AN ALGORITHM TO PROTECT MACHINE LEARNING SYSTEMS FROM THE THREAT OF MODEL MODIFICATION BY DETECTING EMBEDDED MALICIOUS DATA

Chekmarev Maxim Alekseevich

The paper discusses the main security threats to machine learning systems and, in particular, the threat of model modification through the introduction of malicious data. A security algorithm based on the deployment technology of artificial neural networks is proposed. The algorithm has been tested using the developed computer program and a set of low-level features of the learning object, which allows to identify the embedded malicious data, has been obtained. Conclusions are drawn on the need for further research in this area.

Keywords:machine learning, artificial intelligence, security breanch, poisoning attack, artificial neural networks, low-level features