Chaos and Pain in Machine Learning, and the ‘DevOps for ML Manifesto'
summary
Argues that ad hoc deployment causes “chaos and pain,” citing the oft‑quoted 80% of AI projects fail and that models can start spewing garbage in seconds without monitoring. Recommends devops for ml: automatic versioning of environment, code, data and metrics; a git‑like system for terabyte datasets — dot mesh (open source, ZFS‑backed) that diffs large data, tracks runs, diffs notebooks and stores models; CI/CD into Kubernetes and monitoring with Prometheus + Grafana to detect distribution drift and enable reproducibility.
AI-written summary · about
chapters (13)
- 0:00Start
- 0:18Background
- 3:41Key Problems Affecting Ai
- 6:04Auditing of Machine Learning
- 8:06Devops for Ml
- 9:15Four Different Things Required To Achieve Devops for Ml
- 10:06Collaboration
- 10:39Continuous Deployment
- 11:00Why Not Just Use Software Development Tools
- 14:34Version Control
- 16:22Pinning Down the Environment
- 16:45Versioning
- 20:59Model Monitoring
related talks
56mDevOps for Machine Learning
44mDevOps for Machine Learning - why is it different?
59mDevOps for Machine Learning Projects
30mChallenges of machine learning: The new ML lifecycle
21mMastering the DevOps Machine Learning Pipeline
01mMLOps: Applying DevOps to ML
41mSponsored Session: MLOps: Why DevOps Solutions Fall Short in the Machine Learning World
33mAI with a DevOps mindset - experimentation, sharing ..