LEARNING METHOD AND LEARNING APPARATUS OF ARTIFICIAL INTELLIGENCE FIREWALL USING SERVICE COMMUNICATIONS RULES AND COMMUNICATIONS PATTERNS, AND TESTING METHOD AND TESTING APPARATUS USING SAME

摘要

Disclosed is a learning method of an artificial intelligence firewall using service communications rules and communications patterns. More specifically, the learning method comprises: (a) a step in which a learning apparatus allows at least one indication layer included in neural networks to indicate processing methods for a packet corresponding to a learning header value while generating an unstructured indication value in a non-standardized state by applying at least one neural network operation to the learning header value when obtaining the learning header value of packets generated by referring to the service communications rules and the communications patterns; (b) a step in which the learning apparatus allows at least one standardization layer included in the neural networks to generate a standardized indication value indicating at least part of the processing methods for the packet including at least one ALLOW processing method and at least one REFUSE processing method by applying at least one SoftMax operation to the unstructured indication value; and (c) a step in which the learning apparatus determines a processing method for a packet corresponding to the learning header value by referring to the standardized indication value, and allows a loss layer included in the neural network to generate loss by referring to the determined at least part of the processing methods and GT corresponding to the learning header value, and to learn at least part of layers included in the neural network by performing back propagation by referring to the loss.;COPYRIGHT KIPO 2020