Postdoctoral fellow

Institution/Company:
McGill University
Location:
Montreal, QC, Canada
Job Type:
  • Postdoctoral
Degree Level Required:
PhD, Bachelor's
Apply Now

Postdoctoral fellow

Using Online News Media to Assess Community and Public Health Responses to COVID-19​ NPI: David Buckeridge (School of Population and Global Health, McGill University) Co-PI: Yue Li (School of Computer Science, McGill University)

Responsibility and Salary We have budgeted for two postdoctoral scholars (Base salary $60,000 plus $15,000 benefits) for both years of the proposed research.

  • Postdoc 1: The first postdoctoral scholar will have strong fundamentals in probabilistic models, variational Bayesian, and (ideally) topic models. Backgrounds in neural networks especially recurrent neural network are also desired. The postdoc will be working with Dr. Yue Li, who is the co-PI on this grant, on the model development and applications to the media news.
  • Postdoc 2: The second postdoctoral scholar will have a background in public health and data science and will focus on developing the official timeline of COVID-19 and will collaborate with the first postdoctoral scholar to fit these machine learning models. The second postdoctoral scholar will also attempt to explain differences between the evolution of the COVID-19 epidemic according to online media and the official timeline in terms of media biases and other factors.

Summary of Research Project

Introduction Global surveillance is critical for monitoring and guiding public health efforts to control the spread of emerging infectious diseases such as COVID-19 (caused by a coronavirus). Some types of surveillance, such as Digital disease surveillance (DDS) systems developed and operated by the Public Health Agency of Canada (PHAC) and the World Health Organization (WHO), hold particular promise for monitoring societal reactions, including public health control measures. DDS use methods from artificial intelligence to comb through information in online media reports from around the world to detect information about disease epidemics. Currently the focus of DDS is detection, because despite the promise of DDS for monitoring societal reactions and public health control measures, DDS systems do not currently automate the extraction of information about community and public health responses to epidemics.

The Global Population Health Intelligence Network (GPHIN) system developed by Health Canada in collaboration with WHO, which detected the first evidence of SARS (3), uses natural language processing (NLP) and human experts to sift through over twenty thousand online news reports each day in nine languages. Currently the focus of DDS is detection, because despite the promise of DDS for monitoring societal reactions and public health control measures, DDS systems do not currently automate the extraction of information about community and public health responses to epidemics.

To advance this research, our team was recently awarded funding by the CIHR for the Canadian 2019 Novel Coronavirus (COVID-19) Rapid Research Funding Opportunity for 2 years. This fund allow we to make equipment purchase such as GPU-enabled HPC servers and recruit highly qualified personnel to work with us. We are currently actively hiring talented postdocs in machine learning and epidemiology (see details below).

Research Team David Buckeridge (NPI) is a Professor in the School of Population and Global Health where he founded and directs the Surveillance Lab. He has been one of the technical advisors to WHO EIOS, leading scholarly efforts to evaluate early event detection through DDS and guiding the incorporation of new artificial intelligence methods into DDS such as GPHIN and EIOS.

Yue Li (co-PI) is an Assistant Professor in the School of Computer Science at McGill University and an Associate member of the Montreal Institute for Learning Algorithms (MILA). He has a strong background in methodological development for machine learning, computational genomics, and statistical genetics with 32 publications and 12 first-authored methodology papers in high-impact journals. Dr. Li can quickly recruit strong candidates through McGill and MILA.

In addition to our prior research with DDS, the Co-PI (Li) has developed methods in machine learning, and specifically in highly scalable, multimodal topic modeling (Li & Kellis, arxRiv 2018). These methods were used to develop a topic model called MixEHR, which was used to model heterogeneous types of electronic health record data for over 40,000 patients and 13 million clinical observations. Our team also has extensive experience in the research and practice of controlling respiratory infectious diseases (Co-I, Quach-Thanh) and in societal response to infectious disease (Co-I, King).

Objectives Our goal is to adapt and apply machine learning methods to DDS systems operated by PHAC and WHO to characterize how communities are reacting to COVID-19 and to document the implementation and effectiveness of public health responses implemented in Canada and around the world. We will realize our goal by accomplishing the following objectives:

  1. To characterize the global evolution of COVID-19 in terms of community reaction and public health response by applying machine learning methods to online news media from DDS used by PHAC and WHO; and,
  2. To compare the evolution of the COVID-19 epidemic in online news media to the evolution as measured through official sources, assessing how observed differences may be explained by biases in media and other factors.

Research Approach Rapid Mobilization and Data: Due to the urgency of the funding period (2 years) and the spread of COVID-19 We will initiate research immediately drawing on established relationships with PHAC and WHO and our extensive expertise in working with DDS data. From December 1, 2019 through February 17, 2020, the WHO DDS platform identified 1.54 million news articles of relevance to public health, with 488,000 articles about coronavirus. The annotations routinely available for each media report include date, geographical location, disease, source, and original language.

Objective 1: We will identify the evolution in themes related to community reaction and public health measures to control COVID-19 as reported in the media globally and by country. Given the amount of data and the challenge of pre-defining themes, we will use the unsupervised machine learning approach of dynamic topic modelling (DTM) to detect themes. The main challenge with an unsupervised approach such as DTM is interpreting the discovered latent topics, which may not map directly to concepts such as community reaction and public health response.

To address this challenge, we will extend the DTM to allow incorporate the spatial and geographical locations of the latent topics based on the regions and countries. Intuitively, topics learned from adjacent regions are more similar than topics learned from remote regions due to cultural, climate, ethnic factors. Dr. Li is an expert in latent topic models and can develop these extensions by working with a talented postdoc (see below).

Additionally, we will use user-defined constraints on the learned latent topics. For instance, we will consider the hierarchical structure of infectious organisms within the SNOMED vocabulary and the hierarchy of interventions within the ontology of public health interventions developed by the NPI (Buckeridge) as constraints.

Objective 2: As a first step we will gather official documents about the global evolution of COVID-19. We will use our established connections with PHAC and WHO to identify and access relevant documents, including resources such as the daily situation reports from WHO. We will use the same approach described in Objective 1 to model the evolution of themes within official documents using a DTM. We will take two approaches: to learn a DTM from official documents only and to learn a joint model across both online media reports and official documents. Finally, we will attempt to explain observed discrepancies between models learned from the two sources in terms of biases from media and other sources, using the framework that we are developing in another project.

Impact and Knowledge Translation The results will enrich our understanding of global variations in community and public health responses to COVID-19, generate evidence to guide the control of COVID-19, and enhance the capacity of DDS for future epidemics. Using an integrated Knowledge Translation model, end-users from public health agencies will be engaged throughout the project. The methods developed will be made available for incorporation into systems operated by WHO and PHAC. In addition to publication in open-source journals, results and data will be made available through an online portal for further analysis by researchers. Modules based on this research will also be included in the CIHR AI4PH Summer Institute.

1

Responsibilities:

We have budgeted for two postdoctoral scholars (Base salary $60,000 plus $15,000 benefits) for both years of the proposed research.

  • Postdoc 1: The first postdoctoral scholar will have strong fundamentals in probabilistic models, variational Bayesian, and (ideally) topic models. Backgrounds in neural networks especially recurrent neural network are also desired. The postdoc will be working with Dr. Yue Li, who is the co-PI on this grant, on the model development and applications to the media news.
  • Postdoc 2: The second postdoctoral scholar will have a background in public health and data science and will focus on developing the official timeline of COVID-19 and will collaborate with the first postdoctoral scholar to fit these machine learning models. The second postdoctoral scholar will also attempt to explain differences between the evolution of the COVID-19 epidemic according to online media and the official timeline in terms of media biases and other factors.

Qualifications:

  • strong background in probabilistic models and Bayesian inference or neural networks especially recurrent neural network
  • proficient programming in one of Python, R, C++
  • at least two first authored publications
  • fluent English speaking
  • How to Apply

    To apply for this position, please email your CV and/or resume to both yueli@cs.mcgill.ca and david.buckeridge@mcgill.ca