Sequential feature computation for power efficient classification

Patent metadata

Field	Value
Publication number	US-10133329-B2
Application number	US-201313841960-A
Country	US
Kind code	B2
Filing date	Mar 15, 2013
Priority date	Nov 19, 2012
Publication date	Nov 20, 2018
Grant date	Nov 20, 2018

Abstract

Disclosed is an apparatus and method for power efficient processor scheduling of features. In one embodiment, features may be scheduled for sequential computing, and each scheduled feature may receive a sensor data sample as input. In one embodiment, scheduling may be based at least in part on each respective feature's estimated power usage. In one embodiment, a first feature in the sequential schedule of features may be computed and before computing a second feature in the sequential schedule of features, a termination condition may be evaluated.

First claim

What is claimed is:

1. A method comprising:
receiving, by a processor, a sensor data sample from one or more sensors;
determining, by a processor, a classification associated with the sensor data sample based in part on the data sample and a type of each of the one or more sensors;
determining, by a processor, an estimated amount of power required by the processor to compute each of a plurality of features associated with the sensor data sample based in part on one or more prior feature calculations associated with the classification;
determining, by the processor, a sequential schedule for computing the plurality of features associated with the sensor data sample, wherein each of the plurality of features is computed based in part on data received by the processor from the one or more sensors, and wherein the schedule is determined based at least in part on the type of each of the one or more sensors and the estimated amount of power required to compute each of the plurality of features;
computing, by the processor, a first feature according to the sequential schedule; and
determining, by the processor, whether a termination condition is satisfied before computing a second feature in the sequential schedule.

2. The method of claim 1 , further comprising:
determining, by the processor, the computing of the first feature satisfies the termination condition; and
classifying, by the processor, the sensor data sample based on a result of the computing of the first feature.

3. The method of claim 1 , further comprising:
determining, by the processor, the computing of the first feature fails to satisfy the termination condition;
computing, by the processor, upon a failure of the first feature to satisfy the termination condition, the second feature in the sequential schedule for computing the plurality of features for computing;
determining, by the processor, the computing of the second feature satisfies the termination condition; and
classifying, by the processor, the sensor data sample according to results of computing the first and second features.

4. The method of claim 1 , wherein the sequential schedule for computing a plurality of features is further determined based on one or more of: a likelihood of each feature resulting in an unambiguous classification, or a relative frequency of occurrences of classes being classified from the sensor data sample.

5. The method of claim 1 , wherein determining the termination condition is satisfied comprises determining that a classification confidence meets a threshold value, wherein the classification confidence is based on one or more of: a similarity between a feature of the sensor data sample with a feature of a training data sample, or a difference in likelihood between a plurality of possible classifications.

6. The method of claim 1 , wherein the first feature is a mean or standard deviation of an accelerometer, wherein computing the mean or standard deviation results in a stationary classification, and wherein the method further comprises cancelling computation of subsequent motion related feature computation.

7. The method of claim 1 , wherein the first feature is a Mel-Frequency Cepstral Coefficient corresponding to audio energy from a microphone, wherein computing the Mel-Frequency Cepstral Coefficient results in a quiet classification, and wherein the method further comprises cancelling computation of subsequent audio related feature computation.

8. The method of claim 1 , wherein the sensor data sample is from one or more of: an accelerometer, a gyroscope, a magnetometer, a barometric pressure sensor, a temperature sensor, a global positioning sensor, a WiFi sensor, a Bluetooth sensor, an ambient light sensor, a camera, or a microphone.

9. A machine readable non-transitory storage medium containing executable instructions which cause a data processing device to perform a method comprising:
receiving, by a processor, a sensor data sample from one or more sensors;
determining, by a processor, a classification associated with the sensor data sample based in part on the data sample and a type of each of the one or more sensors;
determining, by a processor, an estimated amount of power required by the processor to compute each of a plurality of features associated with the sensor data sample based in part on one or more prior feature calculations associated with the classification;
determining, by the processor, a sequential schedule for computing the plurality of features associated with the sensor data sample, wherein each of the plurality of features is computed based in part on data received by the processor from the one or more sensors, and wherein the schedule is determined based at least in part on the type of each of the one or more sensors and the estimated amount of power required to compute each of the plurality of features;
computing, by the processor, a first feature according to the sequential schedule; and
determining, by the processor, whether a termination condition is satisfied before computing a second feature in the sequential schedule.

10. The machine readable non-transitory storage medium of claim 9 , wherein the method further comprises:
determining, by the processor, the computing of the first feature satisfies the termination condition; and
classifying, by the processor, the sensor data sample based on a result of the computing of the first feature.

11. The machine readable non-transitory storage medium of claim 9 , wherein the method further comprises:
determining, by the processor, the computing of the first feature fails to satisfy the termination condition;
computing, by the processor, upon a failure of the first feature to satisfy the termination condition, the second feature in the sequential schedule for computing the plurality of features for computing;
determining, by the processor, the computing of the second feature satisfies the termination condition; and
classifying, by the processor, the sensor data sample according to results of computing the first and second features.

12. The machine readable non-transitory storage medium of claim 9 , wherein the sequential schedule for computing a plurality of features is further determined based on one or more of: a likelihood of each feature resulting in an unambiguous classification, or a relative frequency of occurrences of classes being classified from the sensor data sample.

13. The machine readable non-transitory storage medium of claim 9 , wherein determining the termination condition is satisfied comprises determining that a classification confidence meets a threshold value, wherein the classification confidence is based on one or more of: a similarity between a feature of the sensor data sample with a feature of a training data sample, or a difference in likelihood between a plurality of possible classifications.

14. The machine readable non-transitory storage medium of claim 9 , wherein the first feature is a mean or standard deviation of an accelerometer, wherein computing the mean or standard deviation results in a stationary classification, and wherein the method further comprises cancelling computation of subsequent motion related feature computation.

15. The machine readable non-transitory storage medium of claim 9 , wherein the first feature is a Mel-Frequency Cepstral Coefficient corresponding to audio energy from a microphone, wherein computing the Mel-Frequency Cepstral Coefficient results in a quiet classification, and wherein the method further comprises cancelling computation of subsequent audio related feature computation.

16. Th

Assignees

• Qualcomm Inc

Inventors

• Grokop Leonard Henry
• Ahuja Disha
• Kulkarni Rashmi
• Sadasivam Shankar
• Teague Edward Harrison
• Nanda Sanjiv

Classifications

• Y02D10/24

  Cross-Sectional Technologies · mapped topic

• G06F9/4893 Primary

  taking into account power or heat criteria (power management in computers in general G06F1/3203; thermal management in computers in general G06F1/206) · CPC title

• G06F1/28 Primary

  Supervision thereof, e.g. detecting power-supply failure by out of limits supervision · CPC title

• Y02D10/00

  Energy efficient computing, e.g. low power processors, power management or thermal management · CPC title