Showing 1–1 of 1 results for author: Seror, T

Spectral Diffusion Processes

Authors: Angus Phillips, Thomas Seror, Michael Hutchinson, Valentin De Bortoli, Arnaud Doucet, Emile Mathieu

Abstract: Score-based generative modelling (SGM) has proven to be a very effective method for modelling densities on finite-dimensional spaces. In this work we propose to extend this methodology to learn generative models over functional spaces. To do so, we represent functional data in spectral space to dissociate the stochastic part of the processes from their space-time part. Using dimensionality reducti… ▽ More Score-based generative modelling (SGM) has proven to be a very effective method for modelling densities on finite-dimensional spaces. In this work we propose to extend this methodology to learn generative models over functional spaces. To do so, we represent functional data in spectral space to dissociate the stochastic part of the processes from their space-time part. Using dimensionality reduction techniques we then sample from their stochastic component using finite dimensional SGM. We demonstrate our method's effectiveness for modelling various multimodal datasets. △ Less

Submitted 28 November, 2022; v1 submitted 28 September, 2022; originally announced September 2022.

Comments: 17 pages, 11 figures, Score-based Method Workshop at 36th Conference on Neural Information Processing Systems (NeurIPS 2022)