arxiv News score 23

SechKAN: Kolmogorov-Arnold Networks with Hyperbolic Secant Functions

AI Digest

Kolmogorov-Arnold网络双曲正割函数参数效率图像分类PDE建模

本文提出SechKAN,基于双曲正割函数改进Kolmogorov-Arnold网络,通过参数缩减保持模型规模,实验证明其在函数拟合、PDE建模和图像分类中表现优异。

SechKAN introduces hyperbolic secant functions to KANs, reducing parameters while maintaining MLP-scale models. Experiments show superior performance in function fitting, PDE surrogate modeling, and image classification.

Key points

  • 采用双曲正割函数实现平滑局部响应与稳定梯度 Uses sech functions for smooth localized responses and stable gradients
  • 通过1D线性投影降低参数量保持模型规模 Reduces parameters via 1D linear projection for MLP-scale models
  • 在图像分类任务中超越现有KAN变体 Outperforms existing KAN variants in image classification
  • 计算成本高于MLP但参数效率更优 Higher computational cost than MLPs but better parameter efficiency
  • 公开代码便于复现与应用 Public code enables reproducibility and application

Takeaway: SechKAN通过函数设计平衡参数效率与性能,为复杂建模提供新思路。 / SechKAN balances parameter efficiency and performance through function design, offering new approaches for complex modeling.

Why it matters 该研究为神经网络设计提供创新函数选择,尤其适合需要参数效率与模型性能平衡的场景。

View original ↗ Back to hot list

This page is an aggregated digest from arxiv; content and hot-score data come from public sources. Copyright belongs to the original authors. We link to originals with nofollow and never republish full text.