MLOps — Understanding and Fighting Model Degradation in Production
summary
How to monitor and diagnose machine-learning model performance after deployment: concrete techniques for tracking production metrics and identifying common causes of degradation such as data drift, changes in usage, bias, and model drift. The talk reviews evaluation and monitoring practices and remediation actions (retraining, data fixes, bias mitigation, etc.) aimed at practitioners with beginner-to-intermediate ML/MLOps knowledge.
AI-written summary · about
chapters (6)
- 0:00Model degradation intro
- 3:53Why models degrade
- 7:23Data vs concept drift
- 10:12Detecting degradation
- 13:21Actions against degradation
- 14:45Prevention best practices
related talks
39mAn SRE Approach to Monitoring ML in Production
47mMachine Learning in Production: An Intro to MLOps.
10mAvoiding the Oops with MLOps
01mMLOps: Applying DevOps to ML
13mMLOps: From AI Research to Production-Ready Systems
47mMLOps tools to automate ML deployment
18mReliable Maintenance of Machine Learning Models
41mTHE STATE OF MLOPS - MACHINE LEARNING IN PRODUCTION AT ENTERPRISE SCALE