Wednesday, October 25, 2023 - 4:00pm
Event Calendar Category
LIDS Seminar Series
Speaker Name
Seth Flaxman
Affiliation
Oxford
Building and Room Number
32-155
Abstract
Machine learning is the computational beating heart of the modern AI renaissance. Behind the hype, a range of machine learning and computational statistical methods are quietly revolutionizing our approach to difficult statistical and scientific inference problems. I will present my perspective on the emerging field of “inferential Artificial Intelligence” (iAI) through a series of case studies on important global health challenges. I conceive of iAI as a big tent, encompassing modern probabilistic programming, replicable data scientific workflows, methods for assessing Big Data quality, uncertainty quantification, active learning, and a range of computational and deep learning approaches to transform applied statistical analyses.
I will discuss iAI in the context of my work during the COVID-19 pandemic as part of the Imperial College COVID-19 Response Team and the collaborations I am now leading through the Machine Learning & Global Health Network (www.MLGH.net).
Case studies will include:
- An open source semi-mechanistic hierarchical Bayesian model of SARS-CoV-2 transmission to estimate R(t), the effectiveness of non-pharmaceutical interventions, and to characterize changed epidemiological properties of Variants of Concern (Flaxman et al, Nature 2020; Volz et al, Nature 2021; Faria et al; Science 2021, Mlcochova et al, Nature 2021; Dhar et al, Science 2021)
- The Big Data Paradox, or: how large surveys of COVID-19 vaccine uptake missed the mark so spectacularly (Bradley et al, Nature 2021)
- Global estimates of COVID-19-associated orphanhood and deaths of caregivers (Hillis et al, Lancet 2021; Unwin et al, Lancet Child & Adolescent Health 2022; Hillis et al, JAMA Pediatrics 2022) and ongoing work strengthening data collection to identify orphans through death certificates in Zambia, Brazil, Colombia, and Utah (Flaxman et al, Science 2023)
- πVAE/PriorVAE: Scalable MCMC inference for computationally challenging prior choices with Bayesian deep generative modeling (Semenova et al, Royal Society Interface 2022; Mishra et al, Statistics & Computing 2022) and other neural network approaches (Giovanni et al, AAAI 2023)
- Adaptive Learning Survey Design (A-LSD): a new survey methodology based on active learning that we are piloting to improve representativeness of vulnerable subpopulations and provide spatially fine-grained maps of food insecurity with the World Food Program
Biography
Seth Flaxman is an associate professor in the Department of Computer Science at Oxford. Originally from the Chicago area, he received his PhD in 2015 from Carnegie Mellon University in machine learning and public policy (School of Computer Science and Heinz College of Information Systems and Public Policy) and has worked for the World Health Organization. Seth’s research is on spatiotemporal statistics and Bayesian machine learning, applied to public policy, global health and social science. He was part of the Imperial College COVID-19 Response Team, leading a number of publications on non-pharmaceutical interventions, computational epidemiology, and COVID-19 orphanhood. He has published on filter bubbles / echo chambers in media, the Big Data paradox, and the regulation of machine learning algorithms. He is the statistical lead for the Global Reference Group on Children Affected by COVID-19 and Crisis. Seth won the Samsung AI Researcher of the Year Award (2020) and the SPI-M-O Award for Modelling and Data Support (2022) for modeling advice provided to the UK government during the COVID-19 pandemic. In 2022, he co-founded the Machine Learning & Global Health network (www.MLGH.net) of researchers spanning three continents with a kickoff workshop held in Kigali, Rwanda at ICLR in 2023.