Reproduce “Degree-Optimized Cumulative Polynomial Kolmogorov-Arnold Networks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Mathew Vanherreweghe, Lirandë Pira, Patrick Rebentrost
Cumulative Polynomial Kolmogorov-Arnold Networks (CP-KAN) introduce a Quadratic Unconstrained Binary Optimization (QUBO) approach to efficiently select the optimal polynomial degree for each activation function in KANs. This method achieves competitive performance and high parameter efficiency, particularly in regression tasks involving smooth functions and temporal dependencies like financial time series prediction, while demonstrating enhanced training stability.
Key insights
- The optimal polynomial degree for learnable activation functions in KANs can be efficiently determined by formulating it as a Quadratic Unconstrained Binary Optimization (QUBO) problem.
- Adapting polynomial degrees for each activation function is crucial, as different parts of a network require varying levels of complexity to capture relationships effectively.
- Polynomial-based KANs, particularly with cumulative polynomial bases, are well-suited for approximating smooth functions and modeling temporal dependencies due to their numerical stability and inherent properties.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.