Aaron Schein

Assistant Professor of Stats & Data Science at UChicago


Curriculum vitae


schein@uchicago.edu


Data Science Institute

University of Chicago

Chicago, IL



Poisson-Randomized Gamma Dynamical Systems


Conference paper


Aaron Schein, Scott LInderman, Mingyuan Zhou, David M. Blei, Hanna M. Wallach
Advances in Neural Information Processing Systems (NeurIPS), 2019

View PDF
Cite

Cite

APA   Click to copy
Schein, A., LInderman, S., Zhou, M., Blei, D. M., & Wallach, H. M. (2019). Poisson-Randomized Gamma Dynamical Systems. In Advances in Neural Information Processing Systems (NeurIPS).


Chicago/Turabian   Click to copy
Schein, Aaron, Scott LInderman, Mingyuan Zhou, David M. Blei, and Hanna M. Wallach. “Poisson-Randomized Gamma Dynamical Systems.” In Advances in Neural Information Processing Systems (NeurIPS), 2019.


MLA   Click to copy
Schein, Aaron, et al. “Poisson-Randomized Gamma Dynamical Systems.” Advances in Neural Information Processing Systems (NeurIPS), 2019.


BibTeX   Click to copy

@inproceedings{aaron2019a,
  title = {Poisson-Randomized Gamma Dynamical Systems},
  year = {2019},
  author = {Schein, Aaron and LInderman, Scott and Zhou, Mingyuan and Blei, David M. and Wallach, Hanna M.},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)}
}

Other materials: [Code] [Poster]
Abstract: This paper presents the Poisson-randomized gamma dynamical system (PRGDS), a model for sequentially observed count tensors that encodes a strong inductive bias toward sparsity and burstiness. The PRGDS is based on a new motif in Bayesian latent variable modeling, an alternating chain of discrete Poisson and continuous gamma latent states that is analytically convenient and computationally tractable. This motif yields closed-form complete conditionals for all variables by way of the Bessel distribution and a novel discrete distribution that we call the shifted confluent hypergeometric distribution. We draw connections to closely related models and compare the PRGDS to these models in studies of real-world count data sets of text, international events, and neural spike trains. We find that a sparse variant of the PRGDS, which allows the continuous gamma latent states to take values of exactly zero, often obtains better predictive performance than other models and is uniquely capable of inferring latent structures that are highly localized in time.

Share



Follow this website


You need to create an Owlstown account to follow this website.


Sign up

Already an Owlstown member?

Log in