The training and creation of deep learning model is usually costly, thus the trained model can be regarded as an intellectual property (IP) of the model creator. However, malicious users who obtain high-performance models may illegally copy, redistribute, or abuse the models without permission. To deal with such security threats, a few deep neural networks (DNN) IP protection methods have been proposed in recent years. This paper attempts to provide a review of the existing DNN IP protection works and also an outlook. First, we propose the first taxonomy for DNN IP protection methods in terms of six attributes: scenario, mechanism, capacity, type, function, and target models. Then, we present a survey on existing DNN IP protection works in terms of the above six attributes, especially focusing on the challenges these methods face, whether these methods can provide proactive protection, and their resistances to different levels of attacks. After that, we analyze the potential attacks on DNN IP protection methods from the aspects of model modifications, evasion attacks, and active attacks. Besides, a systematic evaluation method for DNN IP protection methods with respect to basic functional metrics, attack-resistance metrics, and customized metrics for different application scenarios is given. Lastly, challenges and future research opportunities on DNN IP protection are presented.

Naïve Bayes classifier (NBC) is a fundamental and widely-used data mining tool. To respond to the growing privacy concern, several privacy-preserving NBC schemes have been proposed recently. Generally, the existing schemes mainly exploit two different techniques: one is secure multi-party computation (SMCP); and the other is (local) differential privacy (DP/LDP). However, the approaches with SMCP would employ some encryption algorithms, which results in heavy computation and communication costs and impedes the usage of the approaches in practice. The existing methods with DP/LDP also have some deficiencies. For instance, a trusted data curator is required or the class labels of individuals are revealed. To make up the drawbacks of previous schemes, in this paper, we develop a novel scheme to train a Naïve Bayes classifier with privacy guarantee in the local setting. We first design a method (JESS) for joint distribution estimation under LDP. Then we apply JESS to calculate the conditional probability, the key of the Naïve Bayes classification and train a Naïve Bayes classifier. Additionally, we leverage extensive experiments to evaluate the effectiveness of our schemes designed for joint distribution and Naïve Bayes classification. Compared with the existing schemes, ours can achieve higher accuracy.

To take advantages of location-based services (LBS) while protecting user privacy against untrusted LBS providers, privacy preserving LBS have attracted increasing attention. Considering that users in an LBS system are often equipped with resource-constrained mobile devices, most existing privacy preserving LBS methods are based on anonymization techniques. However, these existing schemes still have some privacy and efficiency limitations. In this paper, we propose a novel privacy preserving LBS scheme, which simultaneously achieves user privacy protection and high query efficiency. Specifically, we utilize the k-anonymity technique and the pseudo random function to protect the location privacy and the query message privacy of users. We design an enhanced one-round blind filter protocol (ORBFe) based on the Paillier cryptosystem to securely filter out redundant records generated by the k-anonymity technique. Compared with existing solutions, our ORBFe protocol not only ensures that users receive exactly satisfying results but also incurs a low computation and communication cost on the server side. Through the theoretical analysis and extensive experiments, we demonstrate the security and efficiency of our proposed scheme.