Abstract: To improve the accuracy of learning result, in observe multiple parties might collaborate through conducting joint Back propagation neural network learning on the union of their several knowledge sets. throughout this method no party needs to disclose her/his non-public knowledge to others. Existing schemes supporting this type of cooperative learning are either restricted within the means of information partition or simply take into account 2 parties. There lacks an answer that enables two or a lot of parties, every with an at random partitioned off information set, to collaboratively conduct the training. This paper solves this open drawback by utilizing the ability of cloud computing. In our planned theme, every party encrypts his/her non-public knowledge regionally and uploads the ciphertexts into the cloud. By firmly offloading the high-ticket operations to the cloud, we tend to keep the computation and communication prices on every party nominal and freelance to the amount of participants. To support versatile operations over ciphertexts, we tend to adopt and tailor the BGN ‘doubly homomorphic’ coding algorithmic rule for the multi-party setting. Numerical analysis and experiments on goods cloud show that our theme is secure, economical and correct.

Keywords: Privacy reserving, Neural Network, Back-Propagation, Cloud computing, Computation Outsource.