Real-world applications of machine learning: Beyond the hype
summary
Covers concrete systems: Sift Science trains online fraud models across 6,000 websites, starting from a naive Bayes prototype and scaling to thousands of online models to handle class imbalance under 1% and thousands of events per second. Blue Apron uses four years of order data to forecast weeks ahead, schedule farm plantings and handle seasonality and recipe-driven spikes. Boxed.com ships auto-replenish emails and runs inventory-forecast and personalized shipment pilots. Mentions tools (scikit-learn, TensorFlow, Spark MLlib, cloud APIs), data engineering, fast prototypes (weeks), and ongoing human-in-the-loop tuning.
AI-written summary · about
chapters (8)
- 0:00Introduction to panel
- 1:45Company ML projects overview
- 6:00Starting ML projects challenges
- 11:15Training algorithms difficulty
- 14:30Data scarcity in ML
- 15:56Build vs buy ML solutions
- 19:49Human role in ML models
- 22:10Closing remarks and thanks
related talks
47mBuilding ML practices to address the top four use cases
54mMachine Learning explained with Examples
21mHow The Application Of Machine Learning Is Changing The Way Companies Use Data
27mA Look Inside Applied Machine Learning
30mMachine Learning: Beyond the Tutorial
24mMachine Intelligence gets real
53mMachine Learning in E‑commerce
06mInterview