The Ethos of Process in ML
summary
Argues that machine learning fails when process is weak, not when tech is missing, contrasting CRISP-DM with a devops-style loop that must capture data bugs like bias or pandemic-driven drift; recommends concrete practices: versioned pipelines, CI/CD and Dockerized deploys, explicit documentation of dataset limits and GDPR considerations, regular two-month review cycles to retire or mature MVPs, and product/change managers plus user research to make models operational.
AI-written summary · about
chapters (7)
- 0:00Start
- 1:40Ethos of Process
- 3:03The Software Development Life Cycle
- 3:45Devops
- 9:40Getting Comfortable with Uncomfortable
- 30:38Final Advice
- 33:56Organizational Alignment
related talks
30mBuilding Teams and Culture that Support ML Innovation
56mDevOps for Machine Learning
27mCulture & Organization for Effective ML at Scale
06mMachine learning is about your data and deployment
21mWhy is Production ML so Hard?
44mAI workflows and teaming - how do teams deliver AI products?
15mHacking ML into Your Organization
01h 09mWorkshop: Product development for identifying and leveraging ML opportunities