Linear Attention as Bayesian Inference

Fuente: Zenodo
Gespeichert in:
Bibliographische Detailangaben
1. Verfasser: Frimane, Âzeddine
Format: Recurso digital
Sprache:Englisch
Veröffentlicht: Zenodo 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866901742876098560
author Frimane, Âzeddine
author_facet Frimane, Âzeddine
contents <p>This paper shows that the recurrent update used in linear attention models is mathematically equivalent to Bayesian inference under a Dependent Dirichlet Process. The forgetting rate, the input-dependent gate, and the multi-head structure each receive a clean probabilistic interpretation. The result is exact and holds by construction, not as an approximation.</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_18975350
institution Zenodo
language eng
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle Linear Attention as Bayesian Inference
Frimane, Âzeddine
linear attention, Dirichlet process, Bayesian nonparametrics, recurrent neural networks, state space models, exponential smoothing, hierarchical Dirichlet process, sequence modeling
<p>This paper shows that the recurrent update used in linear attention models is mathematically equivalent to Bayesian inference under a Dependent Dirichlet Process. The forgetting rate, the input-dependent gate, and the multi-head structure each receive a clean probabilistic interpretation. The result is exact and holds by construction, not as an approximation.</p>
title Linear Attention as Bayesian Inference
topic linear attention, Dirichlet process, Bayesian nonparametrics, recurrent neural networks, state space models, exponential smoothing, hierarchical Dirichlet process, sequence modeling
url https://doi.org/10.5281/zenodo.18975350