Leading machine learning research journeys with happy (into-production) endings
summary
Shows concrete playbooks and outcomes: a team ships live‑streaming AI into production using test-driven machine learning (API-first blocks, unit tests, business-metric functions), enforces weekly pull requests and peer reviews, and stages inner releases so models integrate without exposing users. Use an automatic fallback per feature to avoid full rollbacks, plan data collection a quarter ahead, and measure impact with dashboards; the talk cites that only 54% of projects move from pilot to production and gives methods that reduce that gap.
AI-written summary · about
chapters (13)
- 0:00ML research to production journeys
- 2:55Why ML projects fail production
- 5:22Balancing incremental vs big swings
- 9:11Planning ML journey with targets
- 12:18Test-driven ML development
- 14:46Setting stop criteria
- 16:11Agile research with weekly deliverables
- 18:20Data-driven research early planning
- 20:50Avoiding ML feature rollbacks
- 22:37Inner releases for integration
- 24:02Aligning expectations before release
- 25:04Gradual rollout and monitoring
- 26:07Celebrating ML success
related talks
27mA Look Inside Applied Machine Learning
21mHow The Application Of Machine Learning Is Changing The Way Companies Use Data
35mHow to deploy machine learning models to production (frequently and safely)
08mConsiderations for producing the best AI and Machine Learning models
53mUnleashing the Black Box: Testing Machine Learning Systems
22mFrom experiment to production - a journey of a machine learning model
14mGetting Machine Learning applications into production using MLOps and CD4ML
36mApplied Machine Learning Playbook: From Idea to Production