Data Poisoning Attack on Federated Machine Learning System
Prof. (Mr.) Sagar B. Shinde, Atharva Dedge, Prerana Roy, Pari Zunake
Department of Computer Engineering Modern Education Society’s College of Engineering Pune, India
Abstract — Federated Machine Learning (FML) is a distributed machine learning framework that enables multiple participants to collaborate and train a machine learning model for tasks such as classification, prediction, and recommendation. In FML, raw data owned by different participants is protected through secure and privacy-preserving techniques, ensuring that it cannot be tampered with, disclosed or reverse-engineered. The framework has the potential to be applied in various use cases and provides a solution to the challenges posed by centralized machine learning. The objective of FML is to provide a brief overview of the technological landscape and the underlying principles of the framework, and its applications in real-life scenarios.
Data poisoning attacks on Federated Machine Learning (FML) refer to the malicious manipulation of machine learning models by adversaries. These attacks can undermine the accuracy and reliability of the models and pose a significant threat to the security and privacy of data. To address these challenges, research is being conducted to develop solutions that can defend against data poisoning attacks in FML. This includes the use of optimization methods and the analysis of optimal data poisoning attacks to find solutions to the challenges posed by these attacks in the federated learning setting. The goal is to make FML safer and more secure, protecting the data and the models from malicious activities.
This paper provides an overview of data poisoning attacks on federated machine learning and their implications. We describe the common types of data poisoning attacks, such as label flipping and data injection, and discuss their impact on the performance and security of federated machine learning. Additionally, we discuss the various defense mechanisms, such as data sanitization and robust aggregation, that can be employed to mitigate the effects of data poisoning attacks. We conclude by highlighting the challenges and future research directions in securing federated machine learning systems against data poisoning attacks.
