<ul><li>Federated Learning (FL) is vulnerable to Model Extraction (ME) attacks that threaten Intellectual Property (IP) in Machine Learning as a Service (MLaaS) platforms.</li><li>A study examined the vulnerability of FL-based victim models to ME attacks, evaluating performance across deep learning architectures and image datasets.</li><li>Experimental results show that the accuracy and fidelity of extracted models in FL are influenced by the size of the attack query set.</li><li>Transfer learning is explored as an approach for ME attacks, indicating higher accuracy and fidelity in fine-tuned pretrained extraction models, especially with smaller query sets.</li></ul>

Evaluating Query Efficiency and Accuracy of Transfer Learning-based Model Extraction Attack in Federated Learning

Discover more