Salvato in:
Dettagli Bibliografici
Autori principali: Xu, Chenwei, Huang, Yu-Chao, Hu, Jerry Yao-Chieh, Li, Weijian, Gilani, Ammar, Goan, Hsi-Sheng, Liu, Han
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:https://arxiv.org/abs/2404.03830
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866914868706148352
author Xu, Chenwei
Huang, Yu-Chao
Hu, Jerry Yao-Chieh
Li, Weijian
Gilani, Ammar
Goan, Hsi-Sheng
Liu, Han
author_facet Xu, Chenwei
Huang, Yu-Chao
Hu, Jerry Yao-Chieh
Li, Weijian
Gilani, Ammar
Goan, Hsi-Sheng
Liu, Han
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2404_03830
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle BiSHop: Bi-Directional Cellular Learning for Tabular Data with Generalized Sparse Modern Hopfield Model
Xu, Chenwei
Huang, Yu-Chao
Hu, Jerry Yao-Chieh
Li, Weijian
Gilani, Ammar
Goan, Hsi-Sheng
Liu, Han
Machine Learning
Artificial Intelligence
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.
title BiSHop: Bi-Directional Cellular Learning for Tabular Data with Generalized Sparse Modern Hopfield Model
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2404.03830