AI model and data transforming techniques for cloud edge
Systems and techniques for AI model and data camouflaging techniques for cloud edge are described herein. In an example, a neural network transformation system is adapted to receive, from a client, camouflaged input data, the camouflaged input data resulting from application of a first encoding transformation to raw input data. The neural network transformation system may be further adapted to use the camouflaged input data as input to a neural network model, the neural network model created using a training data set created by applying the first encoding transformation on training data. The neural network transformation system may be further adapted to receive a result from the neural network model and transmit output data to the client, the output data based on the result.
1. A system for processing transformed data with a neural network model in an edge computing environment, comprising:
a first computing device configured to:
generate the neural network model with a transformed training data set, wherein the transformed training data set is created from raw training data applied with a first encoding transformation;
a second computing device configured to:
receive, on an unsecure channel in the edge computing environment, transformed input data from an edge computing device, the transformed input data resulting from application of a second encoding transformation to raw sensor data collected by the edge computing device;
apply a first transformation to the transformed input data, wherein the first transformation includes decoding the transformed input data with the second encoding to produce a first intermediate result and encoding the first intermediate result with the first encoding to produce a second trans formed input data;
provide the second transformed input data as input to the neural network model;
receive transformed output data from the neural network model, wherein the transformed output data is transformed with the first encoding transformation;
apply a second transformation to the transformed output data, wherein the second transformation includes decoding the transformed output data with the first encoding to produce a second intermediate result and encoding the second intermediate result with the second encoding to produce a second transformed output data; and
transmit the second transformed output data to the edge computing device.
2. The system of claim 1 , wherein the transformed input data is part of an inference request to the neural network model and the transformed output data is a response to the inference request by the neural network model.
3. The system of claim 1 , wherein applying the first encoding of the first transformation includes scaling and rotating of the first intermediate result.
4. The system of claim 1 , further comprising instructions to:
receive, from the edge computing device, an identifier; and
select the neural network model from a plurality of neural network models based on the identifier.
5. The system of claim 4 , wherein the identifier is a device identifier.
6. The system of claim 4 , wherein the identifier is an application identifier.
7. The system of claim 4 , further comprising instructions to:
perform a query to a table associating identifiers and device-side encoding transformations using the identifier; and
identify, from the query, the second encoding transformation corresponding to the identifier.
8. The system of claim 7 , wherein the second encoding transformation is further identified based on a time of day.
9. At least one non-transitory computer readable medium including instructions for processing transformed data with a neural network model in an edge computing environment that when executed by at least one processor, cause the at least one processor to:
select the neural network model, the neural network model generated with a transformed training data set, wherein the transformed training data set is created from raw training data applied with a first encoding transformation;
receive, on an unsecure channel in the edge computing environment, transformed input data from an edge computing device, the transformed input data resulting from application of a second encoding transformation to raw sensor data collected by the edge computing device;
apply a first transformation to the transformed input data, wherein the first transformation includes decoding the transformed input data with the second encoding to produce a first intermediate result and encoding the first intermediate result with the first encoding to produce a second transformed input data;
provide the second transformed input data as input to the neural network model;
receive transformed output data from the neural network model, wherein the transformed output data is transformed with the first encoding transformation;
apply a second transformation to the transformed output data, wherein the second transformation includes decoding the transformed output data with the first encoding to produce a second intermediate result and encoding the second intermediate result with the second encoding to produce a second transformed output data; and
transmit the second trans formed output data to the edge computing device.
10. The at least one non-transitory computer readable medium of claim 9 , wherein the transformed input data is part of an inference request to the neural network model and the transformed output data is a response to the inference request by the neural network model.
11. The at least one non-transitory computer readable medium of claim 9 , wherein applying the first encoding transformation includes scaling and rotating of the first intermediate result.
12. The at least one non-transitory computer readable medium of claim 9 , further comprising instructions to:
receive, from the edge computing device, an identifier; and
select the neural network model from a plurality of neural network models based on the identifier.
13. The at least one non-transitory computer readable medium of claim 12 , wherein the identifier is a device identifier.
14. The at least one non-transitory computer readable medium of claim 12 , wherein the identifier is an application identifier.
15. The at least one non-transitory computer readable medium of claim 12 , further comprising instructions to:
perform a query to a table associating identifiers and device-side encoding transformations using the identifier; and
identify, from the query, the second encoding transformation corresponding to the identifier.
16. The at least one non-transitory computer readable medium of claim 15 , wherein the second encoding transformation is further identified based on a time of day.
17. A method for processing transformed data with a neural network model in an edge computing environment, comprising:
selecting the neural network model, the neural network model generated with a transformed training data set, wherein the transformed training data set is created from raw training data applied with a first encoding transformation;
receiving, on an unsecure channel in the edge computing environment, transformed input data from an edge computing device, the transformed input data resulting from application of a second encoding transformation to raw sensor data collected by the edge computing device;
applying a first transformation to the transformed input data, wherein the first transformation includes decoding the transformed input data with the second encoding to produce a first intermediate result and encoding the first intermediate result with the first encoding to produce a second transformed input data;
providing the second transformed input data as input to the neural network model;
receiving transformed output data from the neural network model, wherein the transformed output data is transformed with the first encoding transformation;
applying a second transformation to the transformed output data, wherein the second transformation includes decoding the transformed output data with the first encoding to produce a second intermediate result and encoding the second intermediate result with the second encoding to produce a second transformed output data; and
transmitting the second transformed output data to the edge computing device.
18. The method of claim 17 , wherein the transformed input data is part of an inference request to the neural network model and the transformed output data is a response to the inference request by the neural network model.
19. The method of claim 17 , wherein applying the first encoding of the first transformation includes scaling and rotating of the first intermediate result.
20. The method of claim 17 , further comprising:
receiving, from the edge computing device, an identifier; and
selecting the neural network model from a plurality of neural network models based on the identifier.
21. The method of claim 20 , wherein the identifier is a device identifier.
22. The method of claim 20 , wherein the identifier is an application identifier.
23. The method of claim 20 , further comprising:
performing a query to a table associating identifiers and device-side encoding transformations using the identifier; and
identifying, from the query, the second encoding transformation corresponding to the identifier.
24. The method of claim 23 , wherein the second encoding transformation is further identified based on a time of day.