Explainable Artificial Intelligence for Computer Vision

SoSe 2024

In recent years, deep learning has achieved remarkable advancements across various domains of computer vision. However, the increasing complexity of deep neural networks has led to a situation where users cannot understand their functioning, making their application in safety-critical fields like autonomous driving and medical image analysis challening. This lecture addresses the important topic of explainable artificial intelligence, which focuses on understanding the inner workings of modern vision models.

Organization

Course materials (Moodle) Link
TUCaN Link
Module no. 20-00-1191-vl
Event type Lecture
Being taught Summer semester 2024
Time Lecture: Tuesday 14:25:16:05
First class 16.04.2024 (see Moodle for updates).
Location S217/103
Lecturer Dr. Simone Schaub-Meyer, Robin Hesse
Exam see TUCaN

Content

  • Goals and challenges of explainability for computer vision
  • Interpretability of classical machine learning models
  • Global vs. local explanations
  • Post hoc explanations
  • Intrinsically explainable neural networks
  • Evaluation of explanations
  • Visualization techniques
  • Applications of explanations
  • XAI beyond classification

Basic knowledge of computer vision, machine learning, and deep learning. For example, acquired through the courses Computer Vision I, Introduction to Artificial Intelligence, Deep Learning: Architectures & Methods, and/or Statistical Machine Learning.