English

BiSHop: Bi-Directional Cellular Learning for Tabular Data with Generalized Sparse Modern Hopfield Model

Machine Learning 2024-07-16 v2 Artificial Intelligence Machine Learning

Abstract

We introduce the \textbf{B}i-Directional \textbf{S}parse \textbf{Hop}field Network (\textbf{BiSHop}), a novel end-to-end framework for deep tabular learning. BiSHop handles the two major challenges of deep tabular learning: non-rotationally invariant data structure and feature sparsity in tabular data. Our key motivation comes from the recent established connection between associative memory and attention mechanisms. Consequently, BiSHop uses a dual-component approach, sequentially processing data both column-wise and row-wise through two interconnected directional learning modules. Computationally, these modules house layers of generalized sparse modern Hopfield layers, a sparse extension of the modern Hopfield model with adaptable sparsity. Methodologically, BiSHop facilitates multi-scale representation learning, capturing both intra-feature and inter-feature interactions, with adaptive sparsity at each scale. Empirically, through experiments on diverse real-world datasets, we demonstrate that BiSHop surpasses current SOTA methods with significantly less HPO runs, marking it a robust solution for deep tabular learning.

Cite

@article{arxiv.2404.03830,
  title  = {BiSHop: Bi-Directional Cellular Learning for Tabular Data with Generalized Sparse Modern Hopfield Model},
  author = {Chenwei Xu and Yu-Chao Huang and Jerry Yao-Chieh Hu and Weijian Li and Ammar Gilani and Hsi-Sheng Goan and Han Liu},
  journal= {arXiv preprint arXiv:2404.03830},
  year   = {2024}
}

Comments

31 pages; Code available at https://github.com/MAGICS-LAB/BiSHop

R2 v1 2026-06-28T15:44:43.684Z