Dhayalan.D, Jeevitha.R
To improve the accuracy of learning result in practice multiple parties may collaborate through conducting joint propagation neural network learning on the union of their respective data sets. In this process no party wants to disclose the private data to others. Back-propagation supporting this kind of collaborative learning takes long time to converge and may fall into local minima. One of the possible techniques to escape from local minima is by using a very small learning rate which will slow down the learning process. And can be solved by using “Optical Back-Propagation” algorithm which escapes from local minima with high speed of convergence during the training period. During this process it allows two or more parties with an arbitrarily partitioned data set to conduct the collaborative learning. In cloud computing each party encrypts the private data locally and uploads the cipher text to the cloud then the cloud executes most of the operation pertaining to the learning algorithm over cipher text without knowing the original private data. We keep the computation and communication cost on each party minimal and independent to the number of participants by using secure operations on the cloud. We use doubly homomorphism encryption algorithm for multiparty setting to support flexible operation over cipher texts. This shows that our scheme is secure, flexible and accurate.
Back-Propagation, Cloud Computing, Neural network, Optical Back-Propagation, Privacy preserving.