KAN (Kolmogorov-Arnold Networks): A Starter Guide 🐣

Kolmogorov-Arnold Networks (KAN) aim to innovate neural networks by leveraging the Kolmogorov-Arnold Representation Theorem. Unlike traditional Multi-Layer Perceptrons that rely on the Universal Approximation Theorem, KAN uses a learnable nonlinear function backed by B-splines for smoother computational efficiency. The guide covers necessary foundational topics, including neural networks, UAT, MLPs, Bezier curves, and B-splines, to explain the components and workings of KAN.