US 12,591,794 B2 · Joint prediction and improvement for machine learning models

Abstract

Methods, systems, and computer program products for a joint prediction and improvement framework for machine learning models are provided herein. A method includes obtaining a machine learning model initialized with a set of parameters; identifying one or more actions based on test inputs corresponding to the machine learning model and historical actions related to a task, where the historical actions are dependent on respective historical outputs of the machine learning model; using the identified one or more actions to jointly compute: one or more first values corresponding to inference loss for the machine learning model; and one or more second values based at least in part on a computing cost function associated with the task; and updating the set of parameters of the machine learning model based on the one or more first values and the one or more second values.

Claims20

Dependent claims

A system comprising:

a memory configured to store program instructions;
a processor operatively coupled to the memory to execute the program instructions to:
obtain a machine learning model initialized with a set of parameters;
identify one or more actions based at least in part on a plurality of test inputs corresponding to the machine learning model and a plurality of historical actions related to a downstream task, wherein the plurality of historical actions is dependent on a respective plurality of historical outputs of the machine learning model;
use the identified one or more actions to jointly compute:
one or more first values corresponding to inference loss for the machine learning model that is adjusted based on the identified one or more actions;
and one or more second values based at least in part on a computing cost function associated with the downstream task;
and
update the set of parameters of the machine learning model based at least in part on the one or more first values and the one or more second values.

6 more elements

A computer program product comprising a non-transitory computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computing device to cause the computing device to:

obtain a machine learning model initialized with a set of parameters;
identify one or more actions based at least in part on a plurality of test inputs corresponding to the machine learning model and a plurality of historical actions related to a downstream task, wherein the plurality of historical actions is dependent on a respective plurality of historical outputs of the machine learning model;
use the identified one or more actions to jointly compute:
one or more first values corresponding to inference loss for the machine learning model that is adjusted based on the identified one or more actions;
and one or more second values based at least in part on a computing cost function associated with the downstream task;
and
update the set of parameters of the machine learning model based at least in part on the one or more first values and the one or more second values.

4 more elements

Dependencies are inferred from the claim text.

Joint prediction and improvement for machine learning models

Life of this patent

Joint prediction and improvement for machine learning models

Life of this patent

Abstract

Claims20