Uncertainty Quantification of Spatiotemporal Travel Demand with Probabilistic Graph Neural Networks (preprint)

Recent studies have significantly improved the prediction accuracy of travel demand using graph neural networks. However, these studies largely ignored uncertainty that inevitably exists in travel demand prediction. To fill this gap, this study proposes a framework of probabilistic graph neural networks (Prob-GNN) to quantify the spatiotemporal uncertainty of travel demand. This Prob-GNN framework is substantiated by deterministic and probabilistic assumptions, and empirically applied to the task of predicting the transit and ridesharing demand in Chicago. We found that the probabilistic assumptions (e.g. distribution tail, support) have a greater impact on uncertainty prediction than the deterministic ones (e.g. deep modules, depth). Among the family of Prob-GNNs, the GNNs with truncated Gaussian and Laplace distributions achieve the highest performance in transit and ridesharing data. Even under significant domain shifts, Prob-GNNs can predict the ridership uncertainty in a stable manner, when the models are trained on pre-COVID data and tested across multiple periods during and after the COVID-19 pandemic. Prob-GNNs also reveal the spatiotemporal pattern of uncertainty, which is concentrated on the afternoon peak hours and the areas with large travel volumes. Overall, our findings highlight the importance of incorporating randomness into deep learning for spatiotemporal ridership prediction. Future research should continue to investigate versatile probabilistic assumptions to capture behavioral randomness, and further develop methods to quantify uncertainty to build resilient cities.

Infectiousness of Places: Impact of Multiscale Human Activity Places in the Transmission of COVID-19 (preprint)

COVID-19 raises attention to epistemological risks related to everyday human activities. Our work quantifies infection transmission risks at different human activity places, including different types of settlements at macro-scale and establishments (restaurants, bars, etc.) at micro-scale, using evidences from COVID-19 in 906 urban areas across four continents. Relatively stable rules of how infection risks are distributed across human settlements and establishments are found. At micro-scale, the infection transmission risks at various establishments differ across countries, but generally, physical activity, entertainment and catering establishments lead to more infections than other activity places. At macro-scale, contrary to common beliefs, we find consistent pattern that transmission does not increase with settlement size and density. When considering interaction between the two scales, there is also consistent pattern that a smaller proportion of infections take place at specific establishments in larger settlements, suggesting that general public spaces such as streets play a greater role in transmission due to longer trips. Though with limitations, our work provides the first steps towards a system of knowledge on the linkage between places, human activities and disease transmission.

Impact of Modelling Approach on the Estimation of Government Intervention Effects in COVID-19 (preprint)

The efficacy of government interventions in epidemic has become a hot subject since the onset of COVID-19. There is however much variation in the results quantifying the effects of interventions, which is partly related to the varying modelling approaches employed by existing studies. This paper therefore aims to examine how the choice of modelling approach would affect the estimation results of intervention effects, by experimenting with different modelling approaches on a same data set composed of the 500 most affected U.S. counties. We compare the most frequently used methods from the two classes of modelling approaches, which are Bayesian hierarchical model from the class of computational approach and difference-in-difference from the class of natural experimental approach. We find that computational methods are likely to produce larger estimates of intervention effects due to simultaneous voluntary behavioral changes. In contrast, natural experimental methods are more likely to extract the true effect of interventions. Among different difference-in-difference estimators, the two-way fixed effect estimator seems to be an efficient one. Our work can inform the methodological choice of future research on this topic, as well as more robust re-interpretation of existing works, to facilitate both future epidemic response plans and the science of public health.