A Gated Residual Kolmogorov-Arnold Networks for Mixtures of Experts
Hugo Inzirillo, Remi Genet
TL;DR
The paper addresses how to improve Mixture of Experts gating for complex sequential data by introducing KAMoE, which leverages Gated Residual Kolmogorov-Arnold Networks (GRKAN) as the gating mechanism. The approach combines learnable inputs, GRKAN-based gating, multiple expert networks, and a weighted aggregation to form predictions, with two experiments covering sequential crypto-time series and non-sequential housing data. Results show that GRKAN generally outperforms standard gating, and KAMoE often surpasses traditional MoE across model types—especially with LSTM-based architectures—while highlighting context-dependent gains and trade-offs in model complexity. The work offers a practical, open-source framework that improves MoE performance in finance and real estate forecasting, contributing to more efficient and interpretable adaptive architectures.
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.