4-8 bit quantized DNNs stay exposed to membership inference
Eight attacks across six quantization methods find privacy gains mainly at 2 bits, with large utility loss.
Stories per day · 30 days
Relatedadversarial-ml1arxiv1backdoor1batch-normalization1federated-learning1image-classification1measurement1membership-inference1model-ownership1model-privacy1qnn1quantization1stealthiness1watermarking1
4 stories, newest first · one card per story
Stories tagged #dnn
Data-free attack removes batch-norm watermarks from federated models
Tests on two federated-learning schemes strip ownership marks without training data and with little accuracy loss.
Quiet Triggers, Loud Footprints: A Tri-Space Measurement Study of Backdoor Stealthiness
A study finds image-classifier backdoors that look stealthy in one space often leave clear footprints in another.
OptiPrime: Optimizing Private Inference through Protocol-Hardware Co-design
OptiPrime co-designs HE-MPC protocols with hardware acceleration to remove network communication bottlenecks in private DNN inference, beating Cheetah by up to 5.7x.