Become a data-driven Organization with Machine Learning
summary
Company builds a knowledge graph from 1.5 billion documents a day to extract who uses which product (triples) using NLP, prototypes in Python scikit-learn (SVM, decision trees, ridge) and runs Java in production, keeping infra cheap. Labeling costs run ~$10–$15 per example so semi‑supervised learning reduces labeling needs; data cleaning takes 50–90% of project time. Development treats sprints as experiments with target accuracy (e.g., 90%), and Lucene/Elasticsearch replace Mahout for real‑time recommendations.
AI-written summary · about
chapters (11)
- 0:00Introduction and speaker background
- 9:59What is data-driven
- 12:34Machine learning basics
- 25:40Coding examples in Python
- 30:58Data cleaning challenges
- 40:24Building machine learning models
- 47:42Interpretable machine learning
- 50:34What is a data scientist
- 1:00:19Machine learning tools and frameworks
- 1:10:03Neural networks and AI
- 1:29:33Q&A session
related talks
30mOperationalizing Machine Learning
39mDeveloping Machine Learning for Impact
29mSeveral People Are Tuning: Data and Machine Learning at
23mHow Can We Trust Machine Learning? Exploration, Evaluation...
23mCustom-made games with machine learning and big data
29mMachine Learning in Production: What to do after you’ve trained the models?
22mReal-world applications of machine learning: Beyond the hype
24mMachine Intelligence gets real