Salvato in:
Dettagli Bibliografici
Autori principali: Hu, Pengfei, Zhang, Zhenrong, Ma, Jiefeng, Liu, Shuhang, Du, Jun, Zhang, Jianshu
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:https://arxiv.org/abs/2409.11887
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866912224982859776
author Hu, Pengfei
Zhang, Zhenrong
Ma, Jiefeng
Liu, Shuhang
Du, Jun
Zhang, Jianshu
author_facet Hu, Pengfei
Zhang, Zhenrong
Ma, Jiefeng
Liu, Shuhang
Du, Jun
Zhang, Jianshu
contents In recent years, visually-rich document understanding has attracted increasing attention. Transformer-based pre-trained models have become the mainstream approach, yielding significant performance gains in this field. However, the self-attention mechanism's quadratic computational complexity hinders their efficiency and ability to process long documents. In this paper, we present DocMamba, a novel framework based on the state space model. It is designed to reduce computational complexity to linear while preserving global modeling capabilities. To further enhance its effectiveness in document processing, we introduce the Segment-First Bidirectional Scan (SFBS) to capture contiguous semantic information. Experimental results demonstrate that DocMamba achieves new state-of-the-art results on downstream datasets such as FUNSD, CORD, and SORIE, while significantly improving speed and reducing memory usage. Notably, experiments on the HRDoc confirm DocMamba's potential for length extrapolation.
format Preprint
id arxiv_https___arxiv_org_abs_2409_11887
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle DocMamba: Efficient Document Pre-training with State Space Model
Hu, Pengfei
Zhang, Zhenrong
Ma, Jiefeng
Liu, Shuhang
Du, Jun
Zhang, Jianshu
Computation and Language
Artificial Intelligence
In recent years, visually-rich document understanding has attracted increasing attention. Transformer-based pre-trained models have become the mainstream approach, yielding significant performance gains in this field. However, the self-attention mechanism's quadratic computational complexity hinders their efficiency and ability to process long documents. In this paper, we present DocMamba, a novel framework based on the state space model. It is designed to reduce computational complexity to linear while preserving global modeling capabilities. To further enhance its effectiveness in document processing, we introduce the Segment-First Bidirectional Scan (SFBS) to capture contiguous semantic information. Experimental results demonstrate that DocMamba achieves new state-of-the-art results on downstream datasets such as FUNSD, CORD, and SORIE, while significantly improving speed and reducing memory usage. Notably, experiments on the HRDoc confirm DocMamba's potential for length extrapolation.
title DocMamba: Efficient Document Pre-training with State Space Model
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2409.11887