Covert Cognizance: A New Predictive Modeling Paradigm
Abstract
Predictive modeling is a mainstay technology in modern autonomous environments, as it represents the brain or the nervous system of any digitally-controlled system, e.g., energy systems, materials reprocessing plants, electric grid, etc. Constructed based on system observations via machine learning and/or physics understanding of system behavior, predictive models have to be continuously updated/improved to keep pace with the ever-increasing functional requirements of modern engineering applications which place high premium on computational efficiency, reliability, and trustworthiness. These high level goals necessitate many important requirements, such as -- to name a few Ð a) minimizing the computational resources in terms of memory usage, storage requirements, and CPU time; b) developing measures of confidence in the simulation results by carefully propagating and identifying dominant sources of uncertainties and by early-detection of software crashes; c) developing measures of trustworthiness of the source code and simulation results against unauthorized access; and d) developing surveillance capabilities to monitor code usage and execution history and its adherence to prescribed specification, etc. To meet these requirements, the extant approaches rely on expert-based techniques that are gleaned from years of familiarity with the system models, their domains of applications, and the available computer architecture. The overarching objective of this project is to develop and implement a novel predictive modeling paradigm that is designed to develop self-cognizance for predictive models. In our context, ÒcognizanceÓ implies the softwareÕs ability to develop self-awareness of its own execution characteristics -- including usage history, memory/storage and CPU utilization, and information footprint throughout the code subroutines -- in order to meet the previously noted requirements, e.g., efficiency, confidence, and trustworthiness, etc. The idea is to augment the model with a new physics operator Ð customized to the cognizance goal(s) -- with the sole function of collecting and processing enabling information about the modelÕs execution characteristics in order to achieve the given goal(s). The proposed cognizance technology will allow a given software to be self-aware of its own execution history, inclusive of code changes, and any unauthorized attempts to manipulate its internal functions or input/output variables. Furthermore, the proposed technology can be implemented in a covert manner -- avoiding the use of additional lines of code and conventional logging files, and instead incorporating the augmented physics as an inseparable part of the numerical solver employed to solve the original physics model. Developing covert self-cognizance is expected to be of critical value to a wide range of military applications in both offensive and defensive settings. This short-term project represents a proof-of-concept for the proposed cognizance algorithms. Specifically, it will demonstrate the value of cognizance to tracking the information footprint throughout a model in order to provide guidance on improving efficiency. The proposed cognizance is inspired by recent R&D work at Purdue University on the development of trustworthiness measures for industrial control network in response to the looming threat of state-sponsored insider-assisted attacks aiming to inflict physical damage on industrial systems. The idea is to embed physics-based covert signals into the network to detect manipulation attempts. The concept is further developed in this project to develop self-awareness for a wider range of applications. The PI has received support from Purdue Research Foundation and Purdue Foundry to form a start-up company, Covert Defenses, LLC, for further development and commercialization of this technology.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Oct 01, 2019
- Source ID
- W911NF1910489
Entities
People
- Hany Abdel-khalik
Organizations
- Army Contracting Command
- National Security Agency