1. A system for providing explanation for output generated by an artificial intelligence (AI)
model, the system comprising:
an AI prediction and explanation device in communication with a client device over a communication network, the AI prediction and explanation device comprising at least one processor and a computer-readable medium storing instructions that, when executed by the at least one processor, cause the at least one processor to perform operations comprising:
receiving encrypted input data and a public encryption key from the client device, wherein the encrypted input data is encrypted using the public encryption key;
generating an encrypted AI model by encrypting an AI model using the public encryption key;
generating an encrypted output and an encrypted feature data based on the encrypted input data using the encrypted AI model;
generating an encrypted explanation for the encrypted output based on the encrypted feature data; and
providing the encrypted output and the encrypted explanation to the client device for rendering, wherein the encrypted output and the encrypted explanation are decrypted by the client device using a private encryption key.
2. The system of claim 1, wherein the client device generates the public and the private encryption keys based on a user input using a homomorphic encryption algorithm, and wherein the user input comprises a desired level of security, a desired level of computational complexity, and a number of key generation parameters.
3. The system of claim 1, wherein the operation further comprise training the AI model using unencrypted training data.
4. The system of claim 1, wherein generating the encrypted AI model comprises:

generating an approximate AI model based on the AI model, wherein the approximate AI model comprises an approximate linear function corresponding to the non-linear activation function of the AI model; and
encrypting a plurality of model parameters of the approximate AI model using the public encryption keys, wherein the plurality of model parameters comprises at least the approximate linear function and a plurality of weights of each node in each layer of the AI model.
5. The system of claim 1, wherein the AI model comprises a first artificial neural network (ANN) model and a second ANN model, wherein generating the encrypted output and the encrypted feature data comprises employing the first ANN model, and wherein generating the encrypted explanation comprises employing the second ANN model.
6. The system of claim 5, wherein the first ANN model is a convolutional neural network (CNN) model, and wherein the second ANN model is a long short-term memory (LSTM) model.
7. The system of claim 5, wherein the operations further comprise:
generating a set of encrypted intermediate feature data corresponding to a set of layers of the first ANN model; and
providing the set of encrypted intermediate feature data to the client device for rendering, wherein the set of encrypted intermediate feature data are decrypted by the client device using the private encryption key.
8. The system of claim 7, wherein the operations further comprise validating at least one of
an encryption algorithm employed by the client device for generation of the public and the
private encryption keys or the encrypted AI model, wherein the validating comprises:
at a first pass, generating an output and a set of intermediate feature data corresponding to the set of layers of the first ANN model based on a probe data and the AI model;

at a second pass, generating the encrypted output and the set of encrypted intermediate feature data corresponding to the set of layers of the first ANN model based on an encrypted probe data and the encrypted AI model; and
providing the output and the set of intermediate feature data corresponding to the probe data as well as the encrypted output and the set of encrypted intermediate feature data corresponding to the encrypted probe data to the client device, wherein the encrypted probe data output and the set of encrypted intermediate feature data are decrypted by the client device for comparison with the output and the set of intermediate feature data, and wherein validation is based on the comparison.
9. The method of claim 8, wherein the operation further comprise retuning at least one of the
encryption algorithm or the encrypted AI model based on the comparison.
10. A method of providing explanation for output generated by artificial intelligence (AI)
model, the method comprising:
receiving, by an AI prediction and explanation device, encrypted input data and a public encryption key from a client device, wherein the encrypted input data is encrypted using the public encryption key;
generating, by the AI prediction and explanation device, an encrypted AI model by encrypting an AI model using the public encryption key;
generating, by the AI prediction and explanation device, an encrypted output and an encrypted feature data based on the encrypted input data using the encrypted AI model;
generating, by the AI prediction and explanation device, an encrypted explanation for the encrypted output based on the encrypted feature data; and
providing, by the AI prediction and explanation device, the encrypted output and the encrypted explanation to the client device for rendering, wherein the encrypted output and the encrypted explanation are decrypted by the client device using a private encryption key.
11. The method of claim 10, wherein the AI model comprises a first artificial neural
network (ANN) model and a second ANN model, wherein generating the encrypted output

and the encrypted feature data comprises employing the first ANN model, and wherein generating the encrypted explanation comprises employing the second ANN model.
12. The method of claim 11, further comprising:
generating, by the AI prediction and explanation device, a set of encrypted intermediate feature data corresponding to a set of layers of the first ANN model; and
providing, by the AI prediction and explanation device, the set of encrypted intermediate feature data to the client device for rendering, wherein the set of encrypted intermediate feature data are decrypted by the client device using the private encryption key.
13. The method of claim 12, further comprising validating at least one of an encryption
algorithm employed by the client device for generation of the public and the private
encryption keys or the encrypted AI model, wherein the validating comprises:
at a first pass, generating, by the AI prediction and explanation device, an output and a set of intermediate feature data corresponding to the set of layers of the first ANN model based on a probe data and the AI model;
at a second pass, generating, by the AI prediction and explanation device, the encrypted output and the set of encrypted intermediate feature data corresponding to the set of layers of the first ANN model based on an encrypted probe data and the encrypted AI model; and
providing, by the AI prediction and explanation device, the output and the set of intermediate feature data corresponding to the probe data as well as the encrypted output and the set of encrypted intermediate feature data corresponding to the encrypted probe data to the client device, wherein the encrypted probe data output and the set of encrypted intermediate feature data are decrypted by the client device for comparison with the output and the set of intermediate feature data; and wherein validation is based on the comparison.