Andreas Schauer, Alina El-Keilany and Iuliia Grebeshok explained how a computer turns an image into numbers, how neural networks learn, and why convolutional neural networks are so effective at recognizing patterns.
The highlight was the hands-on part. Participants trained their own image classifier with a webcam and then deliberately tried to trick it. This led to a discussion of AI's weak points, including biased training data, adversarial attacks and privacy, and of how explainable AI can help us understand what a model is actually looking at.
Many thanks to all participants and to the organizers of Tag der KI!
