Geometric Autoencoder with Manifold Learning

Autoencoders are widely recognized as non-probabilistic learning models for extracting useful information from data. Most autoencoder models assume a Euclidean geometry for the underlying nature of the data. However, recent advancements in geometric learning suggest that incorporating curvature information of the intrinsic manifold of data may yield richer representations. In this work, we investigate the performance of a learning method that embeds data under curved geometric constraints. Our method assumes that the data manifold consists of both curved and Euclidean spaces. Experimental results demonstrate that our method achieves performance comparable to state-of-the-art techniques