A Decision Support Tool to Evaluate the Robustness of Federated Learning to Data Poisoning Attacks
Open AccessFederated Learning is a popular privacy-preserving machine learning methodology that has been used in many critical applications. However, more recently, researchers have found that Federated Learning is vulnerable to various poisoning attacks that compromise the integrity of the machine learning model and can cause the model to output incorrect predictions. There are many examples of poisoning attacks on Federated Learning that can have disastrous consequences. For example, Federated Learning is used by hospitals. It is crucial to ensure that Federated Learning is secure as incorrect predictions could potentially impact the lives of patients relying on applications that use Federated Learning for medical imaging diagnostics. This research examines the impact of data poisoning attacks on Federated Learning. The main contributions of this research are developing a set of statistical metrics that determine the resiliency of Federated Learning algorithms to data poisoning attacks and a decision support tool that determines which Federated Learning algorithm is most resilient when faced with data poisoning attacks. The decision support tool developed as a result of this research will enable enterprise users to determine the impact of data poisoning attacks on Federated Learning algorithms in their environment. Therefore, enterprises can make an informed decision as to which Federated Learning algorithm would be the most resilient to data poisoning attacks.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.