DETECTION OF VULNERABILITIES IN SOFTWARE MODELS OF HARDWARE

Abstract

The work deals with the problem of detecting extraneous signs in software models of hardware. To solve it, a strategy based on the use of a genetic algorithm is proposed. After the analysis, it was found that the traditional genetic algorithm model is not suitable due to its low efficiency. This is due to complexity issues that inherently arise both when converting a hash to a function and when mapping a query to a target parameter. In particular, it was problematic to create Hamiltonian completion problems in the graph. This problem is based on solutions to determine the minimum number of edges in a graph that must be added to ensure the existence of a Hamiltonian cycle. Thus, the modifications made to the traditional genetic algorithm model were proposed in addition to the new functions, operators, etc. that were used. Modifications made to the traditional model of the genetic algorithm made it possible to develop a strategy and approach to the detection of extraneous signs included in software models of hardware.

To conduct experiments to establish the effectiveness of the developed approach, an application was developed that implements a genetic algorithm with modifications. Two files are submitted to the input of this application. The first file contains a software model of a hardware device, and the second file contains a reference model of the same hardware device. The database of third-party mark models contains typical models of entering third-party marks. In general, the developed application acts as a classifier. As a result of its use in experimental studies on artificial sets of input models, it demonstrated a classification result evaluated by the F1 metric equal to 82%. Such a value is permissible. To improve it, you need to fill the database of models of third-party signs.

Directions for further research are improvements in the approach based on modifications in the genetic algorithm. These improvements will primarily concern the consideration of a greater number of models for entering third-party characters into software models of hardware.