Model Agnostic Defence Against Backdoor Attacks in Machine Learning

Machine learning (ML) has automated a multitude of our day-to-day decision-making domains, such as education, employment, and driving automation. The continued success ML largely depends on ability to trust the model we are using. Recently, new class attacks called backdoor have been developed. These undermine user’s in models. In this article, present Neo , agnostic framework detect mitigate image classification For given model, approach analyzes inputs it receives determines if is backdoored. addition feature, also these by determining correct predictions poisoned images. We implemented evaluated against three state-of-the-art evaluation, show that can $\approx$ 88% average fast 4.4 ms per input image. compare with defence methodologies proposed for attacks. Our evaluation reveals despite being blackbox approach, more effective thwarting than existing techniques. Finally, reconstruct exact user effectively test their systems.