Intel Machine Inferred Code Similarity (MISIM)

Main articles:

• Evidential programming
• Logical programming
• Neuronets (neural networks)
• Machine learning (Machine Learning)

2020: Announcement of a system of machine programming of MISIM

On July 31, 2020 Intel provided automatic system of machine programming of Machine Inferred Code Similarity (MISIM) which is capable to distinguish, what this or that part of the software product is intended for. For this purpose a system studies structure of the source code and analyzes the code of programs with similar properties. Accuracy of MISIM up to 40 times exceeds the modern systems of verification of the code. The solution can be used for a broad spectrum of tasks — from recommendations about programming before automatic error correction. MISIM was developed by Intel together with the Massachusetts Institute of Technology (MIT) and Institute of Technology of Georgia.

Creation of the systems of exact detection of the similar code meanwhile remains an unresolved problem. It is still extremely difficult to modern computers to define degree of similarity of two fragments of the program on the basis of the analysis of their source code and also to understand that both fragments perform the same functions. The MISIM system is capable to define most precisely when two fragments of the source code execute similar calculations even if they have different algorithms and a data structure.

The key difference of MISIM from the existing systems of determination of the similar code is a context-dependent semantic structure (contest-aware semantic structure, CASS). The purpose of CASS — to define what this or that fragment of the code is intended for. It can be configured on a certain context — it allows it to collect information describing the code more effectively.

As soon as the code format is integrated into CASS, several neural networks estimate degree of similarity of two fragments on the basis of tasks which they should solve. So if two parts of the code look different in structure, but perform the same functions, neuronets will estimate them as similar.

Other feature of MISIM is that she does not use the compiler. It allows a system to analyze incomplete fragments of the code which are in development process that is one of the major properties for implementation of a system of hints and automatic error correction.

Having integrated all these approaches in a single system, researchers of Intel, MIT and Institute of Technology of Georgia found out that MISIM allows to identify more precisely up to 40 times similar fragments of the code, than other solutions existing today.

Now MISIM is in a completion stage, however, the project already passed from a research stage to demonstration models. They should implement the mechanism of recommendations to the source code for programmers who create the applications working in heterogeneous architecture. Such system will be able to distinguish problems of the developed algorithm directly in the course of its creation and to offer semantic similar, but more effective options of its implementation.^[1]