Who Are the `Silent Spreaders'?: Contact Tracing in Spatio-Temporal Memory Models

The novel coronavirus SARS-CoV-2 has been posing severe threat to public health around the world by causing the infectious coronavirus disease 2019 (COVID-19). A key challenge of controlling COVID-19 is the large proportion of asymptomatic COVID-19 cases (ACCs) that makes breaking the transmission chains even harder. Unfortunately, till now there is still a lack of cost-effective approaches to detect ACCs except doing a medical screening of the entire population. This paper presents a neural network model called Spatio-Temporal Episodic Memory for COVID-19 tracing (STEM-COVID) for identifying ACCs from contact tracing data. Based on fusion Adaptive Resonance Theory (fusion ART), STEM-COVID provides a mechanism to encode the collective spatio-temporal episodic memory traces of individuals, based on which parallel searches of ACCs can be performed in a computationally efficient manner by pooling together the episodic traces of the identified positive cases. To illustrate the effectiveness of STEM-COVID, a simulation model of the COVID-19 spreading is implemented based on recent epidemiological findings on ACCs. The experimental results based on multiple simulation scenarios show that the STEM-COVID model is able to identify ACCs with a reasonably high level of accuracy and efficiency.