Applied Machine Learning Playbook: From Idea to Production
summary
Practical checklist of what actually works: simulate scarce sensors by staging activity in an empty hospital room and doubled the dataset in a day; prefer extra labeled examples to exotic few‑shot models; track training data with timestamp hashes so experiments compare apples to apples; enforce data and label sanity tests, unit tests on model tensors, and distribution checks; monitor end‑to‑end latency at the 95/99 percentiles and business KPIs; use Wizard of Oz human‑in‑the‑loop bootstrapping and careful crowdsourcing like Mechanical Turk with review voters.
AI-written summary · about
chapters (12)
- 0:00Intro
- 0:35Collecting data
- 7:56Managing data
- 11:22Data pipelines
- 23:51Training the model
- 25:27Monitoring
- 28:26Examples
- 32:05Example
- 32:59Simple heuristics
- 33:49Wizard of Oz experiments
- 34:18Hidden human
- 35:31Outro
related talks
41mBeyond the Algorithm: What Makes Machine Learning Work?
32mAn Only One Step Ahead Guide for Machine Learning Projects
27mA Look Inside Applied Machine Learning
30mFrom Black Box to a Known Quantity: How to Build Predictable, Reliable ML-based
42mMasterclass: The right path to ML
28mLeading machine learning research journeys with happy (into-production) endings
08mConsiderations for producing the best AI and Machine Learning models
27mBuild Your Own ML Data Feedback Loop