scwyllie/csc413-project

This research project presents a defense against model extraction attacks that uses *proof-of-work* calibrated with differential privacy to significantly increase the computational cost for attackers. It is designed for **machine learning practitioners and researchers** who need to protect publicly deployed models from being stolen via API queries, offering a practical trade-off that minimally impacts legitimate users. The repository provides code to simulate attacks and the defense across standard datasets like MNIST and CIFAR-