Adaptive COVID-19 Forecasting via Bayesian Optimization

Bannur, Nayana; Maheshwari, Harsh; Jain, Sansiddh; Shetty, Shreyas; Merugu, Srujana; Raval, Alpan

doi:10.1145/3430984.3431047

Cited by 3 publications

(4 citation statements)

References 2 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Regarding the optimization of COVID-19 prediction models, three main approaches have been reported in the literature. The first uses the SEIR (Susceptible - Exposed - Infectious - Recovered) model (or its derivatives) as its basis and applies machine learning and optimization methods to determine the epidemiological parameters of the model [6] – [13] , [95] – [106] . The second approach uses a population-based model to simulate the transmission of the virus [14] , [15] .…”

Section: The Four Framework and Literature Reviewmentioning

confidence: 99%

“…Due to the novelty of the virus, its epidemiological parameters are unknown, so the SEIR model is fitted to historical COVID-19 data, and the resulting estimated parameters are used to predict future cases. Bayesian optimization [6] , metaheuristics (e.g., particle swarm optimization, stochastic fractal search) [7] – [10] , [104] , [108] – [114] , neural networks [11] , [115] , [116] , and nonlinear curve-fitting based optimization methods [12] , [13] , [117] – [119] are some of the most popular approaches used to fit the model to the data and estimate the epidemiological parameters of the model, such as the reproduction number. In addition to forecasting COVID-19 cases, some studies considered additional aspects, such as the effect of different non-pharmaceutical intervention policies (social distancing and lockdown) and re-opening plans [101] , [114] , [120] – [127] .…”

Section: The Four Framework and Literature Reviewmentioning

confidence: 99%

“…These challenges required quick solutions and responses and presented many opportunities for optimization. Examples include optimizing prediction models to quantify future COVID-19 cases [6] – [24] , optimizing the allocation of critical hospital supplies to treat COVID-19 patients [25] – [32] , and developing decision support tools to guide data-driven mitigation strategies to combat the spread of the virus [33] – [37] . While numerous studies have tried to tackle these challenges, the COVID-19 pandemic continues to expose the flaws and shortcomings of current systemic policies, many of which were unanticipated and unprepared for.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Optimization in the Context of COVID-19 Prediction and Control: A Literature Review

et al. 2021

View full text Add to dashboard Cite

This paper presents an overview of some key results from a body of optimization studies that are specifically related to COVID-19, as reported in the literature during 2020-2021. As shown in this paper, optimization studies in the context of COVID-19 have been used for many aspects of the pandemic. From these studies, it is observed that since COVID-19 is a multifaceted problem, it cannot be studied from a single perspective or framework, and neither can the related optimization models. Four new and different frameworks are proposed that capture the essence of analyzing COVID-19 (or any pandemic for that matter) and the relevant optimization models. These are: (i) microscale vs. macroscale perspective; (ii) early stages vs. later stages perspective; (iii) aspects with direct vs. indirect relationship to COVID-19; and (iv) compartmentalized perspective. To limit the scope of the review, only optimization studies related to the prediction and control of COVID-19 are considered (public health focused), and which utilize formal optimization techniques or machine learning approaches. In this context and to the best of our knowledge, this survey paper is the first in the literature with a focus on the prediction and control related optimization studies. These studies include optimization of screening testing strategies, prediction, prevention and control, resource management, vaccination prioritization, and decision support tools. Upon reviewing the literature, this paper identifies current gaps and major challenges that hinder the closure of these gaps and provides some insights into future research directions.

show abstract

Section: The Four Framework and Literature Reviewmentioning

confidence: 99%

Section: The Four Framework and Literature Reviewmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Optimization in the Context of COVID-19 Prediction and Control: A Literature Review

et al. 2021

View full text Add to dashboard Cite

show abstract

“…From chocolate chip cookies [15] to COVID [16]: Bayesian Optimization has been applied in a wide range of subjects, including auto-tuning. To grasp the current state of BO in GPU auto-tuning, this section briefly explores related applications of BO in auto-tuning.…”

Section: Related Workmentioning

confidence: 99%

Bayesian Optimization for auto-tuning GPU kernels

Willemsen

Nieuwpoort

Werkhoven

2021

2021 International Workshop on Performance Modeling, Benchmarking and Simulation of High Performance Computer Systems (PMBS)

View full text Add to dashboard Cite

Finding optimal parameter configurations for tunable GPU kernels is a non-trivial exercise for large search spaces, even when automated. This poses an optimization task on a nonconvex search space, using an expensive to evaluate function with unknown derivative. These characteristics make a good candidate for Bayesian Optimization, which has not been applied to this problem before.However, the application of Bayesian Optimization to this problem is challenging. We demonstrate how to deal with the rough, discrete, constrained search spaces, containing invalid configurations. We introduce a novel contextual variance exploration factor, as well as new acquisition functions with improved scalability, combined with an informed acquisition function selection mechanism.By comparing the performance of our Bayesian Optimization implementation on various test cases to the existing search strategies in Kernel Tuner, as well as other Bayesian Optimization implementations, we demonstrate that our search strategies generalize well and consistently outperform other search strategies by a wide margin.

show abstract

Visible and Real Sizes of New COVID-19 Pandemic Waves in Ukraine

Nesteruk

2021

Innov Biosyst Bioeng

View full text Add to dashboard Cite

Background. To simulate the COVID-19 pandemic dynamics, various data sets and different mathematical models can be used. In particular, previous simulations for Ukraine were based on smoothing of the dependence of the number of cases on time, classical and the generalized SIR (susceptible-infected-removed) models. Different simulation and comparison methods were based on official accumulated number of laboratory confirmed cases and the data reported by Johns Hopkins University. Since both datasets are incomplete (a very large percentage of infected persons are asymptomatic), the accuracy of calculations and predictions is limited. In this paper we will try to assess the degree of data incompleteness and correct the relevant forecasts. Objective. We aimed to estimate the real sizes of two new epidemic waves in Ukraine and compare them with visible dynamics based on the official number of laboratory confirmed cases. We also aimed to estimate the epidemic durations and final numbers of cases. Methods. In this study we use the generalized SIR model for the epidemic dynamics and its known exact solution. The known statistical approach is adopted in order to identify both the degree of data incompleteness and parameters of SIR model. Results. We have improved the method of estimating the unknown parameters of the generalized SIR model and calculated the optimal values of the parameters. In particular, the visibility coefficients and the optimal values of the model parameters were estimated for two pandemic waves in Ukraine occurred in December 2020–March 2021. The real number of cases and the real number of patients spreading the infection versus time were calculated. Predictions of the real final sizes and durations of the pandemic in Ukraine are presented. If current trends continue, the end of the pandemic should be expected no earlier than in August 2022. Conclusions. New method of the unknown parameters identification for the generalized SIR model was proposed, which allows estimating the coefficients of data incompleteness as well. Its application for two pandemic waves in Ukraine has demonstrated that the real number of COVID-19 cases is approximately four times higher than those shown in official statistics. Probably, this situation is typical for other countries. The reassessments of the COVID-19 pandemic dynamics in other countries and clarification of world forecasts are necessary.

show abstract

Adaptive COVID-19 Forecasting via Bayesian Optimization

Cited by 3 publications

References 2 publications

Optimization in the Context of COVID-19 Prediction and Control: A Literature Review

Optimization in the Context of COVID-19 Prediction and Control: A Literature Review

Bayesian Optimization for auto-tuning GPU kernels

Visible and Real Sizes of New COVID-19 Pandemic Waves in Ukraine

Contact Info

Product

Resources

About