Overview
This project explores a hardware-aware implementation of Radial Basis Function (RBF) Kolmogorov-Arnold Networks, a neural architecture that learns distinct nonlinear mappings per node and layer instead of using a single shared activation function across the whole model. The goal is to investigate whether KAN-style models can offer a more flexible and interpretable alternative to standard feedforward networks for selected signal processing and regression tasks.
Details are on the way. This page will soon include architecture notes, implementation highlights, performance observations, and the design choices that guided the project.