We need several versions of the latent feature space: - Single Modality to single modality (Single-latent to single-output) - Multi-modality to single modality (multi-latent to single-output) - "Trainable" latent feature space
We need several versions of the latent feature space: