Predictin Biothreat Impacts from Early-Stage Data via Transfer Learning
Abstract
The project’s objectives are to determine the efficacy of machine learning (ML) techniques in predicting epidemic impacts in early-stage, data-sparse conditions, particularly novel biothreats, both intentional (e.g. novel genetically-modified pathogens) and natural (e.g. zoonotic). We define “early stage” as the initial period in an outbreak when only short time-series data are available, prior to rapid growth and geographic spread, and epidemiological parameters are not well-established. Specifically, we will answer three questions that will enable rapid decision-making based on ML models of epidemic impacts: Q1: How much does partial-pooling improve predictions of epidemic impacts, and what information transfers best across diseases? Q2: What variables are most predictive of epidemic impacts at early stages of an outbreak? Q3: At what point in epidemic growth do hybrid models that incorporate mechanistic epidemiological components become more accurate than pure ML models?
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jun 14, 2022
- Source ID
- HDTRA12110023
Entities
People
- Noam Ross
Organizations
- Defense Threat Reduction Agency
- EcoHealth Alliance