Biblio

Deep Neural Networks (DNN) has gained great success in solving several challenging problems in recent years. It is well known that training a DNN model from scratch requires a lot of data and computational resources. However, using a pre-trained model directly or using it to initialize weights cost less time and often gets better results. Therefore, well pre-trained DNN models are valuable intellectual property that we should protect. In this work, we propose DeepTrace, a framework for model owners to secretly fingerprinting the target DNN model using a special trigger set and verifying from outputs. An embedded fingerprint can be extracted to uniquely identify the information of model owner and authorized users. Our framework benefits from both white-box and black-box verification, which makes it useful whether we know the model details or not. We evaluate the performance of DeepTrace on two different datasets, with different DNN architectures. Our experiment shows that, with the advantages of combining white-box and black-box verification, our framework has very little effect on model accuracy, and is robust against different model modifications. It also consumes very little computing resources when extracting fingerprint.

Data security is becoming more and more important in data exchange. Once the data is leaked, it will pose a great threat to the privacy and property security of users. Copyright authentication and data provenance have become an important requirement of the information security defense mechanism. In order to solve the collusion leakage of the data distributed by organization and the low efficiency of tracking the leak provenance after the data is destroyed, this paper proposes a concatenated-group digital fingerprint coding based on CFF code and Reed-solomon (RS) that can resist collusion attacks and corresponding detection algorithm. The experiments based on an asymmetric anti-collusion fingerprint protocol show that the proposed method has better performance to resist collusion attacks than similar non-grouped fingerprint coding and effectively reduces the percentage of misjudgment, which verifies the availability of the algorithm and enriches the means of organization data security audit.

Digital fingerprinting refers to as method that can assign each copy of an intellectual property (IP) a distinct fingerprint. It was introduced for the purpose of protecting legal and honest IP users. The unique fingerprint can be used to identify the IP or a chip that contains the IP. However, existing fingerprinting techniques are not practical due to expensive cost of creating fingerprints and the lack of effective methods to verify the fingerprints. In the paper, we study a practical scan chain based fingerprinting method, where the digital fingerprint is generated by selecting the Q-SD or Q'-SD connection during the design of scan chains. This method has two major advantages. First, fingerprints are created as a post-silicon procedure and therefore there will be little fabrication overhead. Second, altering the Q-SD or Q'-SD connection style requires the modification of test vectors for each fingerprinted IP in order to maintain the fault coverage. This enables us to verify the fingerprint by inspecting the test vectors without opening up the chip to check the Q-SD or Q'-SD connection styles. We perform experiment on standard benchmarks to demonstrate that our approach has low design overhead. We also conduct security analysis to show that such fingerprints are robust against various attacks.