Hidden errors finding in memory programming usage for C++ code by means of static Deep Learning analysis

Abstract

In our time, a memory usage by program code is one of the important problems both: from the side of code writing and from the side of application protection. Resource saving and secure code play an important role today, even before the industrial programs implementation, when memory usage errors cannot be detected by traditional means. Undetected memory usage errors can cause unnecessary resource consumption as well as become a source of intrusions or unauthorized information seizure. This article outlines the main advantages, settings, modeling, code study special features and other aspects of applying the Deep learning static analysis based on neural networks to detect memory usage errors in C++ program code. Some important role in its usage applies the practical study of have already been executed analyzed code samples with known memory allocation problems, saved in the database and used as basic templates. Based on the large amount of code study examples by Deep learning means from several software projects were used to detect errors that were undetected by other traditional methods, and its performance, obtained results visualization, and reduction of false positives were also discussed here