“If we knew what it was we were doing, it would not be called research, would it?” -Albert Einstein.
Data reduction in the era of big data
Funding: Center for Hardware and Embedded System Security and Trust (CHEST)
Cyber-physical systems (CPS) and the Internet of Things (IoT) promise to enable transformational improvements in complex systems through ubiquitous sensing. In this paradigm, a large number of spatially distributed sensor nodes cooperatively monitor a physical system and transmit a potentially vast amount of sensed data to a central base station for analysis. In 2010, Lt. Gen. David A. Deptula, former Air Force deputy chief of staff for intelligence, predicted that “We’re going to ﬁnd ourselves in the not too distant future swimming in sensors and drowning in data.” His words have been conﬁrmed by several reports, recognizing the ﬂood of data coming from the intelligence, surveillance, and reconnaissance systems used by intelligence analysts and commanders as a major concern. To address these concerns, we aim to develop computational algorithms that substantially reduce the volume of data while maintaining its value by minimizing the redundancy within the data. Particularly, these algorithms aim to achieve: a) optimal sensor placement by exploiting mobile sensors and data fusion, b) efﬁcient prediction-based event-triggered data collection, and c) optimal spatial data collection from moving sensors. Together, these algorithms will achieve (near)-optimal data collection from the network of IoT sensors while ensuring accurate system state estimation within acceptable error bounds.
N. Alemazkoor and H. Meidani, “Efficient Collection of Connected Vehicles Data With Precision Guarantees,” in IEEE Transactions on Intelligent Transportation Systems, vol. 21, no. 11, pp. 4637-4645, Nov. 2020, doi: 10.1109/TITS.2019.2942568.
N. Alemazkoor and H. Meidani, “A Data-Driven Multi-Fidelity Approach for Traffic State Estimation Using Data From Multiple Sources,” in IEEE Access, vol. 9, pp. 78128-78137, 2021, doi: 10.1109/ACCESS.2021.3081063.
Expedited analysis for infrastructure systems
Funding: Environmental Resilience Institute at UVA
Infrastructure systems are inherent to various sources of uncertainty. Reliable analysis of these systems is significantly challenged by those uncertainties. This is while optimal operation, control, and planning of infrastructure systems hinge on timely and reliable analysis of the systems. To reduce the computational burden of probabilistic analysis, high-fidelity simulations of the systems are often replaced with surrogate models that approximate the relationship between uncertain inputs and the quantity of interest for the probabilistic analysis. The accuracy of these analytical approximations directly affects the accuracy of system analysis and the optimality of the decisions made based on the analysis. The challenge is that the number of training samples required for deriving accurate approximations grows exponentially with the dimension of uncertain inputs. To address this, we investigate advanced surrogate modeling techniques such as basis adaptation, dimension-reduction, optimal sampling as well as multi-fidelity and physic-based surrogate modeling for surrogate-based probabilistic analysis and surrogate-based optimization of large-scale infrastructure systems under uncertainty.
Alemazkoor, N., Rachunok, B., Chavas, D.R. et al. Hurricane-induced power outage risk under climate change is primarily driven by the uncertainty in projections of future hurricane frequency. Sci Rep 10, 15270 (2020). https://doi.org/10.1038/s41598-020-72207-z
N. Alemazkoor and H. Meidani, “Fast Probabilistic Voltage Control for Distribution Networks With Distributed Generation Using Polynomial Surrogates,” in IEEE Access, vol. 8, pp. 73536-73546, 2020, doi: 10.1109/ACCESS.2020.2987787.
Funding: Engineering in Medicine program at UVA
We use advances in machine learning to improve the treatment and quality of life of end-stage cancer patients. Many terminally ill patients, including those with stage-IV gastric and esophageal cancers, wish to have more time. At the same time, they wish to be symptom-free to be able to enjoy their last few months with their family and friends. Considering the trade-off between side effects from palliative chemotherapy (PC) and its potential survival benefits, many patients choose PC. However, the response to PC varies significantly among the patients, meaning that many patients do not necessarily benefit from PC while they experience its substantial side effects. Early prediction of patients’ response to chemotherapy is, therefore, crucial to avoid unnecessary toxicities from unhelpful PC during the short remaining lifetime of the patients. Here, we conduct the first-ever study on survival prediction of patients with gastric and esophageal cancers under PC treatment. We find that machine learning can predict with high accuracy whether a patient will benefit from PC only after two cycles of chemotherapy. The results from this study not only promise substantial improvement in patients’ quality of life during their last months but also can increase their survival time by timely switching to potentially more beneficial second-line treatments.
Big mobility data for understanding community behavior
Enforcement of hurricane preparedness plans, such as issuing evacuation orders, is an imperative step towards reducing social vulnerability, in terms of both human suffering and economic loss. Yet are these evacuation orders actually effective? To answer this question, our study focuses on utilizing big data to (1) analyze evacuation decisions as a function of government-issued evacuation orders and (2) examine how the evacuation behavior of communities varies based on their socio-economic and demographic factors. In particular, we analyzed the evacuation behavior of 660 census block groups (CBGs) in the state of Florida during Hurricane Dorian by using passively collected high-fidelity mobility data in conjunction with their associated socio-economic and demographic data. The results from analysis show that CBGs with evacuation, compared to those without orders, have statistically significant increases in evacuations once order are issued. However, there does not seem to be significant overall growth in evacuations for CBGs under mandatory evacuation orders issued. The landfall hours, median age, and racial minority percentage of the CBGs are highly relevant to determine the effectiveness of an order. The results are helpful in understanding both the statistical and practical significance of community trust and compliance with evacuation orders. Our analysis could be used as a strategic decision-support aid by policymakers to define effective hurricane evacuation plans.