Plenary Talk WIRN 2026
Neural Mesh, as an alternative to Neural Network
Abstract
The presentation introduces the concept and architecture of the Neural Mesh. The Neural Mesh is an alternative to conventional neural networks. Its primary motivation is that, in many contemporary AI developments, merely achieving highly accurate learning of information is no longer sufficient. Instead, increasing emphasis is placed on learning architectures that yield interpretable and transparent representations, enabling meaningful evaluation, facilitating subsequent design, and allowing the quality of the solution to be formally verified.
The Neural Mesh architecture presented in this talk represents the learned information in a manner that inherently supports these capabilities.
Biography
Peter Baranyi, a notable Hungarian scholar, has made significant contributions to the fields of non-linear control theory and modeling. Among his key inventions is the TP model transformation, a sophisticated form of higher-order singular value decomposition for continuous functions. This transformation is crucial in the development of nonlinear control design theories and facilitates new optimisation techniques. Baranyi’s scientific achievements have been recognised with several prestigious awards, including the Investigator Award from Sigma Xi, the Kimura Award, and the International Dennis Gabor Award. He has published over 100 journal papers and authored four books, resulting in an h-index of 54.