A Protection against the Extraction of Neural Network Models

Given oracle access to a Neural Network (NN), it is possible to extract its underlying model. We here introduce a protection by adding parasitic layers which mostly keep unchanged the underlying NN while complexifying the task of reverse-engineering. Our countermeasure relies on approximating the identity mapping with a Convolutional NN. We explain why the introduction of new parasitic layers complexifies the attacks. We report experiments regarding the performance and the accuracy of the protected NN.

Premium Access to Convolutional Neural Networks

Neural Networks (NNs) are today used for all our daily tasks; for instance, in mobile phones. We here want to show how to restrict their access to privileged users. Our solution relies on a degraded implementation which can be corrected thanks to a PIN. We explain how to select a few parameters in an NN so as to maximize the gap in the accuracy between the premium and the degraded modes. We report experiments on an implementation of our proposal on a deep NN to prove its practicability.