← Back to all news
arXiv Machine Learning By Justin Bunker, Mark Girolami, Hefin Lambley, Andrew M. Stuart, T. J. Sullivan

Autoencoders in Function Space

Read the original on arXiv Machine Learning →

The paper introduces function‑space versions of autoencoders (FAE) and variational autoencoders (FVAE), analysing their theoretical properties and practical deployment. It highlights that the FVAE objective is well‑defined only when the data distribution aligns with the generative model, a restriction often met when data come from stochastic differential equations. In contrast, the FAE objective remains well‑defined in many cases where FVAE fails, and both can be paired with neural operator architectures to enable tasks such as inpainting, super‑resolution, and generative modelling of scientific data.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Machine Learning.

Related stories

arXiv Machine Learning
Oct 2

Function-Space Transformer with Adaptive Anchors

The paper introduces the Function‑Space Transformer (FST), a neural framework that learns from functions by using a spatially adaptive continuous latent representation. FST places features at anchors whose positions are predicted from input samples and refines these anchors through recursive function‑space interactions, allowing the representation to adapt to each input rather than relying on a fixed grid. Experiments on PDEBench (Burgers and Darcy flow) and ImageNet‑1K show that FST outperforms or matches state‑of‑the‑art baselines such as Perceiver IO and the Vision Transformer while using fewer parameters.

By Guorui Sang, Pedram Rooshenas
llmscomputer-vision
More like this →