Low Power Based Cognitive Domain Ontology Solving Approaches

Date of Award


Degree Name

Ph.D. in Electrical and Computer Engineering


Department of Electrical and Computer Engineering


Tarek Taha


The demand for autonomous systems is increasing in multiple domains, including mobile systems (UAVs, cars, and robots) and planning systems, as it improves the performance of the systems beyond human capabilities. In autonomous systems, agents mine a massively large knowledge database to make intelligent and optimal decisions in run-time. Knowledge mining and decision-making are cast as constraint satisfaction problems (CSP), where solutions are generated by satisfying a number of constraints from the domain. CSPs have become a point of interest because of their affiliation with both artificial intelligence and operations research. From resource allocation and automated decision-making to gaming, constraint satisfaction problems are widely noticeable. An autonomous system achieves its autonomy by solving these problems using CSP solving approaches, including Boolean satisfiability, satisfiability modulo theories, answer set programming. Autonomy is the degree of acquired autonomous capability. Within the Air Force, autonomy is defined as the ability to select the required course of action (COA) to achieve higher objectives. The Cognitively Enhanced Complex Event Processing (CECEP) framework being developed at the US Air Force is an autonomous decision support tool that enables enhanced agent-based decision making. CECEP enables the autonomous system to process complex real-world events and select the required course of action to achieve optimal results. CECEP is capable of representing and processing declarative, procedural, and domain-specific knowledge to deal with all forms of real-world events. CECEP also incorporates several task independent knowledge processing frameworks to perform as a generic problem-solving framework. CECEP's problem-solving capability makes it a universal complex event processing framework that can be utilized in both military and civilian domains. CECEP captures its domain knowledge in a cognitive domain ontology (CDO), storing it as a feature diagram. In a feature diagram, knowledge is stored in a hierarchical set of connected elements through constraints. The CECEP framework consumes most of its time and power to process this domain knowledge. The CDO mining process employed in the CECEP framework is cast as a constraint-satisfaction problem (CSP). However, CSPs require enormous amounts of power and time to solve because of their large combinatorial search space and complexity. This thesis examined several accelerated CDO solving approaches for low SWAP platforms to process complex events in real-time. Both spiking-based non-von Neumann and traditional approaches were considered in this thesis. This study developed two novel non-von Neumann approaches to solve medium-sized CDOs using spiking neural networks. The spiking neural methods used convolution neural networks and pattern matching-based approaches to solve CDOs. In the non-spiking field, this study developed a novel Q-learning-based efficient CDO mapping and mining approach to deal with multi-instance-based large CDOs. The 'instance' property is one of the crucial characteristics of the CDOs where the solution search space increases exponentially with the increase in the number of instances. An instance represents the replication of existing CDO branches. This study also developed a portfolio of incomplete SAT-based solvers to solve CDOs as CSP problems. CECEP generates an optimized decision by assessing all the possible solutions with specific criteria. Here this task is classified as the task allocation problem. This thesis also developed a portfolio of the optimized solvers to perform the optimized task allocation.


Electrical Engineering, Cognitive Domain Ontology (CDO), Low power, Optimized decision making, Knowledge mining, Spiking based CDO solving

Rights Statement

Copyright © 2021, author.