multi_modal_ebm

VAE EBM

Two sepratae VAEs for images and text data are trained using standard ELBO objective. I then utilize the latent space embeddings of such multi-modalities to jointly train EBMs in an unsupervised manner.