RL in real life: Bringing reinforcement learning to the enterprise
summary
Argues that reinforcement learning replaces hard-coded business rules and runs in live systems, not just simulators, and that Microsoft builds services for real-time personalization and contextual optimization that must learn from every event. Describes constraints: reality hits with full complexity, baselines shift, you cannot rewind, exploration has consequences, and models must train and deploy in milliseconds — for example MSN News models updated every 15 minutes on every event. Proposes real-time adaptation and cross-system learning with ethics and monitoring built in.
AI-written summary · about
related talks
30mReconceptualizing Machine Learning for the Real-time World
07mRobots, houses & the IoT - Microsoft 's Brandon Rohrer on making things intelligent
19mLessons from Trillion Token Deployments at Fortune 500s
37mA practical perspective on training reinforcement learning bots.
15mAdaptive, AI-Driven Security Assessments
12mMachine Learning for Adaptive, Self-Optimizing Systems
39mChanging Realities - The Final Frontier for Machine Learning?
05mPractical insights into deep reinforcement learning