Logo Universität Regensburg Department Information Systems

Publication at the Conference on Artificial Intelligence

8. Januar 2024

Exciting news 🌟! The full-research paper “Piecewise Linear Transformation – Propagating Aleatoric Uncertainty in Neural Networks” by Thomas Krapf, Michael Hagn, Paul Miethaner, Dr. Alexander Schiller, Lucas Luttner and Bernd Heinrich has been accepted at #AAAI24. The paper develops a novel method for the propagation of probability distributions through neural networks. It will be presented in February 2024 at the 38th Annual AAAI Conference on Artificial Intelligence in Vancouver, Canada, one of the most prestigious AI conferences in the world.

The presented method harnesses the piecewise linear structure of certain neural networks to support the propagation of probability distributions with high exactness. The exactness is proven by means of theoretical error bounds and shown in an experimental evaluation, where the presented method outperforms competing methods on publicly available real-world classification and regression datasets. Paper and code available soon! #neuralnetworks #ai #machinelearning #aaai #computervision #uncertaintypropagation