Methods, systems, articles of manufacture and apparatus to build privacy preserving models
Methods, apparatus, systems and articles of manufacture are disclosed to build privacy preserving models. An example apparatus disclosed herein includes a training manager to generate a first modeling plan for client-side resources, and transmit the first modeling plan to the client-side resources. The example apparatus also includes a data aggregator to search for a primary validation flag in response to retrieving client-side model parameters, and an accuracy calculator to, in response to detecting the primary validation flag, perform a secondary validation corresponding to the client-side model parameters using a server-side ground truth data set, and determine whether to update the global model with the client-side model parameters based on a comparison of results of the secondary validation and a validation threshold.
1. A computer system to update a global model, comprising:
memory;
machine readable instructions; and
processor circuitry to at least one of instantiate or execute the machine readable instructions to:
generate a first modeling plan for client-side resources, the first modeling plan including (a) tokenized model parameters devoid of telemetry labels and (b) a transmission schedule for the client-side resources;
transmit the first modeling plan to the client-side resources;
search for a primary validation flag in response to retrieving client-side model parameters;
in response to detecting the primary validation flag, perform a secondary validation corresponding to the client-side model parameters using a server-side ground truth data set, the secondary validation based on calculating a confidence metric in view of the server-side ground truth data set and determining that the confidence metric satisfies a validation threshold;
authorize updating of the global model in response to results of the secondary validation satisfying the validation threshold;
in response to the confidence metric not satisfying the validation threshold, prohibit updating the global model with the client-side model parameters;
send a second modeling plan to the client-side resources in response to rejecting the client-side modeling parameters; and
cause the client-side resources to execute the second modeling plan instead of the first modeling plan.
2. The computer system as defined in claim 1 , wherein the transmission schedule includes at least one of a target epoch, a quantity of training dataset iterations, a periodicity of communication with a server for a next training iteration, or a quantity of training samples to be processed by the client-side resources.
3. The computer system as defined in claim 1 , wherein the global model is to cause the client-side resources to at least one of identify computing device types or perform anomaly detection.
4. At least one non-transitory machine readable storage medium comprising instructions that cause a processor to at least:
generate a first modeling plan for client-side resources, the first modeling plan including (a) tokenized model parameters devoid of telemetry labels and (b) a transmission schedule for the client-side resources;
transmit the first modeling plan to the client-side resources;
search for a primary validation flag in response to retrieving client-side model parameters;
in response to detecting the primary validation flag, perform a secondary validation corresponding to the client-side model parameters using a server-side ground truth data set, the secondary validation based on calculating a confidence metric corresponding to the server-side ground truth data set and determining that the confidence metric satisfies a validation threshold;
authorize updating of a global model in response to results of the secondary validation satisfying the validation threshold;
in response to the confidence metric not satisfying the validation threshold, prohibit updating the global model with the client-side model parameters;
send a second modeling plan to the client-side resources in response to rejecting the client-side modeling parameters; and
cause the client-side resources to execute the second modeling plan instead of the first modeling plan.
5. The at least one non-transitory machine readable storage medium as defined in claim 4 , wherein the transmission schedule includes at least one of a target epoch, a quantity of training dataset iterations, or a quantity of training samples to be processed by the client-side resources.
6. The at least one non-transitory machine readable storage medium as defined in claim 4 , wherein the instructions cause the processor to cause the client-side resources to at least one of identify computing device types or perform anomaly detection.
7. A method to update a global model, comprising:
generating a first modeling plan for client-side resources, the first modeling plan including (a) tokenized model parameters devoid of telemetry labels and (b) a transmission schedule for the client-side resources;
transmitting the first modeling plan to the client-side resources;
searching for a primary validation flag in response to retrieving client-side model parameters;
in response to detecting the primary validation flag, performing a secondary validation corresponding to the client-side model parameters using a server-side ground truth data set, the secondary validation based on calculating a confidence metric in view of the server-side ground truth data set;
authorize updating of the global model in response to results of the secondary validation satisfying a validation threshold;
in response to determining that the confidence metric does not satisfy a validation threshold prohibit updating the global model with the client-side model parameters;
sending a second modeling plan to the client-side resources in response to rejecting the client-side modeling parameters; and
causing the client-side resources to execute the second modeling plan instead of the first modeling plan.
8. The method as defined in claim 7 , wherein the transmission schedule includes at least one of a target epoch, a quantity of training dataset iterations, or a quantity of training samples to be processed by the client-side resources.
9. The method as defined in claim 7 , further including causing the client-side resources to at least one of identify computing device types or perform anomaly detection.