Learn how to detect silent failures in machine learning models without accessing target data in this 59-minute webinar. Explore the most common causes of ML model failure, including data and concept drift. Discover statistical and algorithmic tools for detecting and addressing these issues, their applications, and limitations. By the end, gain the ability to monitor ML models, detect performance drops without ground truth data, and understand data drift for effective problem-solving. The session includes a practical demo and Q&A to reinforce key concepts and techniques.
How to Detect Silent Failures in ML Models
Overview
Syllabus
Introduction
Data drift and concept drift
Performance estimation
Data and concept drift detection
Summary
Demo
QnA
Taught by
Data Science Dojo
