Diffusion Map Autoencoder

TL;DR

The paper introduces Diffusion-Map-AutoEncoder, a novel autoencoder combining diffusion-map encoding with Gaussian-process decoders for efficient, closed-form mappings and high-quality reconstructions.

Contribution

It presents a new autoencoder architecture that integrates diffusion maps with Gaussian-process decoders, enabling inductive learning and improved reconstruction quality.

Findings

Provides closed-form inductive mappings.

Achieves strong reconstruction performance.

Utilizes Nyström method for efficient diffusion-map encoding.

Abstract

Diffusion-Map-AutoEncoder (DMAE) pairs a diffusion-map encoder (using the Nystr\"om method) with linear or RBF Gaussian-Process latent mean decoders, yielding closed-form inductive mappings and strong reconstructions.