Riemannian Score-Based Generative Modelling

Authors: Valentin De Bortoli, Emile Mathieu, Michael Hutchinson, James Thornton, Yee Whye Teh, Arnaud Doucet

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate our approach on a variety of manifolds, and in particular with earth and climate science spherical data. ... In this section we benchmark the empirical performance of RSGMs along with other manifold-valued methods introduced in Sec. 5. ... We start by evaluating RSGMs on a collection of simple datasets, each containing an empirical distribution of occurrences of earth and climate science events on the surface of the earth. ... In Fig. 3, we observe that RSGMs are able to fit well the target distribution even in high dimension... From Table 5 we observe that, RSGMs perform consistently...
Researcher Affiliation Academia Valentin De Bortoli , Émile Mathieu , Michael Hutchinson , James Thornton , Yee Whye Teh , Arnaud Doucet equal contribution. Dept. of Computer Science ENS, CNRS, PSL University Paris, France. Dept. of Statistics, University of Oxford, Oxford, UK. 36th Conference on Neural Information Processing Systems (Neur IPS 2022).
Pseudocode Yes Algorithm 1 GRW (Geodesic Random Walk) ... Algorithm 2 RSGM (Riemannian Score-Based Generative Model)
Open Source Code Yes Code is available at https://github.com/vdebor/Riemannian-SGM
Open Datasets Yes earthquakes (NGDC/WDS), floods (Brakenridge, 2017) and wild fires (EOSDIS, 2020).
Dataset Splits Yes We use a 80-20% train-test split for the earth and climate science datasets and the SO3(R) dataset, and 5-fold cross-validation for the torus dataset.
Hardware Specification Yes All experiments were run on a single NVIDIA A100 GPU.
Software Dependencies No Our implementation is built on Jax (Bradbury et al., 2018) and Geomstats (Miolane et al., 2020a,b).
Experiment Setup Yes The score network is a 3-layer MLP of 128 hidden units with a Swish activation. We use the Adam optimizer (Kingma and Ba, 2015) with a learning rate of 1e-3 and a batch size of 128. We train for 5000 epochs.