[Paper Review] Neural Network aided quarantine control model estimation of global Covid-19 spread

Raj Dandekar, George Barbastathis|arXiv (Cornell University)|Apr 2, 2020

COVID-19 epidemiological studies22 references46 citations

TL;DR

The paper augments a SIR/SEIR epidemiological model with a neural network to learn quarantine strength and estimates its impact on Rt across Wuhan, Italy, South Korea, and the US, highlighting how rapid interventions curb spread and predicting US outcomes under different policies.

ABSTRACT

Since the first recording of what we now call Covid-19 infection in Wuhan, Hubei province, China on Dec 31, 2019, the disease has spread worldwide and met with a wide variety of social distancing and quarantine policies. The effectiveness of these responses is notoriously difficult to quantify as individuals travel, violate policies deliberately or inadvertently, and infect others without themselves being detected. In this paper, we attempt to interpret and extrapolate from publicly available data using a mixed first-principles epidemiological equations and data-driven neural network model. Leveraging our neural network augmented model, we focus our analysis on four locales: Wuhan, Italy, South Korea and the United States of America, and compare the role played by the quarantine and isolation measures in each of these countries in controlling the effective reproduction number $R_{t}$ of the virus. Our results unequivocally indicate that the countries in which rapid government interventions and strict public health measures for quarantine and isolation were implemented were successful in halting the spread of infection and prevent it from exploding exponentially. We test the predictive ability of our model by matching predictions in the duration 3 March - 1 April 2020 for Wuhan and in the duration 25 March - 1 April 2020 for Italy and South Korea. In the case of the US, our model captures well the current infected curve growth and predicts a halting of infection spread by 20 April 2020. We further demonstrate that relaxing or reversing quarantine measures right now will lead to an exponential explosion in the infected case count, thus nullifying the role played by all measures implemented in the US since mid March 2020.

Motivation & Objective

• Quantify how quarantine and isolation policies affected the effective reproduction number Rt in diverse regions.
• Develop a neural network-augmented epidemiological model that learns quarantine strength Q(t) from public data.
• Assess predictive capability by training on early outbreak data and validating against subsequent observations.
• Provide policy-relevant forecasts for the USA under different quarantine scenarios based on learned Q(t).
• Compare results with traditional SEIR/SIR models to show the added value of incorporating quarantine dynamics.

Proposed method

• Augment a classic SIR/SEIR framework with a neural network that estimates a time-varying quarantine strength Q(t).
• Represent Q(t) as a 2-layer dense neural network with 10 hidden units using ReLU activation.
• Define T(t)=Q(t)I(t) as quarantined individuals and use R_t=β/(γ+Q(t)).
• Train the neural network-augmented model by minimizing the loss between log(I) and log(I_data) and between log(R) and log(R_data) using ADAM optimization.
• Perform region-specific training for Wuhan, Italy, South Korea, and the USA using publicly available infection and recovery counts.
• Compare results to standard SIR/SEIR models to illustrate how quarantine dynamics explain observed plateaus and declines in Rt.

Experimental results

Research questions

• RQ1 How does a learned quarantine strength function Q(t) influence the effective reproduction number Rt in different regions?
• RQ2 Can a neural network-augmented SIR model reproduce observed stagnation/plateau in infection counts that classical models miss?
• RQ3 What are the quantitative effects of quarantine policies on Rt across Wuhan, Italy, South Korea, and the USA?
• RQ4 How do forecasts under current policies compare with forecasts when adopting quarantine strengths learned from other regions?
• RQ5 What are region-specific ranges of Q(t) that align with observed data and intervention timelines?

Key findings

• Quarantine strength learning captures observed stagnation and Rt<1 in Wuhan and South Korea within roughly one month of strong interventions.
• In Italy, a sharp rise in Q(t) coincides with a decrease in Rt following government restrictions.
• In the US, Q(t) increases after mid-March 2020, with Rt showing a decreasing trend and a predicted halt in infections if policies continue as modeled.
• Comparative analyses show stronger quarantine corresponds to higher Q(t) and lower Rt across regions.
• Forecasts suggest relaxing quarantine in the US could lead to exponential growth and possibly up to ~1 million infections under certain scenarios.
• Table 1 reports region-specific β, γ, Q(t) ranges, intervention efficiency, and shows Wuhan (β=1, γ=0.023, Q(t) 0.8-1.1, 30 days), Italy (β=0.74, γ=0.032, Q(t) 0.4-0.7, 27 days), South Korea (β=0.68, γ=0.004, Q(t) 0.4-0.8, 20 days), US (β=0.69, γ=0.008, Q(t) 0.4-0.6, 37 days).