The COVID-19 pandemic has affected everyone’s life and sheds some light on the shortcomings in monitoring, tracking and modeling the spread of diseases. As a result, many funding mechanisms across Canada and the world have taken additional action to support research in these areas. The goal of these programs is not only to help curb COVID-19, but also long-term resiliency against infectious diseases and pandemics.
As part of these efforts, the Programming Languages for Artificial Intelligence (PLAI) group at UBC Computer Science has been awarded a contribution by the Innovation for Defence Excellence and Security (IDEaS) program under their COVID-19 challenge, ‘Rapid response: Real-time insights for pandemic decision-making.’ PLAI envisions a system, where a public health policy analyst with particular public health policy goals in mind (e.g., maintaining the number of patients below a specific threshold, or maintaining ICU capacity) can use a simple web interface to describe these goals and current conditions. In turn, the system will respond with a set of conditions and controls (e.g., closures in particular locales or business-types for a defined period of time) that can enable the achievement of these goals within a particular set of the population.
Building off past success
The aim of current work, through the IDEaS funding, is to continue past research PLAI has already conducted on inference in epidemiological simulators, funded by a CIFAR AI and COVID-19 Catalyst Grant.
Specifically, in their past work, the group connected their PyProb Inference Engine to the agent-based epidemiological simulator ‘FRED’. As a result, they were able to infer the effectiveness of several policies such as school closures and hand-washing on influenza in the context of the pandemic. With the same goal in mind, the group also developed a simpler in-house model for COVID-19 spread. Details of this work is available online as a pre-print, and a code-base.
Building on this past work, the PLAI Group aims to continue work on showcasing the role probabilistic inference can play in situations requiring complex computer simulation.
Principal investigator, CS Professor Frank Wood explains, “Decision-making under uncertainty requires planning, planning can be framed as inference, inference requires a model, and the best models are simulators. Attacking the problems of figuring out what the best vaccine distribution policy is, whether it is better to keep schools open or closed, and how to do better when the next pandemic rolls around -- these are all addressable using software tools rooted in our research.”
The PLAI team will also be working on making the resulting tools more accessible for end-users such as policy analysts, by creating an easy-to-use web-based graphical user interface for the system. The interface will be developed with support from Tamara Munzner, UBC computer science professor who has expertise in information visualization.
She will also be leading a project exploring the accessibility of PyProb and other statistical inference tools for end-users. According to Tamara: “A visual analytics interface that allows epidemiologists or other analysts well-versed in traditional simulation to connect their existing knowledge and experience to the power and flexibility of a probabilistic inference approach could dramatically accelerate its adoption and impact. We can scaffold the transition to this new generation of tools through carefully designed visualization support that bridges between their current mental models and the new set of possibilities.”
Figure 1: Array of 2D histograms showing two-dimensional marginal distributions over controllable policy parameters that give rise to appropriately controlled outcomes in Allegheny county. Marginals for each policy are shown in the bottom row, with the number of samples from the uniform prior indicated by the dashed line. We can clearly see the efficacy of high rates of hand washing and a quick school closure policy, as indicated by the non-uniformity of the marginal distributions.
The end-goal is to make it faster and easier for policy analysts to assess different policies that can be enforced in the context of a pandemic, leading to better decision-making within our healthcare system. To this end, the team will be engaging with policy analysts, and public health experts through a series of online events to be announced shortly.
More about the PLAI Group