Past Meetup

Stan September Presentation Night

This Meetup is past

34 people went

CiBO Technologies

155 Second St. · Cambridge, ma

How to find us

Enter via the CiBO entrance on Second St. Talks will begin at 6:30pm -- please plan accordingly!

Location image of event venue


Hi all,

On Tuesday, September 18, we will have two Stan presentations at CiBO! The first talk will begin at 6:30pm. As per usual, there will be food and drinks for all attendees. Special thanks to CiBO for hosting and sponsoring this event!

The first presentation will be by Peter Komar, R&D engineer at Totient.

Title: Fitting normal model to censored data

I investigate the problem of fitting to censored data, which arise when data points outside of a "detection window" are replaced with the closest points on the window boundary. I present three solutions implemented in STAN dealing with (x,y) data that were generated from a multivariate 2D normal distribution, followed by a censoring step along the y axis. First, a truncated model, which discards censored points. Second, a censored model obtained by exact marginalization of the latent variables. Finally, a model which is generalizable to multiple censoring axes and higher dimensions.

The second presentation will be by Nathan Sanders, Chief Scientist at Warner Media Applied Analytics.

Title: Modeling the rate of public mass shootings in the US with Gaussian Processes in Stan

Abstract: Legislative action to decrease public mass shootings in the United States should be informed by knowledge of the long-term evolution of these events, but this evolution is difficult to measure because of the small number statistics associated with their occurrence. In work presented at StanCon'17 and published in Statistics and Public Policy, we developed a Bayesian model for the annualized rate of public mass shootings in the United States based on a Gaussian process with a time-varying mean function. Using NUTS, we explored the posterior consequences of different prior choices and the correlations between hyperparameters. This work provides a case study in the policy implications of implicit and explicit choices of prior information in model design and the utility of full Bayesian inference in evaluating the consequences of those choices.