《自動機器學習（AutoML）：方法、系統與挑戰》是清華大學出版社於2020年出版的書籍，作者是弗蘭克 · 亨特（Frank Hutter）。本書全面介紹自動機器學習，主要包含自動機器學習的方法、實際可用的自動機器學習系統及所面臨的挑戰。
基本介紹
- 中文名：自動機器學習（AutoML）：方法、系統與挑戰
- 作者：弗蘭克 · 亨特（Frank Hutter）
- 出版社：清華大學出版社
- 出版時間：2020年
- 定價：89 元
- ISBN：9787302552550
圖書簡介,目錄,作者簡介,譯者簡介,
圖書簡介
在自動機器學習方法中，本書涵蓋超參最佳化、元學習、神經網路架構搜尋三個部分，每一部分都包括詳細的內容介紹、原理解讀、具體運用方法和存在的問題等。勸汽此外，本書還具體介紹了現有的各種可用的AutoML系統，如Auto-sklearn、Auto-WEKA及Auto-Net等，並且本書最後一章詳細介紹了具有代表性的AutoML挑戰賽及挑戰賽結果背後所蘊含的理念，有助於從業者設計出自己的AutoML系統。 本書英文版是國際上第一本介紹自動機器學習的英文采烏剃書，內容全面且翔實，尤為重要的是涵蓋了最新的AutoML領域進展和難點。本書作者和譯者學術背景紮實歸愚蒸，保證了本書的內容質量。 對於初步研究者，本書可以作為其研究自動機器學習方法的背景知識和起點；對於工業界從業人員，本書全面介紹了AutoML系統及其實際套用要點；對於已經從事自動機器學習的研究者，本書可以提供一個AutoML最新研究成果和進展的概覽。總體來說，本書客群較為廣泛，既可以作為入門書，也可以作為專業人士的參考書。
目錄
自動機器學習方法
第1章 超參優紙灶芝化 ··································2
1.1 引言 ··············································2
1.2 問題定義 ·······································4
1.2.1 最佳化替代方案：集成與邊緣化 ·············5
1.2.2 多目標最佳化 ···········································5
1.3 黑盒超參最佳化 ·······························6
1.3.1 免模型的黑盒最佳化方法 ························6
1.3.2 貝葉斯最佳化 ···········································8
1.4 多保真度最佳化 ······························13
1.4.1 基於學習曲線預測的早停法 ··············14
1.4.2 基於Bandit的選擇方法 ·····················15
1.4.3 保章欠駝乘真度的適應性選擇 ··························17
1.5 AutoML的相關套用 ····················18
1.6 探討與展望 ··································20
1.6.1 基準測試和基線模型 ··························21
1.6.2 基於梯度的最佳化 ··································22
1.6.3 可擴展性 ·············································22
1.6.4 過擬合和泛化性 ··································23
1.6.5 任意尺度的管道構建 ··························24
參照紙企頸考文獻···············································25
第2章 元學習 ···································36
2.1 引言 ·············································36
2.2 模型評估中學習 ··························37
2.2.1 獨立於任務的推薦 ······························38
2.2.2 配置空間的設計 ··································39
2.2.3 配置遷移 ·············································39
2.2.4 學習曲線 ·············································42
2.3 任務道刪糊特性中學習 ··························43
2.3.1 元特徵 ·················································43
2.3.2 元特徵的學習 ·····································44
2.3.3 基於相似任務熱啟動最佳化過程 ···········46
2.3.4 元模型 ·················································48
2.3.5 管道合成 ·············································49
2.3.6 調優與否 ·············································50
2.4 先前模型中學習 ··························50
第一篇
XVI
2.4.1 遷移學習 ·············································51
2.4.2 針對神經網路的元學習 ······················51
2.4.3 小樣本學習 ·········································52
2.4.4 不止於監督學習 ··································54
2.5 總結 ·············································55
參考文獻···············································56
第3章 神經網路架構搜尋 ··················68
3.1 引言 ·············································68
3.2 搜尋空間 ······································69
3.3 搜尋策略 ······································73
3.4 性能評估策略 ······························76
3.5 未來方向 ······································78
參考文獻···············································80
自動機器學習系統
第4章 Auto-WEKA ···························86
4.1 引言 ·············································86
4.2 準備工作 ······································88
4.2.1 模型選擇 ·············································88
4.2.2 超參最佳化 ·············································88
4.3 算法選擇與超參最佳化結合
（CASH） ···································89
4.4 Auto-WEKA ·································91
4.5 實驗評估 ······································93
4.5.1 對比方法 ·············································94
4.5.2 交叉驗證性能 ·····································96
4.5.3 測試性能 ·············································96
4.6 總結 ·············································98
參考文獻···············································98
第5章 Hyperopt-sklearn ·················101
5.1 引言 ···········································101
5.2 Hyperopt背景 ····························102
5.3 Scikit-Learn模型選擇 ···············103
5.4 使用示例 ····································105
5.5 實驗 ···········································109
5.6 討論與展望 ································111
5.7 總結 ···········································114
參考文獻·············································114
作者簡介
弗蘭克•亨特，德國弗萊堡大學教授，機器學習實驗室負責人。主要研究統計機器學習、知識表示、自動機器學習及其套用，獲得第一屆（2015/2016）、第二屆（2018/2019）自動機器學習比賽的世界冠軍。
拉斯•特霍夫，美國懷俄明大學助理教授。主要研究深度學習、自動機器學習，致力於構建領先且健壯的機器學習系統，領導Auto-WEKA項目的開發和維護。
華昆•萬赫仁，荷蘭埃因霍溫理工大學助理教授。主要研究機器學習的逐步自動化，創建了共享數據開源平台OpenML.org，並獲得微軟Azure研究獎和亞馬遜研究獎。
譯者簡介
何明，中國科學技術大學博士，為上海交通大學電子科學與技術方向博士後研究人員、好未來教育集團數據中台人工智慧算法研究員。
劉淇，中國科學技術大學計算機學院特任教授，博士生導師，中國計算機學會大數據專家委員會委員，中國人工智慧學會機器學習專業委員會委員。
1.6.1 基準測試和基線模型 ··························21
1.6.2 基於梯度的最佳化 ··································22
1.6.3 可擴展性 ·············································22
1.6.4 過擬合和泛化性 ··································23
1.6.5 任意尺度的管道構建 ··························24
參考文獻···············································25
第2章 元學習 ···································36
2.1 引言 ·············································36
2.2 模型評估中學習 ··························37
2.2.1 獨立於任務的推薦 ······························38
2.2.2 配置空間的設計 ··································39
2.2.3 配置遷移 ·············································39
2.2.4 學習曲線 ·············································42
2.3 任務特性中學習 ··························43
2.3.1 元特徵 ·················································43
2.3.2 元特徵的學習 ·····································44
2.3.3 基於相似任務熱啟動最佳化過程 ···········46
2.3.4 元模型 ·················································48
2.3.5 管道合成 ·············································49
2.3.6 調優與否 ·············································50
2.4 先前模型中學習 ··························50
第一篇
XVI
2.4.1 遷移學習 ·············································51
2.4.2 針對神經網路的元學習 ······················51
2.4.3 小樣本學習 ·········································52
2.4.4 不止於監督學習 ··································54
2.5 總結 ·············································55
參考文獻···············································56
第3章 神經網路架構搜尋 ··················68
3.1 引言 ·············································68
3.2 搜尋空間 ······································69
3.3 搜尋策略 ······································73
3.4 性能評估策略 ······························76
3.5 未來方向 ······································78
參考文獻···············································80
自動機器學習系統
第4章 Auto-WEKA ···························86
4.1 引言 ·············································86
4.2 準備工作 ······································88
4.2.1 模型選擇 ·············································88
4.2.2 超參最佳化 ·············································88
4.3 算法選擇與超參最佳化結合
（CASH） ···································89
4.4 Auto-WEKA ·································91
4.5 實驗評估 ······································93
4.5.1 對比方法 ·············································94
4.5.2 交叉驗證性能 ·····································96
4.5.3 測試性能 ·············································96
4.6 總結 ·············································98
參考文獻···············································98
第5章 Hyperopt-sklearn ·················101
5.1 引言 ···········································101
5.2 Hyperopt背景 ····························102
5.3 Scikit-Learn模型選擇 ···············103
5.4 使用示例 ····································105
5.5 實驗 ···········································109
5.6 討論與展望 ································111
5.7 總結 ···········································114
參考文獻·············································114
作者簡介
弗蘭克•亨特，德國弗萊堡大學教授，機器學習實驗室負責人。主要研究統計機器學習、知識表示、自動機器學習及其套用，獲得第一屆（2015/2016）、第二屆（2018/2019）自動機器學習比賽的世界冠軍。
拉斯•特霍夫，美國懷俄明大學助理教授。主要研究深度學習、自動機器學習，致力於構建領先且健壯的機器學習系統，領導Auto-WEKA項目的開發和維護。
華昆•萬赫仁，荷蘭埃因霍溫理工大學助理教授。主要研究機器學習的逐步自動化，創建了共享數據開源平台OpenML.org，並獲得微軟Azure研究獎和亞馬遜研究獎。
譯者簡介
何明，中國科學技術大學博士，為上海交通大學電子科學與技術方向博士後研究人員、好未來教育集團數據中台人工智慧算法研究員。
劉淇，中國科學技術大學計算機學院特任教授，博士生導師，中國計算機學會大數據專家委員會委員，中國人工智慧學會機器學習專業委員會委員。
