Systems and methods for implementing an intelligent tuning of multiple hyperparameter criteria of a model constrained with metric thresholds

1. A method to tune hyperparameters of a machine learning model, the method comprising:

defining a joint tuning function that is based on a combination of a first objective function and a second objective function, the first objective function and the second objective function from a multi-criteria tuning request to tune the hyperparameters of the machine learning model;

identifying a Pareto efficient frontier curve based on the joint tuning function;

demarcating the Pareto efficient frontier curve, based on one or more thresholds, into a first section that is available to search and a second section that is unavailable to search; and

identifying, by configuring a machine, hyperparameter values from the first section of the Pareto efficient frontier curve based on the first section being available to search without causing the machine to search the second section that is demarcated as unavailable to search.

2. The method according to claim 1 , including applying the one or more thresholds to the Pareto efficient frontier curve by:

identifying a first plurality of values along the Pareto efficient frontier curve that do not satisfy the one or more thresholds;

identifying a second plurality of values along the Pareto efficient frontier curve that satisfy the one or more thresholds;

setting the first plurality of values as infeasible; and

setting the second plurality of values as feasible.

3. The method according to claim 2 , including searching the first section of the Pareto efficient frontier curve for one or more proposed hyperparameter values by:

bypassing hyperparameter values associated with ones of the first plurality of values; and

selecting a first proposed hyperparameter value based on an evaluation of hyperparameter values associated with the second plurality of values.

4. The method according to claim 1 , including applying the one or more thresholds to the Pareto efficient frontier curve by:

subjugating ones of a plurality of values along the Pareto efficient frontier curve that do not satisfy the one or more thresholds, wherein the subjugating causes hyperparameter values associated with the plurality of values to be invisible to search.

5. The method according to claim 4 , including searching the first section of the Pareto efficient frontier curve for one or more proposed hyperparameter values that are visible to search.

6. The method according to claim 1 , including applying the one or more thresholds to the Pareto efficient frontier curve by:

designating one or more sections of the Pareto efficient frontier curve having a plurality of values as valid if the plurality of values along the one or more sections satisfy the one or more thresholds;

isolating the one or more sections of the Pareto efficient frontier curve from one or more other sections of the Pareto efficient frontier curve having a plurality of distinct values that do not satisfy the one or more thresholds; and

extracting the one or more sections of the Pareto efficient frontier curve to perform the search based on the isolating.

7. The method according to claim 6 , including searching the first section of the Pareto efficient frontier curve for one or more proposed hyperparameter values by:

searching for the proposed hyperparameter values within the one or more sections of the Pareto efficient frontier curve based on the isolation and extraction.

8. The method according to claim 1 , including applying the one or more thresholds to the Pareto efficient frontier curve by:

rendering opaque to search a plurality of values along the Pareto efficient frontier curve that do not satisfy the one or more thresholds, and maintaining a transparent state to search ones of a plurality of values along the Pareto efficient frontier curve that satisfy the one or more thresholds.

9. The method according to claim 8 , including searching the first section of the Pareto efficient frontier curve for one or more proposed hyperparameter values by:

searching for the proposed hyperparameter values within the one or more sections of the Pareto efficient frontier curve based on the transparent state of ones of the plurality of values along the Pareto efficient frontier curve that satisfy the one or more thresholds.

10. The method according to claim 1 , including:

implementing an application programming interface that is in operable communication with a hyperparameter tuning service, wherein one or more parameters of the multi-criteria tuning request are defined via the application programming interface by:

defining the first objective function, defining the second objective function, defining the one or more thresholds to constrain a hyperparameter search space.

11. The method according to claim 1 , wherein the Pareto efficient frontier curve includes a curve having a convex shape; and including:

searching the first section of the Pareto efficient frontier curve for one or more proposed hyperparameter values by:

searching along the convex shape within the first section of the Pareto efficient frontier curve to identify the one or more proposed hyperparameter values.

12. The method according to claim 1 , wherein the Pareto efficient frontier curve includes a curve having a non-convex shape; and including:

searching the first section of the Pareto efficient frontier curve for one or more proposed hyperparameter values by:

searching along the non-convex shape within the first section of the Pareto efficient frontier curve to identify the one or more proposed hyperparameter values.

13. The method according to claim 1 , including executing a tuning operation by building a population of possible hyperparameter values within a hyperparameter search space based on the joint tuning function;

identifying an emergence of a general convex shape or a general non-convex shape of the Pareto efficient frontier curve; and

applying the one or more thresholds to the Pareto efficient frontier curve based on the emergence of the general convex shape or the general non-convex shape of the Pareto efficient frontier curve.

14. The method according to claim 1 , including executing a tuning operation by building a population of possible hyperparameter values within a hyperparameter search space based on the joint tuning function;

identifying an emerged convex shape or an emerged non-convex shape of the Pareto efficient frontier curve; and

applying the one or more thresholds to the Pareto efficient frontier curve based on the emerged convex shape or the emerged non-convex shape of the Pareto efficient frontier curve.

15. The method according to claim 1 , wherein the Pareto efficient frontier curve relates to a curve positioned along a plurality of hyperparameter parameter values that uniformly improves both the first objective function and the second objective function of the machine learning model.

16. A system to tune hyperparameters of a machine learning model, the system comprising:

interface circuitry;

instructions; and

at least one computer operable, based on the instructions, to:

define a joint tuning function that is based on a combination of a first objective function and a second objective function, the first objective function and the second objective function from a multi-criteria tuning request to tune the hyperparameters of the machine learning model;

identify a Pareto efficient frontier curve based on the joint tuning function;

demarcate the Pareto efficient frontier curve, based on one or more thresholds, into a first section that is available to search and a second section that is unavailable to search; and

identify hyperparameter values from the first section of the Pareto efficient frontier curve based on the first section being available to search without the at least one computer searching the second section that is demarcated as unavailable to search.

17. The system according to claim 16 , wherein the at least one computer is to apply the one or more thresholds to the Pareto efficient frontier curve by:

subjugating ones of a plurality of values along the Pareto efficient frontier curve that do not satisfy the one or more thresholds, wherein the subjugating causes hyperparameter values associated with the plurality of values to be invisible to search.

18. The system according to claim 16 , wherein the at least one computer is to apply the one or more thresholds to the Pareto efficient frontier curve by:

designating one or more sections of the Pareto efficient frontier curve having a plurality of values as valid if the plurality of values along the one or more sections satisfy the one or more thresholds;

isolating the one or more sections of the Pareto efficient frontier curve from one or more other sections of the Pareto efficient frontier curve having a plurality of distinct values that do not satisfy the one or more thresholds; and

extracting the one or more sections of the Pareto efficient frontier curve to perform a search of the first section based on the isolating.

19. The system according to claim 18 , wherein the at least one computer is to search the first section of the Pareto efficient frontier curve for one or more proposed hyperparameter values includes:

searching for the proposed hyperparameter values within the one or more sections of the Pareto efficient frontier curve based on the isolation and extraction.

20. The system according to claim 16 , wherein the first objective function and the second objective function generate metric values that do not improve uniformly.