Advanced Spatial Encoding

Background: The performance and fidelity of reconstructing spatiotemporal fields are constrained by the effectiveness of the spatial feature compression method used in latent-space models.

Question / Future Work: Investigating alternative physical field encoding techniques beyond the Convolutional Autoencoder (CAE), such as Variational Autoencoders (VAEs) or Vision Transformers, is required to potentially overcome limitations in reconstruction fidelity.