The State of Production Machine Learning in 2023
summary
Claims that model lifecycle begins when deployed not when trained and that production ML shifts from model-centric to system- and data-centric; names practical components like experiment tracking, pipeline orchestration, model registry and monitoring plus observability and progressive rollouts; points to resources from The Institute for ethical Ai and machine learning and an ml secops top 10 covering poisoning and supply-chain risks; describes a built example—an embedding-based retrieval system with offline embedding build and online ranking—and stresses role shifts toward ML engineers and MLOps.
AI-written summary · about
chapters (7)
- 0:00Production ML intro
- 2:34Why production machine learning is challenging
- 10:20ML trends in production
- 19:50Security and monitoring
- 21:14Org trends in ML
- 26:51AI impact and responsibility
- 27:54Q&A
related talks
52mThe ML Lifecycle in Combat Conditions
47mMachine Learning in Production: An Intro to MLOps.
21mHow The Application Of Machine Learning Is Changing The Way Companies Use Data
30mChallenges of machine learning: The new ML lifecycle
25mML in Production–What does “Production” even mean?
29mDemystifying Machine Learning in Production: Reasoning about a Large-Scale ML Platform
14mGetting Machine Learning applications into production using MLOps and CD4ML
01h 05mMachine Learning Models and Productionization