Convolutional Neural Network Copy Detection with Neural Network Perceptual Hashing

In recent years, many model intellectual property (IP) proof methods for IP protection have been proposed, such as model watermarking and model fingerprinting. However, as an important part of the model IP protection system, the model copy detection task has not received enough attention. With the increasing number of neural network models transmitted and deployed on the Internet, the search for similar models is in great demand, which concurrently triggers the security problem of copy detection of models for IP protection. Due to the high computational complexity, both model watermarking and model fingerprinting lack the capability to efficiently find suspected infringing models among tens of millions of models. In this paper, inspired by the hash-based image retrieval methods, we propose a perceptual hashing algorithm for convolutional neural networks (CNNs). The proposed perceptual hashing algorithm can convert the weights of CNNs to fixed-length binary hash codes so that the lightly modified version has the similar hash code as the original model. By comparing the similarity of a pair of hash codes between a query model and a test model in the model library, the similar versions of a query model can be retrieved efficiently. To the best of our knowledge, this is the first perceptual hashing algorithm for CNNs. The experiment performed on a model library containing 3,565 models indicates that our proposed perceptual hashing scheme has a superior copy detection performance.

Convolutional Neural Network Copy Detection with Neural Network Perceptual Hashing

In recent years, many model intellectual property (IP) proof methods for IP protection have been proposed, such as model watermarking and model fingerprinting. However, as an important part of the model IP protection system, the model copy detection task has not received enough attention. With the increasing number of neural network models transmitted and deployed on the Internet, the search for similar models is in great demand, which concurrently triggers the security problem of copy detection of models for IP protection. Due to the high computational complexity, both model watermarking and model fingerprinting lack the capability to efficiently find suspected infringing models among tens of millions of models. In this paper, inspired by the hash-based image retrieval methods, we propose a perceptual hashing algorithm for convolutional neural networks (CNNs). The proposed perceptual hashing algorithm can convert the weights of CNNs to fixed-length binary hash codes so that the lightly modified version has the similar hash code as the original model. By comparing the similarity of a pair of hash codes between a query model and a test model in the model library, the similar versions of a query model can be retrieved efficiently. To the best of our knowledge, this is the first perceptual hashing algorithm for CNNs. The experiment performed on a model library containing 3,565 models indicates that our proposed perceptual hashing scheme has a superior copy detection performance.

Secure Data Collaborative Computing Scheme Based on Blockchain

With the rapid development of information technology, different organizations cooperate with each other to share data information and make full use of data value. Not only should the integrity and privacy of data be guaranteed but also the collaborative computing should be carried out on the basis of data sharing. In this paper, in order to achieve the fairness of data security sharing and collaborative computing, a security data collaborative computing scheme based on blockchain is proposed. A data storage query model based on Bloom filter is designed to improve the efficiency of data query sharing. The MPC contract is designed according to the specific requirements. The participants are rational, and the contract encourages the participants to implement the agreement honestly to achieve fair calculation. A secure multiparty computation based on secret sharing is introduced. The problem of identity and vote privacy in electronic voting is solved. The scheme is analyzed and discussed from storage expansion, anticollusion, verifiability, and privacy.

Towards Rank-Aware Data Mashups

Data mashups are web applications that combine complementary (raw) data pieces from different data services or web data APIs to provide value added information to users. They became so popular over the last few years; their applications are numerous and vary from addressing transient business needs in modern enterprises. Even though data mashups have been the focus of many research works, they still face many challenging issues that have never been explored. The ranking of the data returned by a data mashup is one of the key issues that have received little consideration. Top-k query model ranks the pertinent answers according to a given ranking function and returns only the best results. This paper proposes two algorithms that optimize the evaluation of top-k queries over data mashups. These algorithms are built based on the web data APIs' access methods: bind probe and indexed probe.

A Sigmoid based Learning in Heterogeneous Distortion for Data Privacy

Neural network-based learning models along with an access to huge data have made a remarkable outcome in recent years. These models are contributing a lot to improvise the working dimensions of various domains like Speech recognition, Image processing, Text analysis and many more. The well represented data is the main resource in the current research, but this data is often privacy sensitive and it definitely needs a proper attention failing which leads to serious privacy concerns. The proposed work demonstrates how learning models can be applied to analyze the data sensitivity and classify them to various privacy classes. Once the privacy class distribution is performed the model applies Inverse laplacian query model to check the data utility. The data should not get compromised on utility with the curse of privacy. With this intention the given experimental study succeeded in training the network to perform privacy analysis under a modest privacy budget, complexity training efficiency and data utility