Browsing by Author "Wang, Bao"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    Adversarial Attacks on Deep Temporal Point Process
    (IEEE, 2022) Khorshidi, Samira; Wang, Bao; Mohler, George; Computer Science, Luddy School of Informatics, Computing, and Engineering
    Temporal point processes have many applications, from crime forecasting to modeling earthquake aftershocks sequences. Due to the flexibility and expressiveness of deep learning, neural network-based approaches have recently shown promise for modeling point process intensities. However, there is a lack of research on the robustness of such models in regards to adversarial attacks and natural shocks to systems. Precisely, while neural point processes may outperform simpler parametric models on in-sample tests, how these models perform when encountering adversarial examples or sharp non-stationary trends remains unknown. Current work proposes several white-box and blackbox adversarial attacks against temporal point processes modeled by deep neural networks. Extensive experiments confirm that predictive performance and parametric modeling of neural point processes are vulnerable to adversarial attacks. Additionally, we evaluate the vulnerability and performance of these models in the presence of non-stationary abrupt changes, using the crimes dataset, during the Covid-19 pandemic, as an example.
  • Thumbnail Image
    Item
    SOS-EW: System for Overdose Spike Early Warning Using Drug Mover’s Distance-Based Hawkes Processes
    (Springer, 2020) Chiang, Wen-Hao; Yuan, Baichuan; Li, Hao; Wang, Bao; Bertozzi, Andrea; Carter, Jeremy; Ray, Brad; Mohler, George; Computer and Information Science, School of Science
    Opioid addictions and overdoses have increased across the U.S. and internationally over the past decade. In urban environments, overdoses cluster in space and time, with 50% of overdoses occurring in less than 5% of the city and dozens of calls for emergency medical services being made within a 48-hour period. In this work, we introduce a system for early detection of opioid overdose clusters based upon the toxicology report of an initial event. We first use drug SMILES, one hot encoded molecular substructures, to generate a bag of drug vectors corresponding to each overdose (overdoses are often characterized by multiple drugs taken at the same time). We then use spectral clustering to generate overdose categories and estimate multivariate Hawkes processes for the space-time intensity of overdoses following an initial event. As the productivity parameter of the process depends on the overdose category, this allows us to estimate the magnitude of an overdose spike based on the substances present (e.g. fentanyl leads to more subsequent overdoses compared to Oxycontin). We validate the model using opioid overdose deaths in Indianapolis and show that the model outperforms several recently introduced Hawkes-Topic models based on Dirichlet processes. Our system could be used in combination with drug test strips to alert drug using populations of risky batches on the market or to more efficiently allocate naloxone to users and health/social workers.