Unfolded primal-dual algorithm estimating time-varying COVID-19 reproduction numbers
Year: 2026
Type: conference-paper
Abstract: During an epidemic outbreak, accurate real-time monitoring of the intensity of the pathogen propagation is key to implement efficient countermeasures rapidly enough to avoid dramatic epidemic waves. However, this represents an immense challenge for two main reasons. First, for an emerging disease such as COVID-19, very few prior knowledge is available. Second, in a crisis context, the quality of collected data might be very low. One popular indicator to quantify the pathogen transmissibility, is the time-varying reproduction number, defined as the average number of infections generated by a typical contagious individual if all conditions remained the same. Recently, a variational estimator of the reproduction number robust to low data quality has been proposed, relying on a customized primal-dual minimization algorithm. This work proposes to unfold this sophisticated optimization scheme into a residual convolutional neural network. The obtained network, designed to estimate time-varying reproduction numbers from infection counts, is trained on genuinely crafted synthetic data. It is then compared to state-of-the art estimators and shown to produce very consistent estimates of the reproduction number when tested on real COVID-19 counts from several countries at different phases of the pandemic. A documented Python toolbox is publicly available to reproduce all presented results.
Source: HAL
Language: English
Subfield: Modeling and Simulation
Field: Mathematics
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7130489484