Systems and methods for quantifying data leakage from a split layer
A system and method are disclosed for providing an artificial intelligence platform. An example method includes examining part of a global neural network to locate a split layer in the global neural network, creating an equivalent model to the part of the global neural network of a same size but having opposite operations, generating smashed data based on an operation on input data by the part of the global neural network, training the equivalent model by inputting the smashed data to generate a second a mirrored copy of the input data, quantifying a distance between the input data and the second generated set of mirrored data to yield a value and, when the value is below a threshold, determining that a current location of the split layer in the global neural network is safe for a training process.
1. A method comprising:
examining part of a global neural network to locate a split layer in the global neural network, the part of the global neural network implementing first operations, wherein the part of the global neural network implementing first operations comprises one or more neural network layers performing a respective function for each respective layer of the one or more neural network layers;
creating an equivalent model to the part of the global neural network of a same size but implementing second operations that are opposite operations in functionality in terms of model inversion and relative to the first operations of the part of the global neural network;
generating smashed data based on a first operation of the first operations on a first input data by the part of the global neural network;
training the equivalent model by inputting the smashed data to generate a second input data using a second operation of the second operations;
quantifying a distance between the first input data and the second input data to yield a value that represents a data leakage level; and
when the value is below a threshold, determining that a current location of the split layer in the global neural network is safe for a future training process for the global neural network.
2. The method of claim 1 , wherein the equivalent model applies a mean squared error function.
3. The method of claim 1 , wherein the generating of the smashed data comprises a warmup training task of the part of the global neural network.
4. The method of claim 1 , wherein, when the value is below the threshold, beginning a training process of the global neural network.
5. The method of claim 1 , wherein, when the value is at or above the threshold, declining a training process.
6. The method of claim 5 , further comprising requesting to move the split layer deeper into the global neural network.
7. The method of claim 6 , further comprising changing an architecture associated with the global neural network prior to approving of a training task.
8. The method of claim 1 , wherein a server sends the part of the global neural network to a client.
9. The method of claim 8 , wherein the client operates a tool that examines part of the global neural network to locate the split layer in the global neural network.
10. The method of claim 1 , wherein training the equivalent model comprises at least one of a split learning process and a blind learning process.
11. A system comprising:
a processor; and
a computer-readable storage device storing instructions which, when executed by the processor, cause the processor to perform operations comprising:
examining part of a global neural network to locate a split layer in the global neural network, the part of the global neural network implementing first operations, wherein the part of the global neural network implementing first operations comprises one or more neural network layers performing a respective function for each respective layer of the one or more neural network layers;
creating an equivalent model to the part of the global neural network of a same size but implementing second operations that are opposite operations in functionality in terms of model inversion and relative to the first operations of the part of the global neural network;
generating smashed data based on a first operation of the first operations on a first input data by the part of the global neural network;
training the equivalent model by inputting the smashed data to generate a second input data using a second operation of the second operations;
quantifying a distance between the first input data and the second input data to yield a value that represents a data leakage level; and
when the value is below a threshold, determining that a current location of the split layer in the global neural network is safe for a future training process for the global neural network.
12. The system of claim 11 , wherein the equivalent model applies a mean squared error function.
13. The system of claim 11 , wherein the generating of the smashed data comprises a warmup training task of the part of the global neural network.
14. The system of claim 11 , wherein, when the value is below the threshold, beginning a training process of the global neural network.
15. The system of claim 11 , wherein, when the value is at or above the threshold, declining a training process.
16. The system of claim 15 , further comprising requesting to move the split layer deeper into the global neural network.
17. The system of claim 16 , further comprising changing an architecture associated with the global neural network prior to approving of a training task.
18. The system of claim 11 , wherein a server sends the part of the global neural network to a client.
19. The system of claim 18 , wherein the client operates a tool that examines part of the global neural network to locate the split layer in the global neural network.
20. The system of claim 11 , wherein training the equivalent model comprises at least one of a split learning process and a blind learning process.