A Gated Residual Kolmogorov-Arnold Networks for Mixtures of Experts

TL;DR

This paper presents KAMoE, a new Mixture of Experts framework utilizing Gated Residual Kolmogorov-Arnold Networks to improve efficiency and interpretability, demonstrating superior performance in financial and real estate applications.

Contribution

Introduces GRKAN as an innovative gating mechanism for MoE, enhancing performance and interpretability over traditional gating functions.

Findings

KAMoE outperforms traditional MoE architectures across tasks.

GRKAN shows superior performance in LSTM-based sequential models.

Insights into trade-offs between model complexity and performance.

Abstract

This paper introduces KAMoE, a novel Mixture of Experts (MoE) framework based on Gated Residual Kolmogorov-Arnold Networks (GRKAN). We propose GRKAN as an alternative to the traditional gating function, aiming to enhance efficiency and interpretability in MoE modeling. Through extensive experiments on digital asset markets and real estate valuation, we demonstrate that KAMoE consistently outperforms traditional MoE architectures across various tasks and model types. Our results show that GRKAN exhibits superior performance compared to standard Gating Residual Networks, particularly in LSTM-based models for sequential tasks. We also provide insights into the trade-offs between model complexity and performance gains in MoE and KAMoE architectures.