Abstract: System and method for robust machine learning (ML) includes an attack detector comprising one or more deep neural networks trained using adversarial examples generated from a generative adversarial network (GAN), producing an alertness score based on a likelihood of an input being adversarial. A dynamic ensemble of individually robust ML models of various types and sizes and all being trained to perform an ML-based prediction is dynamically adapted by types and sizes of ML models to be deployed during the inference stage of operation. The adaptive ensemble is responsive to the alertness score received from the attack detector. A data protector module with interpretable neural network models is configured to prescreen training data for the ensemble to detect potential data poisoning or backdoor triggers in initial training data.

Abstract: A method for testing software applications in a system under test (SUT) includes building a reference model of the SUT that defines a computer-based neural network. The method includes training the reference model using input data and corresponding output data generated by the SUT, selecting an output value within a domain of possible output values of the SUT representing an output that is not represented in the output data used to train the reference model, applying the selected output value to the reference model, and tracing the selected output through the reference model to identify test input values that when input to the reference model, produce the selected output value. The method can further include using the identified test input values to test the system under test.

Abstract: A method for testing software applications in a system under test (SUT) includes building a reference model of the SUT comprising a computer-based neural network, training the reference model using input data and corresponding output data generated by the SUT, selecting an output value within a domain of possible output values of the SUT representing an output that is not represented in the output data used to train the reference model, applying the selected output value to the reference model and tracing the selected output through the reference model to identify test input values that when input to the reference model, produce the selected output value and using the identified test input values to test the system under test.