AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
Is Diffusion a Tool for Scientific Applications?
The question of which specific architecture and which specific framework can evolve over time, it was GANs a few years ago. And I do think that's important for the scientific applications too, because in sciences we have a lot of constraints. For instance, diffusion models are slow to sample because they have to gradually go from a Gaussian distribution to the distribution of images. One of the frameworks we've developed is to do a decoding in parallel.