Of All StrIPEs: Investigating Structure-informed Positional Encoding for Efficient Music Generation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Agarwal, Manvi, Wang, Changhong, Richard, Gael
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917979601502208
author Agarwal, Manvi
Wang, Changhong
Richard, Gael
author_facet Agarwal, Manvi
Wang, Changhong
Richard, Gael
contents While music remains a challenging domain for generative models like Transformers, a two-pronged approach has recently proved successful: inserting musically-relevant structural information into the positional encoding (PE) module and using kernel approximation techniques based on Random Fourier Features (RFF) to lower the computational cost from quadratic to linear. Yet, it is not clear how such RFF-based efficient PEs compare with those based on rotation matrices, such as Rotary Positional Encoding (RoPE). In this paper, we present a unified framework based on kernel methods to analyze both families of efficient PEs. We use this framework to develop a novel PE method called RoPEPool, capable of extracting causal relationships from temporal sequences. Using RFF-based PEs and rotation-based PEs, we demonstrate how seemingly disparate PEs can be jointly studied by considering the content-context interactions they induce. For empirical validation, we use a symbolic music generation task, namely, melody harmonization. We show that RoPEPool, combined with highly-informative structural priors, outperforms all methods.
format Preprint
id arxiv_https___arxiv_org_abs_2504_05364
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Of All StrIPEs: Investigating Structure-informed Positional Encoding for Efficient Music Generation
Agarwal, Manvi
Wang, Changhong
Richard, Gael
Sound
Artificial Intelligence
Machine Learning
While music remains a challenging domain for generative models like Transformers, a two-pronged approach has recently proved successful: inserting musically-relevant structural information into the positional encoding (PE) module and using kernel approximation techniques based on Random Fourier Features (RFF) to lower the computational cost from quadratic to linear. Yet, it is not clear how such RFF-based efficient PEs compare with those based on rotation matrices, such as Rotary Positional Encoding (RoPE). In this paper, we present a unified framework based on kernel methods to analyze both families of efficient PEs. We use this framework to develop a novel PE method called RoPEPool, capable of extracting causal relationships from temporal sequences. Using RFF-based PEs and rotation-based PEs, we demonstrate how seemingly disparate PEs can be jointly studied by considering the content-context interactions they induce. For empirical validation, we use a symbolic music generation task, namely, melody harmonization. We show that RoPEPool, combined with highly-informative structural priors, outperforms all methods.
title Of All StrIPEs: Investigating Structure-informed Positional Encoding for Efficient Music Generation
topic Sound
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2504.05364