Threat-agnostic neural network discriminator

Abstract

A neural network discriminator includes a neural network autoencoder and a scorer. The scorer compares an original signal or signature with a reconstructed signal or signature generated in the neural network autoencoder. The scorer generates a similarity score based on the comparison. If the similarity score exceeds a threshold value, then the discriminator determines that the source of the original signal is a positive indicator. In some scenarios, the positive indicator refers to a threat, such as a missile or other enemy object that may cause harm, from an open set of signal signatures that the discriminator has never learned or never been exposed to. The discriminator can also determine clutter signatures from the open set of signature that the discriminator has never been exposed to.