System and method of improving compression of predictive models

1. A computer-implemented method for improving compression of predictive models, the computer-implemented method comprising:

generating an unlabeled simulated data set by expanding an initial data set, wherein the initial data set includes a first plurality of fact sets and wherein the unlabeled simulated data set includes a second plurality of fact sets;

generating a labeled data set, at least by predicting the unlabeled simulated data set using a complex model to output a plurality of labels, wherein the labeled data set includes the second plurality of fact sets and the plurality of labels, and wherein each fact set of the second plurality of fact sets corresponds to a respective one of the plurality of labels; and

training, using the labeled data set, a neural network model associated with a plurality of training parameters, wherein training the neural network model includes:

(i) generating a plurality of intermediate predictions, at least by predicting the second plurality of fact sets using the neural network model,

(ii) comparing the plurality of labels to the plurality of intermediate predictions to produce a measure of accuracy, and

(iii) modifying, based on the measure of accuracy, at least one of the plurality of training parameters of the neural network model.

2. The computer-implemented method of claim 1 wherein (i), (ii), and (iii) are iteratively repeated until the measure of accuracy is within a predetermined threshold.

3. The computer-implemented method of claim 1 , wherein the first plurality of fact sets and the second plurality of fact sets both include a plurality of fact types, and wherein generating the unlabeled simulated data set by expanding the initial data set includes generating the unlabeled simulated data set such that a distribution of the plurality of fact types within the second plurality of fact sets is skewed as compared to a distribution of the plurality of fact types within the first plurality of fact sets.

4. The computer-implemented method of claim 1 , further comprising:

generating a graphical depiction of the neural network model, as trained.

5. The computer-implemented method of claim 1 , further comprising:

accessing a remote electronic database; and

obtaining, from the remote electronic database, a definition of the plurality of training parameters of the neural network model.

6. The computer-implemented method of claim 1 , wherein generating the plurality of intermediate predictions includes:

dividing the second plurality of fact sets into fact subsets;

receiving, at each of a plurality of networked computing devices, one of the fact subsets;

generating, by each of the plurality of networked computing devices predicting a respective one of the fact subsets using the neural network model, a respective intermediate prediction; and

receiving, at a single networked computing device, the respective intermediate prediction corresponding to each of the fact subsets.

7. The computer-implemented method of claim 1 , further comprising:

receiving, in a computing device, the neural network model, as trained;

generating an unlabeled new data set based upon data collected by the computing device, wherein the unlabeled new data set includes a plurality of new fact sets; and

generating a plurality of device predictions, at least by predicting the unlabeled new data set using the neural network model, as trained.

8. The computer-implemented method of claim 1 , further comprising:

sending, to a remote computing device, the neural network model, as trained, to enable the remote computing device to analyze one or more unlabeled new data sets using the neural network model, as trained.

9. The computer-implemented method of claim 8 , wherein sending, to the remote computing device, the neural network model, as trained, comprises sending the neural network model, as trained, to a mobile computing device of a user.

10. A computing system comprising:

one or more processors; and

one or more memories storing instructions that, when executed by the one or more processors, cause the computing system to:

generate an unlabeled simulated data set by expanding an initial data set, wherein the initial data set includes a first plurality of fact sets and wherein the unlabeled simulated data set includes a second plurality of fact sets,

generate a labeled data set, at least by predicting the unlabeled simulated data set using a complex model to output a plurality of labels, wherein the labeled data set includes the second plurality of fact sets and the plurality of labels, and wherein each fact set of the second plurality of fact sets corresponds to a respective one of the plurality of labels, and

train, using the labeled data set, a neural network model associated with a plurality of training parameters, wherein training the neural network model includes:

(i) generating a plurality of intermediate predictions, at least by predicting the second plurality of fact sets using the neural network model,

(ii) comparing the plurality of labels to the plurality of intermediate predictions to produce a measure of accuracy, and

(iii) modifying, based on the measure of accuracy, at least one of the plurality of training parameters of the neural network model.

11. The computing system of claim 10 wherein the instructions further cause (i), (ii), and (iii) to iteratively repeat until the measure of accuracy is within a predetermined threshold.

12. The computing system of claim 10 , wherein the first plurality of fact sets and the second plurality of fact sets both include a plurality of fact types, and wherein the instructions cause the computing system to generate the unlabeled simulated data set such that a distribution of the plurality of fact types within the second plurality of fact sets is skewed as compared to a distribution of the plurality of fact types within the first plurality of fact sets.

13. The computing system claim 10 , wherein the instructions further cause the computing system to generate a graphical depiction of the neural network model, as trained.

14. The computing system of claim 10 , further comprising:

a remote electronic database,

wherein the instructions further cause the computing system to

access the remote electronic database, and

obtain, from the remote electronic database, a definition of the plurality of training parameters of the neural network model.

15. The computing system of claim 10 , wherein the instructions further cause the computing system to:

send, to a remote computing device, the neural network model, as trained, to enable the remote computing device to analyze one or more unlabeled new data sets using the neural network model, as trained.

16. The computing system of claim 15 , wherein the remote computing device is a mobile computing device of a user.

17. A non-transitory computer readable medium containing program instructions that when executed, cause a computer to:

generate an unlabeled simulated data set by expanding an initial data set, wherein the initial data set includes a first plurality of fact sets and wherein the unlabeled simulated data set includes a second plurality of fact sets;

generate a labeled data set, at least by predicting the unlabeled simulated data set using a complex model to output a plurality of labels, wherein the labeled data set includes the second plurality of fact sets and the plurality of labels, and wherein each fact set of the second plurality of fact sets corresponds to a respective one of the plurality of labels; and

train, using the labeled data set, a neural network model associated with a plurality of training parameters, at least in part by:

(i) generating a plurality of intermediate predictions, at least by predicting the second plurality of fact sets using the neural network model;

(ii) comparing the plurality of labels to the plurality of intermediate predictions to produce a measure of accuracy; and

(iii) modifying, based on the measure of accuracy, at least one of the plurality of training parameters of the neural network model.

18. The non-transitory computer readable medium of claim 17 containing further program instructions that when executed, cause the computer to:

iteratively repeat (i), (ii), and (iii) until the measure of accuracy is within a predetermined threshold.

19. The non-transitory computer readable medium of claim 17 , wherein the first plurality of fact sets and the second plurality of fact sets both include a plurality of fact types, and wherein the program instructions, when executed, cause the computer to generate the unlabeled simulated data set such that a distribution of the plurality of fact types within the second plurality of fact sets is skewed as compared to a distribution of the plurality of fact types within the first plurality of fact sets.

20. The non-transitory computer readable medium of claim 17 , further comprising program instructions that, when executed, cause the computer to:

generate a graphical depiction of the neural network model, as trained.