Reducing computation complexity and increasing power efficiency in multi-variant inference models

Patent metadata

Field	Value
Publication number	US-2025307687-A1
Application number	US-202418616906-A
Country	US
Kind code	A1
Filing date	Mar 26, 2024
Priority date	Mar 26, 2024
Publication date	Oct 2, 2025
Grant date	—

Abstract

An information handling system may define a first grouping of inputs to a first inference model, determine a first number of inference stages for the first inference model, and calculate an accuracy of an output of the first inference model. When the accuracy is within a threshold accuracy, the system may load the first inference model to multiple computing devices.

First claim

What is claimed is:

1 . An information handling system, comprising:
a memory device to store code; and a processor configured to execute code to:
define a first grouping of inputs to a first inference model;
determine a first number of inference stages for the first inference model;
calculate an accuracy of an output of the first inference model;
compare the accuracy with a threshold accuracy; and
when the accuracy is within the threshold accuracy, load the first inference model to a plurality of computing devices.

2 . The information handling system of claim 1 , wherein, when the accuracy is not within the threshold accuracy the processor is further configured to:
define a second grouping of the inputs to a second inference model; determine a second number of inference stages for the second inference model; calculate the accuracy of an output of the second inference model; compare the accuracy with the threshold accuracy; and when the accuracy is within the threshold accuracy, load the second inference model to the plurality of computing devices.

3 . The information handling system of claim 1 , wherein in defining the first grouping, the processor is further configured to determine that the inputs in each of a sub-group of the first grouping are related inputs.

4 . The information handling system of claim 3 , wherein the related inputs include at least one of related application variable inputs, related hardware parameter inputs, and power range inputs.

5 . The information handling system of claim 1 , wherein in defining the first grouping, the processor is further configured to apply at least one of a Bayesian analysis, a conditional analysis, an absolute probability analysis, and a contingency grouping analysis to the inputs.

6 . The information handling system of claim 1 , wherein determining the first number of inference stages is based on the first grouping.

7 . The information handling system of claim 1 , wherein the first number of inference stages is at least two inference stages.

8 . The information handling system of claim 1 , wherein the first number of inference stages is not more than three inference stages.

9 . The information handling system of claim 1 , wherein each of the first number of inference stages applies an artificial intelligence/machine learning (AI/ML) model.

10 . The information handling system of claim 8 , wherein the AI/ML model includes at least one of a regression model, a decision tree model, a support vector means model, a Naïve Bayes model, a K-nearest neighbors model, a K-means model, a random forest model, a dimensional reduction model, and a gradient boosting model.

11 . A method, comprising:
defining, by a processor, a first grouping of inputs to a first inference model; determining a first number of inference stages for the first inference model; calculating an accuracy of an output of the first inference model; comparing the accuracy with a threshold accuracy; and when the accuracy is within the threshold accuracy, loading the first inference model to a plurality of computing devices.

12 . The method of claim 11 , wherein, when the accuracy is not within the threshold accuracy the method further comprises:
defining a second grouping of the inputs to a second inference model; determining a second number of inference stages for the second inference model; calculating the accuracy of an output of the second inference model; comparing the accuracy with the threshold accuracy; and when the accuracy is within the threshold accuracy, loading the second inference model to the plurality of computing devices.

13 . The method of claim 11 , wherein in defining the first grouping, the method further comprises determining that the inputs in each of a sub-group of the first grouping are related inputs.

14 . The method of claim 13 , wherein the related inputs include at least one of related application variable inputs, related hardware parameter inputs, and power range inputs.

15 . The method of claim 11 , wherein in defining the first grouping, the method further comprises applying at least one of a Bayesian analysis, a conditional analysis, an absolute probability analysis, and a contingency grouping analysis to the inputs.

16 . The method of claim 11 , wherein determining the first number of inference stages is based on the first grouping.

17 . The method of claim 11 , wherein the first number of inference stages is at least two inference stages.

18 . The method of claim 11 , wherein the first number of inference stages is not more than three inference stages.

19 . The method of claim 11 , wherein each of the first number of inference stages applies an artificial intelligence/machine learning (AI/ML) model, including at least one of a regression model, a decision tree model, a support vector means model, a Naïve Bayes model, a K-nearest neighbors model, a K-means model, a random forest model, a dimensional reduction model, and a gradient boosting model.

20 . An information handling system, comprising:
a memory device to store code; and a processor configured to execute code to:
define a grouping of related inputs to an inference model, wherein the related inputs include at least one of related application variable inputs, related hardware parameter inputs, and power range inputs;
determine a number of inference stages for the inference model based on the grouping of inputs;
calculate an accuracy of an output of the inference model;
compare the accuracy with a threshold accuracy; and
when the accuracy is within the threshold accuracy, load the inference model to a plurality of computing devices.

Assignees

• Dell Products Lp

Inventors

• Khosrowpour Farzad
• Vichare Nikhil

Classifications

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06N20/20

  Ensemble learning · CPC title

• G06N20/00 Primary

  Machine learning · CPC title