Explore a comprehensive analysis of shortcut learning in deep neural networks through this 49-minute video lecture. Delve into the framework for understanding out-of-distribution generalization failures in modern deep learning, where models learn false shortcuts present in training data. Examine why and when shortcut learning occurs, and discover recommendations for countering its effects. Investigate examples of shortcut learning, decision rules, cost functions, and taxonomy. Compare natural versus out-of-distribution data, with a focus on ImageNet. Gain insights from related issues in Comparative Psychology, Education, and Linguistics, suggesting shortcut learning as a common characteristic across biological and artificial learning systems.
Overview
Syllabus
Introduction
Shortcut Learning Example
Shortcut Learning Examples
Decision Rules
The Problem
Cost Function
Taxonomy
ImageNet
Natural vs OOD Data
Examples
Taught by
Yannic Kilcher
