A classifier scores highly in testing and fails in deployment. Investigation shows it was detecting an artefact of how the positive cases were collected rather than the condition itself. What is this called, and which failure family does it belong to?
Written from the published competencies and our own syllabus, written from the published curricula of BlueDot Impact, the Center for AI Safety, DeepMind and Stanford. Not actual exam questions, and not affiliated with or endorsed by Product Digest.