Privacy preservation of data over a shared network
System and method are disclosed for preserving privacy of shared data over a shared network. A vector encoder transforms received data into a feature vector. An autoencoder includes a neural network-based encoder transforms the feature vector into a fixed size latent space representation of the received data. A neural network-based decoder of the autoencoder is configured to reconstruct the feature vector from the latent space representation. The autoencoder is trained using training data with an objective to minimize reconstruction error. A vector decoder transforms the reconstructed feature vector into reconstructed data. The latent space representation of data from the trained autoencoder is shared as anonymized data with at least one trusted party over the shared network, decoded offline using a replica of the trained decoder.
1. A computer-based method for preserving privacy of shared data across a shared network, comprising:
transforming, by a vector encoder, input data into a feature vector;
transforming, by a neural network-based encoder of a trained autoencoder, the feature vector into anonymized data comprising a fixed size latent space representation of the input data,
transmitting the anonymized data to a trusted party over a shared network,
reconstructing by a neural network-based decoder of the trained autoencoder used by the trusted party, the feature vector from the anonymized data comprising the latent space representation; and
transforming, by a vector decoder used by the trusted party, the reconstructed feature vector into reconstructed data,
wherein the autoencoder including the neural network-based encoder and the neural net-work-based decoder is trained using training data with an objective of minimizing reconstruction error.
2. The method of claim 1 , wherein the shared data is text data.
3. The method of claim 1 , wherein the shared data is image data.
4. The method of claim 1 , wherein the shared data is sensor data.
5. The method of claim 1 , wherein the shared data is KKS tags.
6. The method of claim 1 , further comprising:
securely transmitting a replica of the trained neural network-based decoder and the vector de-coder to the trusted party subsequent to the training of the autoencoder.
7. The method of claim 6 , wherein the trusted party uses the replica of the trained neural network-based decoder and the vector decoder offline to decode the latent space representation to obtain reconstructed data corresponding to the input data.
8. A system for preserving privacy of shared data across a shared network, the system comprising:
one or more processors; and
a memory storing instructions that, when executed by the one or more processors, cause the system to perform a method according to claim 1 .
9. A non-transitory computer-readable medium storing instructions that, when executed by one or more processors, cause the one or more processors to perform a method according to claim 1 .