The Multimodality of a Modeling Approach?

We have a hyper network that takes late maps, a tenuro network. This architecture is used to discover the manifold brain, like shapes, images, video. But at the same time, we don't learn manifolds over all these multimodisons at once. It's the mechanism that we can use if you wanted to maybe de proteins or like some other exotic thing, you could also just plug it in. There's no conbolutions or particular data augmentations or features that rely on a particular data typewrite. That's really interesting. We actually had vincent, who i think simon was t one of vincentscens ment's papers, be on the poncast