Speaker
Simon Klüttermann
Description
In this poster, I show Polyra Swarms, a novel approach to machine learning that shifts focus from function approximation to shape approximation. While these swarms are still less developed, I show that they can still hold their own when compared to neural networks and on some tasks outperform them. I also present an automated abstraction mechanism that enhances generalization and interpretability by simplifying the learned swarm structure. By operating on principles fundamentally different from neural networks, Polyra Swarms offer new perspectives and open up fresh research directions in the design of learning systems.
