Privacy-Preserving Nonlinear SVM Classifier Training Based on Blockchain

Abstract

The SVM classifier has been a significant and prevailing technique in machine learning applications. Training a high-quality SVM classifier usually requires a huge amount of data, which makes collaborative training by multiple parties become an inevitable trend. However, it causes privacy risks when sharing sensitive data with others. There are some existing methods to solve this problem. These methods mainly contain computation-intensive cryptographic techniques which are inefficient and not suitable for practical use. Therefore, it is important to realize efficient SVM classifier training while protecting privacy. In this paper, we propose a novel privacy-preserving SVM classifier training scheme based on blockchain. We establish a blockchain-based SVM classifier training mechanism which realizes collaboratively training while protecting privacy. We adopt the additive secret sharing technique to design several computation protocols, which are much more efficient than the schemes which contain complex cryptographic primitives. We conduct a thorough analysis of the security properties of our scheme. Experiments over a real dataset show that our scheme achieves high accuracy and practical efficiency.