Restricted Research - Award List, Note/Discussion Page
Fiscal Year: 2023
1994 The University of Texas at El Paso (143882)
Principal Investigator: Wagler,Amy E
Total Amount of Contract, Award, or Gift (Annual before 2011): $ 20,000
Exceeds $250,000 (Is it flagged?): No
Start and End Dates: 3/17/23 - 9/30/24
Restricted Research: YES
Academic Discipline: Mathematical Sciences
Department, Center, School, or Institute: Mathematical Sciences
Title of Contract, Award, or Gift: (Differentially Private) Synthetic Data Generation, SDG
Name of Granting or Contracting Agency/Entity:
PACIFIC NORTHWEST NATIONAL LABORATORY
CFDA Link: NSF
47.049
Program Title:
Mathematical and Physical Sciences
CFDA Linked: Mathematical and Physical Sciences
Note:
As we operationalize model training on personal data, there is a strong need to guarantee the privacy for the data upon which the models are trained. Modern machine learners excel in extracting features from data, and in some instances, these features might inadvertently reveal sensitive information about the data. To this end, we propose to synthesize data that guarantees differential privacy for individuals but retains population statistics upon which we can make inferences as well as predictions. To this end, we propose to use and expand the use of generative adversarial models to synthesize artificial electronic health record data. The use f generative adversarial models has one primary benefit where we would not need a priori knowledge of the underlying distribution of the data for synthesis; the adversarial model would learn these distributions via feature extraction. Building upon the generative models, we will develop statistical metrics to ascertain the efficacy of the model as well as the robustness of the data. Our final deliverables will consist of software to synthesize data at scale on HPC, methods to evaluate the models and the data, and a computational framework that would generalize beyond the biomedical domain.
Discussion: No discussion notes