M$^2$VAE - Derivation of a Multi-Modal Variational Autoencoder Objective from the Marginal Joint Log-Likelihood

This work provides a theoretical foundation for multi-modal VAEs, which is incremental as it builds on existing VAE theory.

The paper tackles the problem of deriving a trainable objective for multi-modal variational autoencoders by presenting an in-depth derivation of the evidence lower bound from the marginal joint log-likelihood, resulting in a framework for training such models.