A recent pre-print of research published in ArXiv might supercharge the neural networks that we know about today.
These researchers introduced Kolmogorov-Arnold Networks (KANs) which are promising alternatives to the currently dominant Multi-Layer Perceptrons (MLPs) architecture.
This story is a deep dive into what KANs are, how they work, why they might be a great alternative to Multi-Layer Perceptron (MLPs) in the near future, and how to train one from scratch.
