Systems and methods for distributed training of deep learning models

1. A local device to train deep learning models, the local device comprising:

memory;

instructions;

a programmable circuitry to execute the instructions to implement:

reference generator circuitry to label input data received at the local device to generate training data;

trainer circuitry to train a local deep learning model and to transmit the local deep learning model to a server that is to receive a plurality of local deep learning models from a plurality of local devices, the server to determine a set of weights for a global deep learning model;

sample controller circuitry to sample an output of the local deep learning model; and

updater circuitry to update the local deep learning model based on the set of weights received from the server.

2. A local device as defined in claim 1 , further including a data receiver to receive the input data directly at the local device.

3. A local device as defined in claim 2 , wherein the input data does not pass through the server.

4. A local device as defined in claim 2 , wherein the local device includes a sensor to collect the input data.

5. A local device as defined in claim 2 , wherein the data receiver is communicatively coupled to a sensor that collects the input data and transmits the input data to the local device.

6. A local device as defined in claim 1 , wherein the local device does not transmit the input data to the server.

7. A local device as defined in claim 1 , wherein the set of weights are aggregated weights based on the plurality of local deep learning models from the plurality of local devices.

8. A local device as defined in claim 1 , wherein the sample controller is to sample the input data.

9. A local device as defined in claim 8 , wherein the sample controller is to sample the input data by selecting a pseudo-random portion of the input data.

10. A local device as defined in claim 8 , wherein the sample controller is to sample the input data by down sampling the input data to reduce a data size of the input data.

11. A local device as defined in claim 1 , wherein the trainer is further to determine a difference between a label determined by the labelling and the output of the local deep learning model.

12. A local device as defined in claim 11 , wherein the sample controller is to sample the output of the local deep learning model prior to the trainer determining the difference.

13. A local device as defined in claim 11 , wherein the trainer is further to adjust the local deep learning model based on the difference.

14. A non-transitory computer readable medium comprising instructions that, when executed, cause a local device to at least:

label input data received at the local device to generate training data;

train a local deep learning model;

transmit the local deep learning model to a server, the server to receive a plurality of local deep learning models from a plurality of local devices, the server to determine a set of weights for a global deep learning model;

sample an output of the local deep learning model; and

update the local deep learning model based on the set of weights received from the server.

15. A non-transitory computer readable medium as defined in claim 14 , wherein the input data is received directly at the local device.

16. A non-transitory computer readable medium as defined in claim 15 , wherein the input data does not pass through the server.

17. A non-transitory computer readable medium as defined in claim 15 , wherein the local device includes a sensor to collect the input data.

18. A non-transitory computer readable medium as defined in claim 15 , wherein the local device is communicatively coupled to a sensor that collects the input data and transmits the input data to the local device.

19. A non-transitory computer readable medium as defined in claim 14 , wherein the input data is not transmitted to the server.

20. A non-transitory computer readable medium as defined in claim 14 , wherein the set of weights are aggregated weights based on the plurality of local deep learning models from the plurality of local devices.

21. A non-transitory computer readable medium as defined in claim 14 , wherein the instructions, when executed, cause the local device to sample the input data.

22. A non-transitory computer readable medium as defined in claim 21 , wherein the instructions, when executed, cause the local device to sample the input data by selecting a pseudo-random portion of the input data.

23. A non-transitory computer readable medium as defined in claim 21 , wherein the instructions, when executed, cause the local device to sample the input data by down sampling the input data to reduce a data size of the input data.

24. A method to train deep learning models, the method comprising:

labelling, by executing an instruction with at least one processor at a local device, input data received at the local device to generate training data;

training, by executing an instruction with the at least one processor, a local deep learning model;

transmitting the local deep learning model to a server, the server to receive a plurality of local deep learning models from a plurality of local devices, the server to determine a set of weights for a global deep learning model;

sampling an output of the local deep learning model; and

updating, by executing an instruction with the at least one processor at the local device, the local deep learning model based on the set of weights received from the server.

25. A method as defined in claim 24 , wherein the input data is received directly at the local device.