APS
APS Virtual Poster Showcase · 2020
What the AI Saw: Mismatches in Human Expectations and AI Classifier Performance
- Anne Linja
Applied Cognitive Science and Human Factors Graduate Student
Abstract
Deep Image Classifiers are limited, and easily fooled by image distortions that don’t impact human recognition. Humans need to understand the types of errors these classifiers make, and the conditions that lead to these errors. This study examines the accuracy and classification errors classifiers make, and how humans perceive them.
Media/Technology