Saved in:
Bibliographic Details
Main Authors: Biegun, Kai, Dolga, Rares, Cunningham, Jake, Barber, David
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2407.07239
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910634031972352
author Biegun, Kai
Dolga, Rares
Cunningham, Jake
Barber, David
author_facet Biegun, Kai
Dolga, Rares
Cunningham, Jake
Barber, David
contents Linear recurrent neural networks, such as State Space Models (SSMs) and Linear Recurrent Units (LRUs), have recently shown state-of-the-art performance on long sequence modelling benchmarks. Despite their success, their empirical performance is not well understood and they come with a number of drawbacks, most notably their complex initialisation and normalisation schemes. In this work, we address some of these issues by proposing RotRNN -- a linear recurrent model which utilises the convenient properties of rotation matrices. We show that RotRNN provides a simple and efficient model with a robust normalisation procedure, and a practical implementation that remains faithful to its theoretical derivation. RotRNN also achieves competitive performance to state-of-the-art linear recurrent models on several long sequence modelling datasets.
format Preprint
id arxiv_https___arxiv_org_abs_2407_07239
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle RotRNN: Modelling Long Sequences with Rotations
Biegun, Kai
Dolga, Rares
Cunningham, Jake
Barber, David
Machine Learning
Linear recurrent neural networks, such as State Space Models (SSMs) and Linear Recurrent Units (LRUs), have recently shown state-of-the-art performance on long sequence modelling benchmarks. Despite their success, their empirical performance is not well understood and they come with a number of drawbacks, most notably their complex initialisation and normalisation schemes. In this work, we address some of these issues by proposing RotRNN -- a linear recurrent model which utilises the convenient properties of rotation matrices. We show that RotRNN provides a simple and efficient model with a robust normalisation procedure, and a practical implementation that remains faithful to its theoretical derivation. RotRNN also achieves competitive performance to state-of-the-art linear recurrent models on several long sequence modelling datasets.
title RotRNN: Modelling Long Sequences with Rotations
topic Machine Learning
url https://arxiv.org/abs/2407.07239