Date of Award

January 2020

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Computer Science

First Advisor

Emanuel Grant

Abstract

Program synthesis systems can be highly advantageous in that users can automatically generate code to fit a wide variety of applications from high-level specifications without needing any low-level programming skills or knowledge of which type of data structures and algorithms should be used. NASA has developed and uses two of these systems, AUTOFILTER and AUTOBAYES. Though much is gained in terms of time and cost efficiency in the use of these systems, they suffer from an issue that is inherent in all code generator systems, the verifiability of the correctness of the generated code against the input specifications. Many times, this verification process can take just as long, if not longer than manually developing and testing the code would have been. Because of this, much work has been done by NASA and others to develop methods for automatic certification that can be produced along with the program and are easy to use. However, there is still more work to be done in this area, especially in the area of automatic visual verification (e.g., by using UML diagrams to provide visual aid in the verification of the generated code). Work has been done by Grant et al. in collaboration with NASA to develop a rigorous approach to system correctness verification that uses domain-specific graphical meta-models of the expected input/output systems with identified constraints on the input/output and their relationships. Though this approach has been applied to AUTOFILTER, it has yet to be applied to other domains. In this work, Grant’s approach is extended to the data analysis domain by being applied to AUTOBAYES. A model of the input specification for AUTOBAYES was obtained for the case in which a normal distribution of data is assumed. This model, derived from the AUTOBAYES input files, the n-dimensional Gaussian equation, and allowed priors, is a UML class diagram (CD). Similarly, a UML CD model of the AUTOBAYES program output was derived. These CD's were then used to develop 30 constraints on the input, the output, and the relationship between them. These constraints were then transformed into the OCL formal specification language and analyzed with the USE tool, along with the derived comprehensive CD (i.e., a combination of the input CD, output CD, and the relationships between each other). These models and constraints were used to successfully check that all of the developed constraints were satisfied with the model representing AUTOBAYES. Unfortunately, a configuration for a full validation with USE was not obtained, after several iterations, due to project time restrictions. However, the results obtained adequately demonstrate that this method can be extended to the domain of AUTOBAYES. This work was motivated both due to its relevance to NASA in the chosen case study of AUTOBAYES as well to show that Grant’s approach can be extended to other domains beyond AUTOFILTER.

Share

COinS