Back to results
US 2024/0256984 A1Published application
Efficient real time serving of ensemble models
AssigneeIntuit Inc.
InventorsAviv BEN ARIE, Omer ZALMANSON
Life of this patent
Estimating the term…
Abstract
A method implements efficient real time serving of ensemble models. The method includes receiving an input and processing the input with an abridged model to generate a set of component scores and an abridged score. The method further includes processing the set of component scores with a deviation threshold to select one of the abridged score and an ensemble score as an output and presenting the output.
Claims20
A method, comprising:
- training an ensemble model using a set of training data to generate a trained ensemble model, wherein the trained ensemble model comprises a plurality of component models;
- processing, by the trained ensemble model, a set of non-training data to generate a first set of output and a plurality of component scores for the plurality of component models:
- generating an abridged model by selecting a set of component models from the plurality of component models according to a deviation between a first set of component scores of the set of component models and an accuracy of the set of component models as compared to the first set of output, wherein the first set of component scores is in the plurality of component scores:
- receiving an input;
- processing the input by the abridged model to generate a second set of component scores for the set of component models and an abridged score for the abridged model,
- wherein the abridged score is an aggregation of the set of component scores;
- processing the second set of component scores with a deviation threshold to select one of the abridged score and an ensemble score of the ensemble model as an output;
- and
- presenting the output.
6 more elements
A system comprising:
- an output controller configured to generate an output;
- an application executing on one or more servers and configured for:
- training an ensemble model using a set of training data to generate a trained ensemble model, wherein the trained ensemble model comprises a plurality of component models:
- processing, by the trained ensemble model, a set of non-training data to generate a first set of output and a plurality of component scores for the plurality of component models;
- generating an abridged model by selecting a set of component models from the plurality of component models according to a deviation between a first set of component scores of the set of component models and an accuracy of the set of component models as compared to the first set of output, wherein the first set of component scores is in the plurality of component scores;
- receiving an input;
- processing the input by the abridged model to generate a second set of component scores for the set of component models and an abridged score for the abridged model, and wherein the abridged score is an aggregation of the set of component scores;
- processing, by the output controller, the second set of component scores with a deviation threshold to select one of the abridged score and an ensemble score of the ensemble model as an output;
- and
- presenting the output.
7 more elements
A non-transitory computer program product comprising computer-readable program code that, when executed by a computer processor of a computing system, causes the computing system to perform operations comprising:
- training an ensemble model using a set of training data to generate a trained ensemble model, wherein the trained ensemble model comprises a plurality of component models:
- processing, by the trained ensemble model, a set of non-training data to generate a first set of output and a plurality of component scores for the plurality of component models:
- generating an abridged model by selecting a set of component models from the plurality of component models according to a deviation between a first set of component scores of the set of component models and an accuracy of the set of component models as compared to the first set of output, wherein the first set of component scores is in the plurality of component scores:
- receiving an input;
- processing the input by the abridged model to generate a second set of component scores for the set of component models and an abridged score for the abridged model, wherein the abridged score is an aggregation of the set of component scores;
- processing the second set of component scores with a deviation threshold to select one of the abridged score and an ensemble score of the ensemble model as an output;
- and
- presenting the output.
5 more elements
Dependencies are inferred from the claim text.