METHOD FOR LEARNING AND TESTING USER LEARNING NETWORK TO BE USED FOR RECOGNIZING OBFUSCATED DATA CREATED BY CONCEALING ORIGINAL DATA TO PROTECT PERSONAL INFORMATION AND LEARNING DEVICE AND TESTING DEVICE USING THE SAME

摘要

The present invention provides a method for learning a user learning network for recognizing modulated data in which the original data is de-identified for protection of personal information, (a) the learning device modulates the original data, and the original data cannot be distinguished. In the learning network, through a modulation network trained to generate modulated data that is recognized as identical to the original data, 1_1 original data for training to 1_m - Where m is an integer greater than or equal to 1 - First modulation data for learning by modulating each of the original data for training Obtain an m-th modulated data for training from a data provider, and 1_i in the 1_1 modulated data for learning to the 1_m modulated data for learning - The i is an integer greater than or equal to 1 and equal to or less than the m - Input the modulated data for learning to a user learning network to cause the user learning network to modulate the 1_i learning through at least one task layer and at least one first batch normalization layer that adjusts the average and variance in the output of the at least one task layer. The first learning characteristic information is generated by performing a learning operation on the data, and the first learning task specific output generated using the first learning characteristic information or the first learning characteristic information, and the first_i first learning modulated data training the user learning network by updating parameters of the at least one first batch normalization layer and the at least one task layer to minimize a first error generated by referring to 1 ground truth; and (b) the learning device obtains, by the learning device, the 2_1th original data for learning to 2_nth - where n is an integer greater than or equal to 1 - from the user, and the second_1th original data for learning to the 2_nth original data for learning. 2_j - the j is an integer greater than or equal to 1 and less than or equal to n - input raw data for training into the user learning network to cause the user learning network to perform the at least one task layer and the average in the output of the at least one task layer Through at least one second batch normalization layer that adjusts variance, the second_j original data for learning is subjected to a learning operation to generate second feature information for learning, and the second feature information for learning or the second feature information for learning is used. The at least one second batch normalization layer and the at least one task to minimize a second error generated by referring to the second learning task specific output generated by training the user learning network by updating the parameters of the layer; It relates to a method comprising