Scalable-effort classifiers for energy-efficient machine learning

• US 9,916,540 B2
• Filed: 01/22/2015
• Issued: 03/13/2018
• Est. Priority Date: 01/22/2015
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Shoaib, Mohammed, Liu, Jie, Venkataramani, Swagath
• Primary Examiner(s): Chen, Alan
• Application Number:
  US14/603,222
• Publication Number:
  US 20160217390A1
• Time in Patent Office: 1,146 Days
• Field of Search: None
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 20/10 using kernel methods, e.g. support vector machines [SVM]

Abstract

Scalable-effort machine learning may automatically and dynamically adjust the amount of computational effort applied to input data based on the complexity of the input data. This is in contrast to fixed-effort machine learning, which uses a one-size-fits-all approach to applying a single classifier algorithm to both simple data and complex data. Scalable-effort machine learning involves, among other things, classifiers that may be arranged as a series of multiple classifier stages having increasing complexity (and accuracy). A first classifier stage may involve relatively simple machine learning models able to classify data that is relatively simple. Subsequent classifier stages have increasingly complex machine learning models and are able to classify more complex data. Scalable-effort machine learning includes algorithms that can differentiate among data based on complexity of the data.

Claims

1. A system comprising:

one or more processors; and

memory storing instructions that, when executed by the one or more processors, configure the one or more processors to perform operations comprising;

receiving an input value;

applying a first level of complexity of a machine learning model to the input value by applying at least one biased first level of complexity to the input value to generate at least one class label;

determining whether the first level of complexity is able to classify the input value; and

if the first level of complexity is not able to classify the input value, applying a second level of complexity of the machine learning model to the input value, wherein the second level of complexity is more complex than the first level of complexity.

2. The system of claim 1, wherein an amount of computing time used to classify the input value depends, at least in part, on the first level of complexity of the machine learning model.

3. The system of claim 1, the operations further comprising:

if the first level of complexity is able to classify the input value, classifying the input value into one of two or more categories.

4. The system of claim 1, wherein applying the first level of complexity of the machine learning model to the input value by applying at least one biased first level of complexity to the input value to generate at least one class label comprises:

applying a negatively biased first level of complexity to the input value to generate a first class label; and

applying a positively biased first level of complexity to the input value to generate a second class label.

5. The system of claim 4, wherein determining whether the first level of complexity is able to classify the input value comprises:

comparing the first class label to the second class label; and

determining whether a consensus exists between the negatively biased first level of complexity and the positively biased first level of complexity based, at least in part, on the comparing.

6. The system of claim 4, the operations further comprising:

adjusting the negatively biased first level of complexity and/or the positively biased first level of complexity to modify a likelihood that the first level of complexity is able to classify the input value with a predetermined confidence level.

7. The system of claim 1, wherein the input value is based, at least in part, on collected information from one or more of the following:

capturing an image, capturing an audio sample, or receiving a search query.

8. A computing device comprising:

an input port to receive an input value having a level of complexity;

a memory device storing a plurality of machine learning models, wherein abilities of the machine learning models to classify the input value are different from one another; and

a processor to apply one or more of the plurality of the machine learning models based, at least in part, on the level of complexity of the input value, wherein the processor is configured to apply at least one biased first level of complexity to the input value to generate at least one class label.

9. The computing device of claim 8, wherein the abilities of the machine learning models to classify the input value comprise:

the abilities of the machine learning models to classify the input value into one of two or more categories.

10. The computing device of claim 8, wherein the configuration of the processor to apply at least one biased first level of complexity to the input value to generate at least one class label includes the processor being configured to:

apply a negatively biased level of complexity to the input value to generate a first class label; and

apply a positively biased level of complexity to the input value to generate a second class label.

11. The computing device of claim 10, wherein the processor is configured to:

compare the first class label to the second class label; and

determine whether a consensus exists between the negatively biased level of complexity and the positively biased level of complexity based, at least in part, on the comparing.

12. The computing device of claim 10, wherein the processor is configured to:

adjust the negatively biased level of complexity and/or the positively biased level of complexity to modify a likelihood that the level of complexity is able to classify the input value.

13. The computing device of claim 8, wherein the processor is configured to apply the plurality of the machine learning models on the input value sequentially in order of increasing ability of the machine learning models to classify the input value.

14. The computing device of claim 8, wherein a computing cost of classifying the input value is proportional to the level of complexity of the input value.

15. One or more computer-readable storage media of a client device storing computer-executable instructions that, when executed by one or more processors of the client device, configure the one or more processors to perform operations comprising:

receiving an input value;

applying a first level of complexity of a machine learning model to the input value by applying at least one biased first level of complexity to the input value to generate at least one class label;

determining whether the first level of complexity is able to classify the input value; and

if the first level of complexity is not able to classify the input value, applying a second level of complexity of the machine learning model to the input value, wherein the second level of complexity is more complex than the first level of complexity.

16. The one or more computer-readable storage media of claim 15, the operations further comprising:

if the first level of complexity is able to classify the input value, classifying the input value into one of two or more categories.

17. The one or more computer-readable storage media of claim 15, wherein applying the first level of complexity of the machine learning model to the input value by applying at least one biased first level of complexity to the input value to generate at least one class label comprises:

applying a negatively biased first level of complexity to the input value to generate a first class label; and

applying a positively biased first level of complexity to the input value to generate a second class label.

18. The one or more computer-readable storage media of claim 17, the operations further comprising:

adjusting the negatively biased first level of complexity and/or the positively biased first level of complexity to modify a likelihood that the first level of complexity is able to classify the input value.

19. The one or more computer-readable storage media of claim 17, wherein determining whether the first level of complexity is able to classify the input value comprises:

comparing the first class label to the second class label; and

determining whether a consensus exists between the negatively biased first level of complexity and the positively biased first level of complexity based, at least in part, on the comparing.

20. The one or more computer-readable storage media of claim 19, wherein determining whether the first level of complexity is able to classify the input value comprises:

calculating a confidence value based, at least in part, on the comparing.