Infinite-Dimensional Feature Interaction

TL;DR

This paper introduces InfiNet, a neural network architecture that leverages RBF kernels to enable feature interactions in an infinite-dimensional space, leading to improved performance over existing models.

Contribution

The paper presents InfiNet, a novel architecture that incorporates infinite-dimensional feature interactions via RBF kernels, surpassing finite-dimensional interaction limitations.

Findings

InfiNet achieves state-of-the-art results on benchmark datasets.

Infinite-dimensional interactions significantly improve model performance.

Kernel-based feature interactions outperform traditional finite-dimensional methods.

Abstract

The past neural network design has largely focused on feature representation space dimension and its capacity scaling (e.g., width, depth), but overlooked the feature interaction space scaling. Recent advancements have shown shifted focus towards element-wise multiplication to facilitate higher-dimensional feature interaction space for better information transformation. Despite this progress, multiplications predominantly capture low-order interactions, thus remaining confined to a finite-dimensional interaction space. To transcend this limitation, classic kernel methods emerge as a promising solution to engage features in an infinite-dimensional space. We introduce InfiNet, a model architecture that enables feature interaction within an infinite-dimensional space created by RBF kernel. Our experiments reveal that InfiNet achieves new state-of-the-art, owing to its capability to leverage infinite-dimensional interactions, significantly enhancing model performance.