Planning for Mistakes -- ML in Production Course @ CMU -- Lecture 4 скачать в хорошем качестве

Planning for Mistakes -- ML in Production Course @ CMU -- Lecture 4

This is the fourth lecture of the Machine Learning in Production course (17-645/11-695) at Carnegie Mellon University by Claire Le Goues and Christian Kaestner, discussing system-level designs that can mitigate mistakes from ML models, as well as how to anticipate what problems to design for with hazard analysis and fault tree analysis. Lecture recorded over Zoom after a snow storm. Course website: https://mlip-cmu.github.io/s2026/ 00:00 Introduction 01:31 Planning for Mistakes: Anticipation and Mitigation 02:20 Design Strategy: Human in the Loop 09:13 Design Strategy: Undoable Actions 11:00 Design Strategy: Guardrails 14:29 Design Strategy: Mistake Detection and Recovery 17:53 Design Strategy: Redundancy 19:10 Design Strategy: Containment and Isolation 23:22 Breakout Session 30:15 Hazard Analysis 49:25 Fault Tree Analysis Explained 52:56 Fault Tree Analysis Example and Questions 01:03:34 Wrap Up