Privacy Preserving Back-Propagation Based on BGV on Cloud
Fanyu Bu, Yu Ma, Zhikui Chen, Han Xu · 2015
Back-propagation is the most effective algorithm for training deep learning models that are proved to have a great ability for big data feature learning. However, back-propagation is of high time complexity, leading to a low efficiency of big data learning. Aiming at this problem, the paper proposes a privacy preserving back-propagation algorithm based on the BGV encryption scheme on cloud. The proposed algorithm improved the efficiency of back-propagation learning by offloading the expensive operations on the cloud. Furthermore, the BGV encryption scheme is used to protect the private data during the learning process using the power of the cloud computing. Experiments show that our proposed scheme is secure and efficient.