Optimizing deep neural networks to predict the effect of social distancing on COVID-19 spread.

Deep Neural Networks (DNN) form a powerful deep learning model that can process unprecedented volumes of data. The hyperparameters of DNN have a significant influence on its prediction performance. Evolutionary algorithms (EAs) form a heuristic-based approach that provides an opportunity to optimize deep learning models to obtain good performance. Therefore, we propose an evolutionary deep learning model called IPSO-DNN based on DNN for prediction and an improved Particle Swarm Optimization (IPSO) algorithm to optimize the kernel hyperparameters of DNN in a self-adaptive evolutionary way. In the IPSO algorithm, a micro population size setting is introduced to improve the search efficiency of the algorithm, and the generalized opposition-based learning strategy is used to guide the population evolution. In addition, the IPSO algorithm employs a self-adaptive update strategy to prevent premature convergence and then improves the exploitation and exploration parameter optimization performance of DNN. In this paper, we show that the IPSO algorithm provides an efficient approach for tuning the hyperparameters of DNN with saving valuable computational resources. We explore the proposed IPSO-DNN model to predict the effect of social distancing on the spread of COVID-19 based on the social distancing metrics. The preliminary experimental results reveal that the proposed IPSO-DNN model has the least computation cost and yields better prediction accuracy results when compared to the other models. The experiments of the IPSO-DNN model also illustrate that aggressive and extensive social distancing interventions are crucial to help flatten the COVID-19 epidemic curve in the United States.