US 12,112,249 B2Granted

Multi-objective automated machine learning

AssigneeInternational Business Machines Corporation

InventorsVaibhav Saxena, Aswin Kannan, Saurabh Manish Raje, Parikshit Ram, Yogish Sabharwal, Ashish Verma

Life of this patent

Estimating the term…

Loading patent…

Abstract

A system, computer program product, and method are presented for performing multi-objective automated machine learning, and, more specifically, to identifying a plurality of machine learning pipelines as Pareto-optimal solutions to optimize a plurality of objectives. The method includes receiving input data directed toward one or more subjects of interest and determining a plurality of objectives to be optimized. The method also includes ingesting at least a portion of the input data through one or more machine learning (ML) models. The method further includes aggregating the plurality of objectives into one or more aggregated single objectives. The method also includes determining a plurality of Pareto-optimal solutions, thereby defining a plurality of ML pipelines that optimize the one or more aggregated single objectives. The method further includes selecting one ML pipeline from the plurality of ML pipelines.

Claims23

Dependent claims

A computer system comprising:

one or more processing devices and at least one memory device operably coupled to the one or more processing devices, the one or more processing devices are configured to:
receive input data directed toward one or more subjects of interest;
determine a plurality of objectives to be optimized;
ingest at least a portion of the input data through one or more machine learning (ML) models;
apply a weight to each of the plurality of objectives, thereby to generate a plurality of weighted objectives, thereby to generate a plurality of weighted aggregated single objectives;
determine a plurality of Pareto optimal solutions, thereby defining a plurality of ML pipelines that optimize the plurality of weighted aggregated single objectives, wherein the Pareto optimal solutions are utilized to graphically define a Pareto-front curve which is displayed to a user in a graphical user interface;
identify at least a portion of the Pareto-front curve to be refined based on a plurality of additional constraints identified by the user by selecting at least a portion of the Pareto-front curve in the graphical user interface for at least one or more of the plurality of objectives to be optimized;
determine whether the plurality of additional constraints identified are supported or not supported for generating a plurality of additional single objective optimizations without modifying the plurality of weighted aggregated single objectives;
determine a plurality of additional Pareto-optimal solutions, wherein the Pareto-front curve is refined based on the plurality of additional Pareto-optimal solutions;
and
select one ML pipeline from the plurality of ML pipelines.

8 more elements

A computer program product, comprising:

one or more computer readable storage media;
and
program instructions collectively stored on the one or more computer storage media, the program instructions comprising:
program instructions to receive input data directed toward one or more subjects of interest;
program instructions to determine a plurality of objectives to be optimized;
program instructions to apply a weight to each of the plurality of objectives, thereby to generate a plurality of weighted objectives, thereby to generate a plurality of weighted aggregated single objectives;
program instructions to determine a plurality of Pareto optimal solutions, thereby defining a plurality of ML pipelines that optimize the plurality of weighted aggregated single objectives, wherein the Pareto optimal solutions are utilized to graphically define a Pareto-front curve which is displayed to a user in a graphical user interface;
program instructions to identify at least a portion of the Pareto-front curve to be refined based on a plurality of additional constraints identified by the user by selecting at least a portion of the Pareto-front curve in the graphical user interface for at least one or more of the plurality of objectives to be optimized;
program instructions to determine whether the plurality of additional constraints identified are supported or not supported for generating a plurality of additional single objective optimizations without modifying the plurality of weighted aggregated single objectives;
and
program instructions to determine whether the plurality of additional constraints identified are supported or not supported for generating a plurality of additional single objective optimizations without modifying the plurality of weighted aggregated single objectives;
program instructions to determine a plurality of additional Pareto-optimal solutions, wherein the Pareto-front curve is refined based on the plurality of additional Pareto-optimal solutions;
and
program instructions to select one ML pipeline from the plurality of ML pipelines.

11 more elements

A computer-implemented method comprising:

receiving input data directed toward one or more subjects of interest;
determining a plurality of objectives to be optimized;
ingesting at least a portion of the input data through one or more machine learning (ML) models;
applying a weight to each of the plurality of objectives, thereby to generate a plurality of weighted objectives, thereby to generate a plurality of weighted aggregated single objectives;
determining a plurality of Pareto optimal solutions, thereby defining a plurality of ML pipelines that optimize the plurality of weighted aggregated single objectives, wherein the Pareto optimal solutions are utilized to graphically define a Pareto-front curve which is displayed to a user in a graphical user interface;
identifying at least a portion of the Pareto-front curve to be refined based on a plurality of additional constraints identified by the user by selecting at least a portion of the Pareto-front curve in the graphical user interface for at least one or more of the plurality of objectives to be optimized;
determining whether the plurality of additional constraints identified are supported or not supported for generating a plurality of additional single objective optimizations without modifying the plurality of weighted aggregated single objectives;
determining a plurality of additional Pareto-optimal solutions, wherein the Pareto-front curve is refined based on the plurality of additional Pareto-optimal solutions;
and
selecting one ML pipeline from the plurality of ML pipelines.

7 more elements

A computer-implemented method comprising:

receiving input data directed toward one or more subjects of interest;
determining a plurality of objectives to be optimized;
ingesting at least a portion of the input data through one or more machine learning (ML) models;
applying a first weight to a first objective of the plurality of objectives, and applying one or more second weights to one or more second respective objectives of the plurality of objectives, the first weight and the one or more second weights are selected through a uniform weight generation scheme, wherein a sum of the first weight and the one or more second weights is unity, thereby generating a first weighted objective and one or more second weighted objectives;
aggregating the first weighted objective and the one or more second weighted objectives into a first weighted aggregated single objective;
determining a first Pareto-optimal solution from the first weighted aggregated single objective;
defining a first ML pipeline through the first Pareto-optimal solution, wherein the first ML pipeline optimizes the first weighted aggregated single objective;
applying a third weight to the first objective of the plurality of objectives, and applying one or more fourth weights to the one or more second respective objectives of the plurality of objectives, the third weight and the one or more fourth weights are user-selected, wherein a sum of the third weight and the one or more fourth weights is unity, and wherein the first weight is different from the third weight and the one or more second weights are different from the one or more fourth weights, thereby generating a third weighted objective and one or more fourth weighted objectives;
aggregating the third weighted objective and the one or more fourth weighted objectives into a second weighted aggregated single objective;
determining a second Pareto-optimal solution from the second weighted aggregated single objective;
defining a second ML pipeline through the second Pareto-optimal solution, wherein the second ML pipeline optimizes the second weighted aggregated single objective;
utilizing the first Pareto-optimal solution and the second Pareto-optimal solution to graphically define a Pareto-front curve which is displayed to a user in a graphical user interface;
identify at least a portion of the Pareto-front curve to be refined based on a plurality of additional constraints identified by the user by selecting at least a portion of the Pareto-front curve in the graphical user interface for at least one or more of the plurality of objectives to be optimized;
and
selecting one ML pipeline from the first and second ML pipelines.

12 more elements

A computer-implemented method comprising:

receiving input data directed toward one or more subjects of interest;
determining a plurality of objectives to be optimized;
ingesting at least a portion of the input data through one or more machine learning (ML) models;
applying a first weight to a first objective of the plurality of objectives, and applying one or more second weights to one or more second respective objectives of the plurality of objectives, the first weight and the one or more second weights are automatically determined, wherein a sum of the first weight and the one or more second weights is unity, thereby generating a first weighted objective and one or more second weighted objectives;
aggregating the first weighted objective and the one or more second weighted objectives into a first weighted aggregated single objective;
determining a first Pareto-optimal solution from the first weighted aggregated single objective, the first Pareto-optimal solution including one or more first objective values;
defining a first ML pipeline through the first Pareto-optimal solution, wherein the first ML pipeline optimizes the first weighted aggregated single objective;
applying a third weight to the first objective of the plurality of objectives, and applying one or more fourth weights to the one or more second respective objectives of the plurality of objectives, the third weight and the one or more fourth weights are automatically determined at least partially based on the one or more first objective values, wherein a sum of the third weight and the one or more fourth weights is unity, and wherein the first weight is different from the third weight and the one or more second weights are different from the one or more fourth weights, thereby generating a third weighted objective and one or more fourth weighted objectives;
aggregating the third weighted objective and the one or more fourth weighted objectives into a second weighted aggregated single objective;
determining a second Pareto-optimal solution from the second weighted aggregated single objective;
defining a second ML pipeline through the second Pareto-optimal solution, wherein the second ML pipeline optimizes the second weighted aggregated single objective;
utilizing the first Pareto-optimal solution and the second Pareto-optimal solution to graphically define a Pareto-front curve which is displayed to a user in a graphical user interface;
identify at least a portion of the Pareto-front curve to be refined based on a plurality of additional constraints identified by the user by selecting at least a portion of the Pareto-front curve in the graphical user interface for at least one or more of the plurality of objectives to be optimized;
and
selecting one ML pipeline from the first and second ML pipelines.

12 more elements

Dependencies are inferred from the claim text.