The Diffusion Encoder

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Premkumar, Akhil, Lucioni, Sarah
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866914563274833920
author Premkumar, Akhil
Lucioni, Sarah
author_facet Premkumar, Akhil
Lucioni, Sarah
contents We construct a new kind of encoder, leveraging the expressive power of diffusion models. In a traditional variational autoencoder, the encoder and decoder jointly negotiate a latent representation of the input. This is made possible by the reparameterization trick, which simplifies training at the cost of restricting the encoder to a simple family of distributions. Replacing this encoder with a diffusion model requires rethinking how the decoder pressure can be transmitted back to the encoder, given that they tend to update their internal estimates of the latent in opposing directions. We solve this problem with an alternating training scheme, inspired by the expectation-maximization algorithm. Our method enables more reliable synchronization between encoder and decoder, while preserving the simple and efficient training objective of standard diffusion models.
format Preprint
id arxiv_https___arxiv_org_abs_2605_13399
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle The Diffusion Encoder
Premkumar, Akhil
Lucioni, Sarah
Machine Learning
Information Theory
We construct a new kind of encoder, leveraging the expressive power of diffusion models. In a traditional variational autoencoder, the encoder and decoder jointly negotiate a latent representation of the input. This is made possible by the reparameterization trick, which simplifies training at the cost of restricting the encoder to a simple family of distributions. Replacing this encoder with a diffusion model requires rethinking how the decoder pressure can be transmitted back to the encoder, given that they tend to update their internal estimates of the latent in opposing directions. We solve this problem with an alternating training scheme, inspired by the expectation-maximization algorithm. Our method enables more reliable synchronization between encoder and decoder, while preserving the simple and efficient training objective of standard diffusion models.
title The Diffusion Encoder
topic Machine Learning
Information Theory
url https://arxiv.org/abs/2605.13399