Machine Learning 资讯 | 基于深度学习识别姑息治疗患者

“Hiring a Machine Learning engineer or Data Scientist in Silicon Valley is becoming like hiring a professional athlete. That’s how demanding it is” — The New York Times

image

基于深度学习识别姑息治疗患者

Stanford ML Group 建立了一个使用深度学习算法的程序，根据电子健康记录（Electronic Health Record ，EHR，包括病历、心电图、医疗影像等信息）数据确定在未来3-12个月高风险死亡的住院患者。这些病人的预警信息将发送给姑息治疗小组，这有助于姑息护理小组尽早介入、提供服务。

姑息治疗（Palliative Care ，在日本、中国台湾翻译为舒缓医学）起源于 hospice运动，最早起源于公元四世纪。根据世界卫生组织的定义，姑息治疗强调控制疼痛及患者有关症状，并对心理、社会和精神问题予以重视，目的是为病人和家属赢得最好的生活质量。

image

image

预测模型是一个 18 层的深度神经网络，输入参数为一个病人的 EHR 数据，输出为未来 3-12 个月死亡的概率。训练数据采用斯坦福医院 EHR 数据库中的历史数据，包含超过 200 万名患者的数据。EHR 数据包括患者过去 12 个月的诊断结论、治疗程序、处方和相关细节（经过脱敏和技术处理，以替代码的形式表示），所有数据被转换成 13654 维的特征向量。训练好的模型 AUROC 评分达到 0.93 ，交叉验证的平均精度为0.69 分。

对于机器学习系统来说，使用户可以根据预测结果采取行动，需要提供预测结果的详细解释，这点对于建立用户信心至关重要。Stanford 的程序可以自动生成一个报告，在病人的 EHR 数据中高亮突出对于预测结果具有重要影响因子的条目。

分类

• 图像处理 Image Manipulation
• 风格转换 Style Transfer
• 图像分类 Image Classification
• 脸部识别 Face Recognition
• 视频稳定化 Video Stabilization
• 目标检测 Object Detection
• 自动驾驶汽车 Self Driving Car
• 智能推荐 Recommendation Al
• 智能游戏 Gaming Al
• 智能下棋 Chess Al
• 智能医学 Medical Al
• 智能演说 Speech Al
• 智能音乐 Music Al
• 自然语言处理 Natural Language Processing
• 智能预测 Prediction

Mybridge AI 在 20000 篇关于创建机器学习应用的文章中挑选了前 50 名。从有实践经验的数据科学家那里学习是一个好方法，我们可以的分享中获得构建、运营机器学习应用的经验教训。50 篇文章大致可以分为 15 个主题，如下所示：

Recommended Learning

• The Beginner’s Guide to Building an Artificial Intelligence in Unity.
• Deep Learning and Computer Vision A-Z™: Learn OpenCV, SSD & GANs and create image recognition apps.

图像处理 Image Manipulation

• High-Resolution Image Synthesis and Semantic Manipulation with Conditional GANs
• Using Deep Learning to Create Professional-Level Photographs
• High Dynamic Range (HDR) Imaging using OpenCV (Python)

风格转换 Style Transfer

• Visual Attribute Transfer through Deep Image Analogy
• Deep Photo Style Transfer: A deep-learning approach to photographic style transfer that handles a large variety of image content while faithfully transferring the reference style
• Deep Image Prior

图像分类 Image Classification

• Feature Visualization: How neural networks build up their understanding of images
• An absolute beginner's guide to Image Classification with Neural Networks
• Background removal with deep learning

人脸识别 Face Recognition

• Large Pose 3D Face Reconstruction from a Single Image via Direct Volumetric CNN Regression
• Eye blink detection with OpenCV, Python, and dlib
• DEAL WITH IT in Python with Face Detection

视频稳定化 Video Stabilization

• Fused Video Stabilization on the Pixel 2 and Pixel 2 XL

目标检测 Object Detection

• How HBO’s Silicon Valley built “Not Hotdog” with mobile TensorFlow, Keras & React Native
• Object detection: an overview in the age of Deep Learning
• How to train your own Object Detector with TensorFlow’s Object Detector API
• Real-time object detection with deep learning and OpenCV

自动驾驶汽车 Self Driving Car

• Self-driving Grand Theft Auto V with Python : Intro [Part I] - Sentdex
• Recognizing Traffic Lights With Deep Learning: How I learned deep learning in 10 weeks and won $5,000

智能推荐 Recommendation AI

• Spotify’s Discover Weekly: How machine learning finds your new music
• Artwork Personalization at Netflix

智能游戏 Gaming AI

• MariFlow - Self-Driving Mario Kart w/Recurrent Neural Network
• OpenAI Baselines: DQN. Reproduce reinforcement learning algorithms with performance on par with published results.
• Reinforcement Learning on Dota 2 [Part II]
• Creating an AI DOOM bot
• Phase-Functioned Neural Networks for Character Control
• The Game Imitation: Deep Supervised Convolutional Networks for Quick Video Game AI - Stanford University
• Introducing: Unity Machine Learning Agents – Unity Blog

智能下棋 Chess AI

• Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm
• AlphaGo Zero: Learning from scratch | DeepMind
• How Does DeepMind's AlphaGo Zero Work?
• A step-by-step guide to building a simple chess AI

智能医学 Medical AI

• CheXNet: Radiologist-Level Pneumonia Detection on Chest X-Rays with Deep Learning
• Can you improve lung cancer detection? 2nd place solution for the Data Science Bowl 2017.
• Improving Palliative Care with Deep Learning - Andrew Ng
• Heart Disease Diagnosis with Deep Learning

智能演说 Speech AI

• Tacotron: A Fully End-to-End Text-To-Speech Synthesis Model - Data Scientists at Google
• Sequence Modeling with CTC
• Deep Voice: Real-time Neural Text-to-Speech - Baidu Silicon Valley AI Lab
• Deep Learning for Siri’s Voice: On-device Deep Mixture Density Networks for Hybrid Unit Selection Synthesis - Apple

智能音乐 Music AI

• Computer evolves to generate baroque music!
• Make your own music with WaveNets: Making a Neural Synthesizer Instrument

自然语言处理 Natural Language Processing

• Learning to communicate: Agents developing their own language - OpenAI Research
• Big Picture Machine Learning: Classifying Text with Neural Networks and TensorFlow
• A novel approach to neural machine translation - Facebook AI Research
• How to make a racist AI without really trying

预测 Prediction

• Using Machine Learning to Predict Value of Homes On Airbnb
• Engineering Uncertainty Estimation in Neural Networks for Time Series Prediction at Uber
• Using Machine Learning to make parking easier
• How to Predict Stock Prices Easily - Intro to Deep Learning #7

扩展阅读:《The Machine Learning Master》

• Machine Learning(一):基于 TensorFlow 实现宠物血统智能识别
• Machine Learning:机器学习项目
• Machine Learning:机器学习算法
• Machine Learning:机器学习书单
• Machine Learning:机器学习参考文集
• Machine Learning:机器学习技术与知识产权法
• Machine Learning:人工智能媒体报道集
• 数据可视化（三）基于 Graphviz 实现程序化绘图