù)初始化........................................244
7.4.1基于固定方差的參數(shù)初始化...........................244
7.4.2基于方差縮放的參數(shù)初始化...........................245
7.5逐層規(guī)范化........................................250
7.5.1批量規(guī)范化....................................250
7.5.2層規(guī)范化.....................................257
7.6網(wǎng)絡正則化方法.....................................259
7.6.1數(shù)據(jù)集構建....................................260
7.6.2模型構建.....................................260
7.6.3?1和?2正則化..................................266
7.6.4權重衰減.....................................268
7.6.5暫退法......................................269
7.7小結............................................272
第8章注意力機制273
8.1基于雙向LSTM和注意力機制的文本分類.......................274
8.1.1數(shù)據(jù)介紹.....................................275
8.1.2模型構建.....................................275
8.1.3使用加性注意力模型進行實驗.........................282
8.1.4使用點積注意力模型進行實驗.........................284
8.2基于雙向LSTM和多頭自注意力的文本分類實驗...................287
8.2.1自注意力模型..................................287
8.2.2基于LSTM和多頭自注意力的文本分類的模型構建..............297
8.2.3模型訓練.....................................299
8.2.4模型評價.....................................300
8.2.5模型預測.....................................301
8.3實踐：基于自注意力模型的文本語義匹配........................302
8.3.1數(shù)據(jù)集構建....................................303
8.3.2模型構建.....................................306
8.3.3模型訓練.....................................316
8.3.4模型評價.....................................317
8.3.5模型預測.....................................318
8.3.6注意力可視化..................................318
8.4小結............................................321
了解更多
我們?yōu)榇蠹姨峁?個名額，贈送這本書給大家！中獎的朋友請記得留下聯(lián)系方式和地址！歡迎大家掃碼參與！
