User Tools

Site Tools


Action unknown: copypageplugin__copy
projects:year7:7a.014.uva

7a.014.UVA - Data-Driven Dynamic Models for Exploring Systemic Risk with Unknown Inputs

Project - Summary

This research project is working to develop a general methodology that will take data streams (time series of data) and/or simulation models (agent-based or other) and create state-based, dynamic macro models, and perform analysis of the created models to explore critical issues. These include: i) what ranges of parameter values would cause the model to become unstable, ii) what is the likelihood of certain critical states occurring and at what recurrence, iii) can we create controls to manage the dynamic system, etc. Such a formal dynamic model would allow for organizations, such as regulators, to explore the impacts of policy changes and modifications to existing polices. This research would also have broader applications in other data-driven systems, such as the internet, telecommunications, healthcare, and the intelligence community.

Project - Team

Team Member Role Email Phone Number Academic Site/IAB
William Scherer PI wts@virginia.edu (434) 982-2069 University of Virginia
Hunter Moore Student/Graduate Researcher Hm2t@virginia.edu (276) 732-5009 University of Virginia
Anil Deane Project Mentor anil.deane@ngc.com Not Available Northrop Gumman

Project - Novelty of Approach

This approach combines previous work on systemic risk, data with unknown inputs, modeling of agents (algorithmic and human), and combines them in a way to forecast regulatory changes to a system in an effort to create data-driven regulation

Project - Deliverables

Deliverables
1 Literature review (review of literature in progress)
2 Preliminary methodology (review/research of methods in progress)
3 Exploration of sample problem (acquisition of multiple datasets, initial analysis of one dataset)
4 Report including explanation of method, model and testing results (development of report content underway)

Project - Benefits to IAB

This project has the potential to decrease modeling time by self-detecting its states and discovering states in a system that may have been overlooked by a modeler alone. In addition, given an online implementation of this system, it will have the possibility to discover the creation of new states that may not have been present at the time of training and can warn operators of the potential existence of these states.

Project - Presentation Video

Project - Documents

projects/year7/7a.014.uva.txt · Last modified: 2019/08/20 10:42 by sally.johnson