《MACHINE LEARNING YEARNING》翻译——开篇

《MACHINE LEARNING YEARNING》是Andrew NG最近出的本新书，目前正在陆续发布书的手稿。打算翻译一下这本书，并借机梳理一下机器学习方面的知识。翻译中的任何不足之处，欢迎大家不吝指出。

Table of Contents (draft)目录

1. Why Machine Learning Strategy 为什么需要机器学习策略

2. How to use this book to help your team 如何使用这本书来帮助你的团队

3. Prerequisites and Notation 预备知识和符号约定

4. Scale drives machine learning progress 规模促进了机器学习的发展

5. Your development and test sets 你的开发集和测试集

6. Your dev and test sets should come from the same distribution 你的开发集和测试机应该来自同一分布

7. How large do the dev/test sets need to be? 开发集和测试集多大合适

8. Establish a single-number evaluation metric for your team to optimize 为你的团队进行算法优化建立一个单一数字的评估指标

9. Optimizing and satisficing metrics 优化指标和满足指标

10. Having a dev set and metric speeds up iterations 有一个开发集和评估指标来加速迭代

11. When to change dev/test sets and metrics 何时更改开发/测试集和评估指标

12. Takeaways: Setting up development and test sets 小结：建立开发集和测试集

13. Build your first system quickly, then iterate

14. Error analysis: Look at dev set examples to evaluate ideas 错误分析：查看开发集样本来评估idea

15. Evaluate multiple ideas in parallel during error analysis 错误分析时并行评估多个想法

16. If you have a large dev set, split it into two subsets, only one of which you look at

17. How big should the Eyeball and Blackbox dev sets be?

18. Takeaways: Basic error analysis

19. Bias and Variance: The two big sources of error

20. Examples of Bias and Variance

21. Comparing to the optimal error rate

22. Addressing Bias and Variance

23. Bias vs. Variance tradeoff

24. Techniques for reducing avoidable bias

25. Techniques for reducing Variance

26. Error analysis on the training set

27. Diagnosing bias and variance: Learning curves

28. Plotting training error

29. Interpreting learning curves: High bias

30. Interpreting learning curves: Other cases

31. Plotting learning curves

32. Why we compare to human-level performance

33. How to define human-level performance

34. Surpassing human-level performance

35. Why train and test on different distributions

36. Whether to use all your data

37. Whether to include inconsistent data

38. Weighting data

39. Generalizing from the training set to the dev set

40. Addressing Bias and Variance

41. Addressing data mismatch

42. Artificial data synthesis

43. The Optimization Verification test

44. General form of Optimization Verification test

45. Reinforcement learning example

46. The rise of end-to-end learning

47. More end-to-end learning examples

48. Pros and cons of end-to-end learning

49. Learned sub-components

50. Directly learning rich outputs

51. Error Analysis by Parts

52. Beyond supervised learning: What’s next?

53. Building a superhero team - Get your teammates to read this

54. Big picture

55. Credits

书稿下载

Machine_Learning_Yearning_V0.5_01.pdf

Machine_Learning_Yearning_V0.5_02.pdf

Machine_Learning_Yearning_V0.5_03.pdf