Automating Operations with ML
summary
Survey finds most ML-for-ops ideas fail in practice because labeled data is scarce, systems are dynamic, and engineering costs outweigh gains; anomaly detection on logs remains an open research problem, alert triage can eliminate two-thirds of false positives with modest true positive loss, and security detection needs active labeling and rapid retraining. The clear win is production auto scaling using historical patterns and periodic retraining which auto-scaling pays for itself; recommended tools are preprocessing/standardization, active learning, and frequent model refresh.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 0:38Agenda
- 1:20Identity
- 2:38Teleology
- 4:33Use Cases
- 6:11What Goes Wrong
- 7:26Evaluation
- 8:40Anomaly Detection
- 12:16Alert Triage
- 15:08Security
- 18:32Autoscaling
- 20:48Key Results
- 22:10Future Work
related talks
47mMachine Learning in Production: An Intro to MLOps.
13mMLOps: From AI Research to Production-Ready Systems
44mDevOps for Machine Learning - why is it different?
04mManaging Machines
09mAI/ML Won't Save you
13mHuman-in-the-Loop MLOps: Boost Model Performance by 40%
24mEmpowering High-Performance Teams with DevOps & Machine Learning
13mMLOps at Scale