SYSTEM AND METHOD FOR TRAINING HARMFUL CONTAINER DETECTION MODEL Russian patent published in 2019 - IPC G06F21/56 G06N20/00 

FIELD: physics.

SUBSTANCE: invention relates to computer engineering. Disclosed is system for detecting harmful containers, wherein the container is a file containing at least two and more objects, which are logically separate data areas of said container (hereinafter - objects), which contains: parameter determining means for determining parameters of each object selected from the analysed container, wherein one of parameters characterizes functional connection of objects among themselves; convolution generating means for convolution formation based on object parameters determined by the parameter determining means, which is a multidimensional vector, each element of which corresponds to its unique parameter from certain parameters, and value of said element corresponds to number of objects for which said parameter was determined; training apparatus for machine learning of harmful containers detection model based on at least one convolution formed for safe container and one harmful container, wherein said detection model is an algorithm for calculating the degree of harmfulness of a container, and the degree of harmfulness of the container is a numerical value which characterizes the probability that the said container is harmful.

EFFECT: technical result is training malware detection model.

10 cl, 5 dwg, 2 tbl