- Tue, Apr 24, 2018
- 6:30 PM - 9:00 PM
- Sourced by
Added Apr 15 2018
Our April meeting will feature a talk from Jonathan Huggins, researcher at Harvard University, about scalable Bayesian inference. Join us to learn about this interesting topic and share your story with fellow Bayesians.
Scaling Bayesian inference by constructing approximating exponential families
Modern statistical applications create many challenges for standard approximate Bayesian inference algorithms. Existing algorithms have so far remained unsatisfactory because they either (1) fail to scale to large data sets, (2) provide limited approximation quality, or (3) fail to provide theoretical guarantees on the quality of inference. To simultaneously overcome these three possible limitations, we leverage the critical insight that in the large-scale setting, much of the data is redundant. Therefore, it is possible to compress data into a form that admits more efficient inference. We first discuss how exact compression is possible in the case of exponential families. Since typically models are not exponential families, we propose a method called PASS-GLM, which constructs an exponential family model that approximates the original model. The data is compressed by calculating the finite-dimensional sufficient statistics of the data under the exponential family. Time permitting, we describe a priori guarantees on the quality of the approximate posteriors produced by PASS-GLM.
Jonathan Huggins is a postdoctoral research fellow in Harvard's Department of Biostatistics. He recently completed his PhD at MIT's Computer Science and Artificial Intelligence Lab (CSAIL), where he was advised by Tamara Broderick. His research focuses on developing better algorithms for statistical inference and improving our theoretical understanding of existing inference algorithms.
* 6:30pm : Meet and greet. Networking
* 7pm : Talk by Jonathan Huggins + QA
* 8pm : Networking
* 8:45pm: End of the event
Note to attendees:
Please bring a photo id to check in with security when accessing the building.
This event is sponsored by QuantumBlack, a McKinsey Company