Why Machine Learning Algorithms Fall Short...
summary
Argues that production ML needs more than algorithms: IBM Watson Machine Learning beta serializes models to JSON and ships a separate efficient serving engine rather than scoring in Spark, and an AutoML-like cognitive assistant uses bandit-style pipeline search to match a data scientist’s two-week effort in about two days. Shows concrete gains from using mathematical optimization (CPLEX) to add constraints — e.g. forcing pixels to be between 0 and 1 or ratings between 1 and 5 yields faster convergence and better RMSE, and gives a telco churn example where raw accuracy hurts business outcomes.
AI-written summary · about
chapters (5)
- 0:00Start
- 1:16What Is Machine Learning
- 5:43Scalable and Deployment
- 18:43Mathematical Optimization
- 22:18Example for Image Reconstruction
related talks
06mMachine learning is about your data and deployment
41mBeyond the Algorithm: What Makes Machine Learning Work?
08mConsiderations for producing the best AI and Machine Learning models
35mIntelligent Ruby: Getting Started with Machine Learning
29mMachine Learning in Production: What to do after you’ve trained the models?
13mOperationalizing machine learning for the enterprise
32mMen and Machine: An Artifical Love Affair?
15mMachine Learning: 5 Additional Metrics to Avoid High‑Performing but Useless Models