Privacy Preserving Back-Propagation Neural Network Learning Made Practical with Cloud Computing
To improve the accuracy of learning result, in practice multiple parties may collaborate through conducting joint Back-propagation neural network learning on the union of their respective data sets. During this process no party wants to disclose her/his private data to others. Existing schemes supporting this kind of collaborative learning are either limited in the way of data partition or just consider two parties. There lacks a solution that allows two or more parties, each with an arbitrarily partitioned data set, to collaboratively conduct the learning. This paper solves this open problem by utilizing the power of cloud computing. In our proposed scheme, each party encrypts his/her private data locally and uploads the ciphertexts into the cloud. The cloud then executes most of the operations pertaining to the learning algorithms over ciphertexts without knowing the original private data. By securely offloading the expensive operations to the cloud, we keep the computation and communication costs on each party minimal and independent to the number of participants. To support flexible operations over ciphertexts, we adopt and tailor the BGN`doubly homomorphic’ encryption algorithm for the multi-party setting. Numerical analysis and experiments on commodity cloud show that our scheme is secure, efficient and accurate.
In the Existing schemes supporting this kind of collaborative learning are either limited in the way of data partition or just consider two parties. There lacks a solution that allows two or more parties, each with an arbitrarily partitioned data set, to collaboratively conduct the learning. Several privacy preserving BPN network learning schemes have been proposed recently. Schlitter introduces a privacy preserving BPN network learning scheme that enables two or more parties to jointly perform BPN network learning without disclosing their respective private data sets. But the solution is proposed only for horizontal partitioned data. Moreover, this scheme cannot protect the intermediate results, which may also contain sensitive data, during the learning process.
In the propose system each participant first encrypts her/his private data with the system public key and then uploads the ciphertexts to the cloud; cloud servers then execute most of the operations pertaining to the learning process over the ciphertexts and return the encrypted results to the participants; the participants jointly decrypt the results with which they update their respective weights for the BPN network. During this process, cloud servers learn no privacy data of a participant even if they collude with all the rest participants. Through off-loading the computation tasks to the resource-abundant cloud, our scheme makes the computation and communication complexity on each participant independent to the number of participants and is thus highly scalable. For privacy preservation, we decompose most of the sub-algorithms of BPN network into simple operations such as addition, multiplication, and scalar product. To support these operations over cipher texts, we adopt the BGN doubly homomorphism encryption algorithm and tailor it to split the decryption capability among multiple participants for collusion-resistance decryption.