Reading Group Discussion: Kolmogorov-Arnold Networks

Join us for an exciting discussion on Kolmogorov-Arnold Networks (KANs), a novel approach inspired by the Kolmogorov-Arnold representation theorem. Unlike traditional Multi-Layer Perceptrons (MLPs), which use fixed activation functions on nodes, KANs feature learnable activation functions on edges, parametrized as splines. This innovative change eliminates the need for linear weights, enhancing both accuracy and interpretability.

In this session, we will explore how KANs outperform MLPs, achieving comparable or better accuracy with significantly smaller models, particularly in data fitting and PDE solving. We will also explore the theoretical and empirical advantages of KANs, including faster neural scaling laws and intuitive visualization for better human interaction. Discover how KANs can pave the way for advancements in deep learning models.

IMPORTANT NOTE: We are considering converting this discussion into an open-source initiative. This would focus on hands-on experimentation and implementation of KANs. Committed attendees will have the opportunity to contribute to this project, making this session particularly beneficial for those interested in practical applications and collaborative research. We welcome those dedicated to contributing to join us in making significant advancements in deep learning models.

Paper: https://arxiv.org/abs/2404.19756