[译] 15 大领域、50 篇文章，2018 年应当这样学习机器学习

整理 | 胡永波

根据《纽约时报》的说法，“在硅谷招募机器学习工程师、数据科学家的情形，越来越像NFL选拔职业运动员，没有苛刻的训练很难上场了。”毕竟，高达124472美元的平均年薪可不是谁想挣就能挣到的。

正如职业运动员每天都要训练一样，机器学习的日常练习也是工程师生涯得以大踏步前进的基本保障。仅2017年一年，机器学习领域总结此类实战经验的文章便已超过20000篇，该领域相关职位的热度自是可见一斑。

从中，我们筛选出50篇最好的经验和心得，囊括了机器学习在15大细分领域的各项典型应用：

1. 图像处理
2. 风格迁移
3. 图像分类
4. 面部识别
5. 视频稳像
6. 目标检测
7. 自动驾驶
8. 推荐系统
9. AI游戏
10. AI棋手
11. AI医疗
12. AI语音
13. AI音乐
14. 自然语言处理
15. 学习预测

当然，如果你只是一个刚要准备上手机器学习的新人，我们推荐你优先考虑以下两个高分实战课程：

A) AI游戏【推荐：5041；评分：4.7/5】

The Beginner’s Guide to Building an Artificial Intelligence in Unity

• 链接：https://www.udemy.com/artificial-intelligence-in-unity/

B) 计算机视觉【推荐：8161；评分：4.5/5】

Deep Learning and Computer Vision A-Z™: Learn OpenCV, SSD & GANs and create image recognition apps

• 链接：https://www.udemy.com/computer-vision-a-z/

而对具体的实战经验，接下来我们分领域一一来看：

图像处理

1、High-Resolution Image Synthesis and Semantic Manipulation with Conditional GANs

• GitHub：https://github.com/NVIDIA/pix2pixHD
• 论文：https://arxiv.org/abs/1711.11585
• 博客：https://tcwang0509.github.io/pix2pixHD/

来源：NVIDIA & UC Berkeley

2、Using Deep Learning to Create Professional-Level Photographs

• GitHub：https://github.com/google/creatism
• 论文：https://arxiv.org/abs/1707.03491
• 博客：https://research.googleblog.com/2017/07/using-deep-learning-to-create.html

来源：Google Research

3、High Dynamic Range (HDR) Imaging using OpenCV (Python)

• 项目：https://www.learnopencv.com/high-dynamic-range-hdr-imaging-using-opencv-cpp-python/
• 课程主页：https://courses.learnopencv.com/p/opencv-for-beginners

作者：Satya Mallick

风格迁移

4、Visual Attribute Transfer through Deep Image Analogy

• GitHub：https://github.com/msracver/Deep-Image-Analogy
• 论文：https://arxiv.org/abs/1705.01088

来源：微软研究院 & 上海交大

5、Deep Photo Style Transfer

• GitHub：https://github.com/luanfujun/deep-photo-styletransfer
• 论文：https://arxiv.org/abs/1703.07511

来源：Cornell University & Adobe

6、Deep Image Prior

• GitHub：https://github.com/DmitryUlyanov/deep-image-prior
• 论文：https://arxiv.org/abs/1711.10925
• 博客：https://dmitryulyanov.github.io/deep_image_prior

来源：SkolTech & Yandex & Oxford University

图像分类

7、Feature Visualization: How neural networks build up their understanding of images.

• 论文：https://distill.pub/2017/feature-visualization/
• 代码：https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/tutorials/deepdream/deepdream.ipynb
• 博客：https://colah.github.io/

来源：Google Brain

8、An absolute beginner’s guide to Image Classification with Neural Networks

• Github【4491收藏】：https://github.com/humphd/have-fun-with-machine-learning
• 中文版：https://github.com/humphd/have-fun-with-machine-learning/blob/master/README_zh-tw.md

来源：Mozilla

9、Background removal with deep learning

• 模型：https://towardsdatascience.com/background-removal-with-deep-learning-c4f2104b3157
• 部署：https://medium.com/@burgalon/deploying-your-keras-model-35648f9dc5fb

作者：Gidi Shperber

面部识别

10、Large Pose 3D Face Reconstruction from a Single Image via Direct Volumetric CNN Regression

• GitHub：https://github.com/AaronJackson/vrn
• 论文：https://arxiv.org/abs/1703.07834
• 博客：https://aaronsplace.co.uk/papers/jackson2017recon/
• Demo：https://cvl-demos.cs.nott.ac.uk/vrn/

作者：Aaron Jackson

11、Eye blink detection with OpenCV, Python, and dlib

• 项目：https://www.pyimagesearch.com/2017/04/24/eye-blink-detection-opencv-python-dlib/
• 论文：https://vision.fe.uni-lj.si/cvww2016/proceedings/papers/05.pdf

作者：Adrian Rosebrock

12、DEAL WITH IT in Python with Face Detection

• GitHub：https://github.com/burningion/automatic-memes
• 博客：https://www.makeartwithpython.com/blog/deal-with-it-generator-face-recognition/

作者：Kirk Kaiser

视频稳像

13、Fused Video Stabilization on the Pixel 2 and Pixel 2 XL

• 博客：https://research.googleblog.com/2017/11/fused-video-stabilization-on-pixel-2.html
• 测评：https://www.dxomark.com/google-pixel-2-reviewed-sets-new-record-smartphone-camera-quality/

来源：Google Research

目标检测

14、How HBO’s Silicon Valley built “Not Hotdog” with mobile TensorFlow and Keras

• 博客：https://medium.com/@timanglade/how-hbos-silicon-valley-built-not-hotdog-with-mobile-tensorflow-keras-react-native-ef03260747f3
• 项目：https://github.com/kmather73/NotHotdog-Classifier

作者：Tim Anglade

15、Object detection: an overview in the age of Deep Learning

• GitHub：https://github.com/tryolabs/luminoth
• 论文：https://tryolabs.com/blog/2017/08/30/object-detection-an-overview-in-the-age-of-deep-learning/

来源：Tryolabs

16、How to train your own Object Detector with TensorFlow’s Object

Detector API

• 博客：https://towardsdatascience.com/how-to-train-your-own-object-detector-with-tensorflows-object-detector-api-bec72ecfe1d9
• 数据集：https://github.com/datitran/raccoon_dataset
• 产品化：https://towardsdatascience.com/building-a-real-time-object-recognition-app-with-tensorflow-and-opencv-b7a2b4ebdc32
• 产品代码：https://github.com/datitran/object_detector_app

作者：Dat Tran

17、Real-time object detection with deep learning and OpenCV

• 实战：https://www.pyimagesearch.com/2017/09/18/real-time-object-detection-with-deep-learning-and-opencv/
• 入门：

①https://www.pyimagesearch.com/2017/09/11/object-detection-with-deep-learning-and-opencv/；②https://www.pyimagesearch.com/2016/01/04/unifying-picamera-and-cv2-videocapture-into-a-single-class-with-opencv/

③https://www.pyimagesearch.com/2017/08/21/deep-learning-with-opencv/

作者：Adrian Rosebrock

自动驾驶

18、Self-driving Grand Theft Auto V with Python : Intro [Part I]

• GitHub：https://github.com/sentdex/pygta5
• 视频：https://www.youtube.com/playlist?list=PLQVvvaa0QuDeETZEOy4VdocT7TOjfSA8a
• 博客：https://pythonprogramming.net/game-frames-open-cv-python-plays-gta-v/

作者：Sentdex

19、Recognizing Traffic Lights With Deep Learning: How I learned deep learning in 10 weeks and won $5,000

• GitHub：https://github.com/davidbrai/deep-learning-traffic-lights
• 博客：https://medium.freecodecamp.org/recognizing-traffic-lights-with-deep-learning-23dae23287cc
• 相关比赛：https://www.getnexar.com/challenge-1/

作者：David Brailovsky

推荐系统

20、Spotify’s Discover Weekly: How machine learning finds your new music

• 实战：https://hackernoon.com/spotifys-discover-weekly-how-machine-learning-finds-your-new-music-19a41ab76efe
• 演讲：https://www.youtube.com/watch?v=A259Yo8hBRs
• 相关博客：②https://notes.variogr.am/2012/12/11/how-music-recommendation-works-and-doesnt-work/

作者：Sophia Ciocca

21、Artwork Personalization at Netflix

• 博客：https://medium.com/netflix-techblog/artwork-personalization-c589f074ad76
• 论文：https://arxiv.org/abs/1003.5956
• 原理介绍：https://highscalability.com/blog/2017/12/11/netflix-what-happens-when-you-press-play.html

来源：Netflix

AI游戏

22、MariFlow — Self-Driving Mario Kart w/Recurrent Neural Network

• 文档：https://docs.google.com/document/d/1p4ZOtziLmhf0jPbZTTaFxSKdYqE91dYcTNqTVdd6es4
• 视频：https://www.youtube.com/watch?v=Ipi40cb_RsI

作者：SethBling

 

23、OpenAI Baselines: DQN

• GitHub：https://github.com/openai/baselines
• 项目主页：https://blog.openai.com/openai-baselines-dqn/

来源：OpenAI

24、Reinforcement Learning on Dota 2 [Part II]

• 博客：https://blog.openai.com/more-on-dota-2/
• 视频：https://openai.com/the-international/

来源：OpenAI

25、Creating an AI DOOM bot

• 博客：https://www.codelitt.com/blog/doom-ai/
• 工具：https://vizdoom.cs.put.edu.pl/

作者：Abel Castilla

26、Phase-Functioned Neural Networks for Character Control

• 博客：https://theorangeduck.com/page/phase-functioned-neural-networks-character-control
• 代码：https://theorangeduck.com/media/uploads/other_stuff/pfnn.zip
• 论文：https://theorangeduck.com/media/uploads/other_stuff/phasefunction.pdf
• 视频：https://theorangeduck.com/media/uploads/other_stuff/phasefunction.mov

作者：Daniel Holden

27、The Game Imitation: Deep Supervised Convolutional Networks for Quick Video Game AI

• 论文：https://arxiv.org/abs/1702.05663
• 视频：https://www.youtube.com/playlist?list=PLegUCwsQzmnUpPwVv8ygMa19zNnDgJ6OC

来源：Stanford

28、Introducing: Unity Machine Learning Agents

• GitHub：https://github.com/Unity-Technologies/ml-agents
• 博客：https://blogs.unity3d.com/cn/2017/09/19/introducing-unity-machine-learning-agents/
• 文档：https://github.com/Unity-Technologies/ml-agents/tree/master/docs

来源：Unity

AI棋手

29、Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm

• 论文：https://arxiv.org/abs/1712.01815
• 演讲：https://ktiml.mff.cuni.cz/~bartak/ui_seminar/talks/2017ZS/KarelHa_AlphaZero.pdf模型：https://deepmind.com/research/alphago/alphazero-resources/
• 相关实现：②https://web.stanford.edu/~surag/posts/alphazero.html

来源：Deepmind

30、AlphaGo Zero: Learning from scratch

• 博客：https://deepmind.com/blog/alphago-zero-learning-scratch/
• 论文：https://deepmind.com/documents/119/agz_unformatted_nature.pdf
• 棋谱：https://www.alphago-games.com/

来源：DeepMind

31、How Does DeepMind’s AlphaGo Zero Work?

• GitHub：https://github.com/llSourcell/alphago_demo
• 视频：https://www.youtube.com/watch?v=vC66XFoN4DE

作者：Siraj Raval

32、A step-by-step guide to building a simple chess AI

• GitHub：https://github.com/lhartikk/simple-chess-ai
• 博客：https://medium.freecodecamp.org/simple-chess-ai-step-by-step-1d55a9266977
• Wiki：https://chessprogramming.wikispaces.com/

作者：Lauri Hartikka

AI医疗

33、CheXNet: Radiologist-Level Pneumonia Detection on Chest X-Rays with Deep Learning

• 项目主页：https://stanfordmlgroup.github.io/projects/chexnet/
• 论文：https://arxiv.org/abs/1711.05225
• 博客：https://lukeoakdenrayner.wordpress.com/2017/11/18/quick-thoughts-on-chestxray14-performance-claims-and-clinical-tasks/

作者：吴恩达 & Stanford ML Group

34、Can you improve lung cancer detection? 2nd place solution for the Data Science Bowl 2017

• Kaggle：https://www.kaggle.com/c/data-science-bowl-2017
• GitHub：https://github.com/dhammack/DSB2017/
• 博客：https://juliandewit.github.io/kaggle-ndsb2017/

作者：Julian de Wit

35、Improving Palliative Care with Deep Learning

• 项目主页：https://stanfordmlgroup.github.io/projects/improving-palliative-care/
• 论文：https://arxiv.org/abs/1711.06402

作者：吴恩达 & Stanford ML Group

36、Heart Disease Diagnosis with Deep Learning

• GitHub：https://github.com/chuckyee/cardiac-segmentation
• 博客：https://blog.insightdatascience.com/heart-disease-diagnosis-with-deep-learning-c2d92c27e730
• 文章：https://chuckyee.github.io/cardiac-segmentation/

作者：Chuck-Hou Yee

AI语音

37、Tacotron: A Fully End-to-End Text-To-Speech Synthesis Model

• GitHub：https://github.com/Kyubyong/tacotron
• 论文：https://arxiv.org/abs/1703.10135
• 项目主页：https://google.github.io/tacotron/

来源：Google

38、Sequence Modeling with CTC

• GitHub：https://github.com/awni/speech
• 论文：https://distill.pub/2017/ctc/

作者：Awni Hannun

39、Deep Voice: Real-time Neural Text-to-Speech

• GitHub：https://github.com/israelg99/deepvoice
• 论文：https://arxiv.org/abs/1702.07825
• 博客：https://research.baidu.com/deep-voice-production-quality-text-speech-system-constructed-entirely-deep-neural-networks/

来源：百度

40、Deep Learning for Siri’s Voice: On-device Deep Mixture Density Networks for Hybrid Unit Selection Synthesis

• 博客：https://machinelearning.apple.com/2017/08/06/siri-voices.html

来源：Apple

AI音乐

41、Computer evolves to generate baroque music!

• 视频：https://www.youtube.com/watch?v=SacogDL_4JU
• 相关博客：https://karpathy.github.io/2015/05/21/rnn-effectiveness/

作者：Cary Huang

42、Make your own music with WaveNets: Making a Neural Synthesizer Instrument

• GitHub：https://github.com/tensorflow/magenta/tree/master/magenta/models/nsynth
• 论文：https://arxiv.org/abs/1704.01279
• 博客：https://magenta.tensorflow.org/nsynth-instrument

作者：Jesse Engelberg

自然语言处理

43、Learning to communicate: Agents developing their own language

• 博客：https://blog.openai.com/learning-to-communicate/
• 论文：https://arxiv.org/abs/1703.04908

来源：OpenAI

44、Big Picture Machine Learning: Classifying Text with Neural Networks and TensorFlow

• GitHub：https://github.com/dmesquita/understanding_tensorflow_nn
• 博客：https://medium.freecodecamp.org/big-picture-machine-learning-classifying-text-with-neural-networks-and-tensorflow-d94036ac2274

作者：Déborah Mesquita

45、A novel approach to neural machine translation

• GitHub：https://github.com/facebookresearch/fairseq
• 论文：https://arxiv.org/abs/1705.03122
• 博客：https://code.facebook.com/posts/1978007565818999/a-novel-approach-to-neural-machine-translation

来源： Facebook

46、How to make a racist AI without really trying

• Jupyter Python：https://gist.github.com/rspeer/ef750e7e407e04894cb3b78a82d66aed
• 博客：https://blog.conceptnet.io/2017/07/13/how-to-make-a-racist-ai-without-really-trying/

作者：Rob Speer

学习预测

47、Using Machine Learning to Predict Value of Homes On Airbnb

• 博客：https://medium.com/airbnb-engineering/using-machine-learning-to-predict-value-of-homes-on-airbnb-9272d3d4739d
• 中文：https://github.com/xitu/gold-miner/blob/master/TODO/using-machine-learning-to-predict-value-of-homes-on-airbnb.md

作者：Robert Chang

48、Engineering Uncertainty Estimation in Neural Networks for Time Series Prediction at Uber

• 论文：https://arxiv.org/abs/1709.01907
• 博客：https://eng.uber.com/neural-networks-uncertainty-estimation/

来源：Uber

49、Using Machine Learning to make parking easier

• 博客：https://research.googleblog.com/2017/02/using-machine-learning-to-predict.html
• 产品介绍：https://blog.google/products/maps/know-you-go-parking-difficulty-google-maps/

来源：Google

50、How to Predict Stock Prices Easily — Intro to Deep Learning #7

• 视频：https://www.youtube.com/watch?v=ftMq5ps503w
• 说明：https://github.com/llSourcell/How-to-Predict-Stock-Prices-Easily
• Demo：GitHub：https://github.com/erilyth/DeepLearning-Challenges/tree/master/Image_Classifier

作者：Siraj Raval

原文链接：
https://github.com/Mybridge/learn-machine-learning
https://medium.mybridge.co/learn-to-build-a-machine-learning-application-from-top-articles-of-2017-cdd5638453fc

你也许感兴趣的：

• Google 辟谣放弃 TensorFlow，它还活着！
• 做了 5 年机器学习研究，我发现了这 7 个真相
• 做了 5 年机器学习研究，我发现了这 7 个真相
• 亚马逊机器学习服务Amazon SageMaker 又添九项新功能！
• 用程序员的独特视角解读《隐秘的角落》
• 吴恩达的 CS229，有人把它浓缩成 6 张中文速查表
• 闲鱼是如何处理复杂背景内容识别的
• 如何判断你的项目是不是真的需要用到 AI？
• 谷歌刷新机器学习世界纪录！2 分钟搞定 ImageNet 训练
• 机器学习项目失败的9个原因