109
Memory Backdoor Attacks on Neural Networks
arXiv:2411.14516v2 Announce Type: replace
Abstract: Neural networks are often trained on proprietary datasets, making them attractive attack targets. We present a novel dataset extraction method leveraging an innovative training time backdoor attack, allowing a malicious federated learning server t…
Abstract: Neural networks are often trained on proprietary datasets, making them attractive attack targets. We present a novel dataset extraction method leveraging an innovative training time backdoor attack, allowing a malicious federated learning server t…