TrojAI

TrojAI software simulates real-world adversarial attacks (natural and malicious attacks) on AI models to monitor model or dataset security during training. As a result, TrojAI provides insights into the weaknesses of the AI model and dataset, enabling users to know precisely howt to alter their model or dataset to achieve better performing and more secure AI models. This enables users to identify points of vulnerability in their model and data set before deployment and thus reduces the time needed to retrain the model for failures that would occur post-deployment.