The ML system endeavors
summary
A computer vision pipeline built for smart buildings captures fisheye ceiling video, runs object detection and tracking on AWS (hundreds of ms per frame, ~5s initial latency), and turns detections into dashboards and real‑time alerts (reduced loitering alerts in a Bronx deployment). The team adds MLflow experiment logging, separates Dev/Stage/Prod, adopts code promotion with full retraining in prod for reproducibility, and plans Jetson edge deployments.
AI-written summary · about
chapters (17)
- 0:00ML-powered space management
- 1:53Real-time video analytics
- 3:22AI tracking for hospitality
- 4:27ML system journey
- 7:35Introducing MLOps
- 9:44Robust ML pipeline
- 14:07Code vs model promotion
- 17:45ML system benefits
- 18:50Jetson edge deployment
- 19:14Q&A on alerts and privacy
- 21:48Ethics and data privacy
- 23:14Measuring AI interactions
- 25:01Adding age/gender detection
- 25:43Training across stages
- 27:53Staging prevents breakage
- 28:38Latency in inference
- 30:20Monitoring system health
related talks
36mVideo analytics at scale: Deep Learning, Computer Vision, ML & Spark
21mML Infrastructure for Autonomous Vehicles
30mMachine learning for computer vision in pharmaceutical manufacturing
46mML Use Cases at Think Big Analytics
59mML and the IoT: Living on the Edge
28mVision, Big Data and AI
17mAI/ML demo derby
10mML-based Log Analysis for Faster Debugging