Research

My research focuses on approximate Bayesian inference.

My primary application domain lies in variational inference for large-scale astronomy datasets. A current project involves deep generative modeling for telescope images of galaxies. I combine ideas from deep learning, Bayesian probabilistic modeling, and physics knowledge for the detection of gravitational lensing.

I also work on methods for efficient quantification of Bayesian sensitivity to prior specifications in the context of variational inference.

Publications

Liu R., McAuliffe J. D., Regier J. “Variational Inference for Deblending Crowded Starfields.” https://arxiv.org/abs/2102.02409.

Liu R., Regier J., Tripuraneni N., Jordan M. I., McAuliffe J. D. “Rao-Blackwellized Stochastic Gradients for Discrete Distributions.” International Conference on Machine Learning. June 2019. https://arxiv.org/abs/1810.04777.

Giordano R., Stephenson W., Liu R., Jordan M. I., Broderick T. “A Swiss Army Infinitesimal Jackknife .” International Conference on Artificial Intelligence and Statistics. April 2019. https://arxiv.org/abs/1806.00550.
– Selected for Notable Paper Award and oral presentation.

Liu R., Giordano R., Jordan M. I., Broderick T. “Evaluating Sensitivity to the Stick Breaking Prior in Bayesian Nonparametrics.” NIPS, All of Bayesian Nonparametrics Workshop. December 2018. https://arxiv.org/abs/1810.06587.
– Selected for ISBA@NIPS Award and contributed talk.

Giordano R., Liu, R., Varoquaux N., Jordan M. I., Broderick T. “Measuring Cluster Stability for Bayesian Nonparametrics Using the Linear Bootstrap.” NIPS, Advances in Approximate Bayesian Inference Workshop. December 2017. https://arxiv.org/abs/1712.01435.