AI/ML Strategies & Techniques for a Post Pandemic World
summary
Claims that pandemic-driven distribution shifts break models and that simple defenses restore performance: a reported drift detector lifts accuracy from ~54% to ~69%, vaccine sentiment falls from ~70% to ~30% (fastText) and 50% to 15% (word methods). Uses concrete tools and methods: Google BigQuery/Google Data Studio for forecasting, ARMA and LSTM for time series, Facebook Prophet for changepoints and confidence intervals, statistical tests (ADF), and mitigation patterns like concept drift, adaptive sliding window retraining and multi-objective reinforcement learning for nonstationary policies.
AI-written summary · about
chapters (7)
- 0:00Introduction to post-pandemic AI
- 1:52Consumer behavior shifts explained
- 6:32Understanding concept drift basics
- 13:15Drift detection and mitigation strategies
- 22:04Reinforcement learning for concept drift
- 28:00MLOps and risk management practices
- 34:23Conclusion and future outlook
related talks
24mEngaging new and diverse audiences with ML
23mApplying AI and ML in the Current Pandemic World
34mTraining & Monitoring AI - Drift Detection
41mMachine Learning Models in a Dynamic Environment
12mRevolutionizing Predictive Analytics
37mAI for Good- Then and Now
56mData, AI, Hype and Fear - Recipe for Disaster?
46mThe AI dominated world – skills required for successfully navigating it