Real Time Poisoning Attacks and Privacy Strategies on Machine Learning Systems
A T Archa, K. Kartheeban · 2024
Machine Learning aims to learn computer systems and predict output. Nowadays, Deep learning uses continuous training based on a large network of interconnected neurons to mimic the way humans think, analyze, and make decisions. It plays an important role in real-time applications. There are some security and privacy threats faced by deep learning systems. One of the major threats faced by such learning systems is attacks during training. The manipulation of training data leads to poisoning attacks. The resulting models are severely impacted by poisoning attacks, which may prevent them from converging or alter the outcomes of their predictions. The need to defend against such poisoning attacks is urgent and difficult. This study explains a thorough and detailed analysis of different types of poisoning attacks in real time applications. Its countermeasures to mitigate such attacks and various privacy-enhancing methods are also discussed here.