Saved in:
Bibliographic Details
Main Authors: Kiruluta, Andrew, Lundy, Eric, Burity, Priscilla
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2505.07862
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912372749238272
author Kiruluta, Andrew
Lundy, Eric
Burity, Priscilla
author_facet Kiruluta, Andrew
Lundy, Eric
Burity, Priscilla
contents Existing sequence to sequence models for structured language tasks rely heavily on the dot product self attention mechanism, which incurs quadratic complexity in both computation and memory for input length N. We introduce the Graph Wavelet Transformer (GWT), a novel architecture that replaces this bottleneck with a learnable, multi scale wavelet transform defined over an explicit graph Laplacian derived from syntactic or semantic parses. Our analysis shows that multi scale spectral decomposition offers an interpretable, efficient, and expressive alternative to quadratic self attention for graph structured sequence modeling.
format Preprint
id arxiv_https___arxiv_org_abs_2505_07862
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Graph Laplacian Wavelet Transformer via Learnable Spectral Decomposition
Kiruluta, Andrew
Lundy, Eric
Burity, Priscilla
Computation and Language
Existing sequence to sequence models for structured language tasks rely heavily on the dot product self attention mechanism, which incurs quadratic complexity in both computation and memory for input length N. We introduce the Graph Wavelet Transformer (GWT), a novel architecture that replaces this bottleneck with a learnable, multi scale wavelet transform defined over an explicit graph Laplacian derived from syntactic or semantic parses. Our analysis shows that multi scale spectral decomposition offers an interpretable, efficient, and expressive alternative to quadratic self attention for graph structured sequence modeling.
title Graph Laplacian Wavelet Transformer via Learnable Spectral Decomposition
topic Computation and Language
url https://arxiv.org/abs/2505.07862