Value of information with streaming evidence based on a prediction of a future belief at a future time

• US 9,836,700 B2
• Filed: 03/15/2013
• Issued: 12/05/2017
• Est. Priority Date: 03/15/2013
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Bohus, Dan, Horvitz, Eric J., Pomerantz, Stephanie Rosenthal, Eden, Semiha E. Kamar
• Primary Examiner(s): Chang, Li-Wu
• Application Number:
  US13/831,688
• Publication Number:
  US 20140279719A1
• Time in Patent Office: 1,726 Days
• Field of Search: None
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 5/02 Knowledge representation; Symbolic representation

Abstract

The subject disclosure is directed towards processing evidence, which may include high-dimensional streaming evidence, into a future belief state. The existing evidence is used to project a belief about a future state. The future belief state may be used to determine whether to wait for additional evidence, or to act now without waiting for additional evidence, e.g., based on a cost of the delay. For example, an autonomous assistant may decide based upon the belief whether to engage a person or not, or to wait for more information before the engagement decision is made.

Claims

1. A method for executing an assistant on a computing device to determine whether or not to act, the method comprising:

receiving, by the assistant, streaming evidence from one or more sensors;

constructing, by the assistant, one or more belief projection models to predict a future belief over a state at a future time, the one or more belief projection models using existing evidence that is collected until a defined period of time to determine a probability of a future state that is based on evidence to be collected between the defined period of time and the future time, the existing evidence comprising the streaming evidence and other evidence obtained prior to the streaming evidence;

based on the predicted future belief and the one or more belief projection models, determining, by the assistant device, a likelihood of obtaining particular evidence between the defined period of time and the future time to determine whether to act at the defined period of time for a next action or wait for the future time to act for the next action; and

based on determining to act at the defined period of time or wait for the future time to act for the next action, transmitting, by the assistant device, data using an output device at the defined period of time without obtaining the particular evidence or upon receipt of the particular evidence at the future time.

2. The method of claim of claim 1 wherein determining the timing of the next action comprises computing a cost of taking the next action at the future time.

3. The method of claim of claim 1 wherein constructing the one or more belief inference projection models comprises processing at least part of the existing evidence into percepts.

4. The method of claim of claim 1 wherein constructing the one or more belief inference projection models comprises projecting models for at least one of:

on-calendar, activity or F-formation related percepts.

5. The method of claim of claim 1 further comprising based on a determination to act without obtaining the particular evidence, initiating engagement with a person by using the output device.

6. The method of claim 1 further comprising associating a cost with activating the at least one other sensor.

7. The method of claim 1 wherein activating the at least one other sensor comprises seeking a response from one or more humans.

8. The method of claim of claim 1 further comprising one or more of the following:

a) learning the belief projection model automatically from data via one or more parametric machine learning approaches, b) learning the belief projection model automatically from data via non-parametric machine learning approaches, and c) constructing the belief projection model manually via a set of heuristic rules.

9. The method of claim of claim 1 further comprising automatically collecting training data while receiving the streaming evidence.

10. A system for executing an assistant on a computing device to determine whether or not to act, the system comprising:

a processor;

processing logic;

a sensor set comprising one or more sensors; and

an output mechanism set comprising one or more output devices;

the processing logic, when executed by the processor, configured to;

process evidence received via the sensor set, including streaming the evidence into one or more belief projection models, the one or more belief projection models using existing evidence that is collected until a defined period of time to determine a probability of a future state that is based on evidence to be collected between the defined period of time and the future time, the existing evidence comprising the streaming evidence and other evidence obtained prior to the streaming evidence;

using a predicted future belief obtained from the one or more belief projection models to determine a likelihood of obtaining particular evidence between the defined period of time and the future time to determine whether to act at the defined period of time for a next action or wait for the future time to act for the next action; and

based on determining to act at the defined period of time or wait for the future time to act for the next action, transmitting data using one or more of the output devices at the defined period of time without obtaining the particular evidence or upon receipt of the particular evidence at the future time.

11. The system of claim 10 wherein the processing logic is incorporated into an autonomous assistant that uses the predicted future belief over the state to determine whether to initiate engagement with a person.

12. The system of claim 10 wherein the sensor set comprises a camera, and wherein the processing logic processes streaming evidence received via the camera.

13. The system of claim 10 wherein the sensor set comprises a connection to computing devices that interface with humans that respond with at least some of the particular evidence.

14. The system of claim 10 wherein the output mechanism set comprises a speaker.

15. The system of claim 14 wherein the audio data comprises speech, including speech to initiate engagement with a person.

16. One or more computer-readable hardware storage devices having computer-executable instructions for executing an assistant on a computing device to determine whether or not to act, that cause the assistant to perform operations comprising:

receiving sensory evidence, including high-dimensional streaming evidence into one or more belief projection models, the one or more belief projection models using existing evidence that is collected until a defined period of time to determine a probability of a future state that is based on evidence to be collected between the defined period of time and the future time, the existing evidence comprising the streaming evidence and other evidence obtained prior to the streaming evidence;

determining a likelihood of obtaining particular evidence is between the defined period of time and the future time to determine whether to act at the defined period of time for a next action or wait for the future time to act for the next action;

based on determining to act at the defined period of time or wait for the future time to act for the next action, transmitting data using an output device at the defined period of time without obtaining the particular evidence or upon receipt of the particular evidence at the future time.

17. The one or more computer-readable hardware storage devices of claim 16 having further computer-executable instructions that cause the computing device to perform an operation comprising making a decision to wait for the particular evidence to be received.

18. The one or more computer-readable hardware storage devices of claim 16 having further computer-executable instructions that cause computing device to perform an operation comprising making a decision to act without waiting for the particular evidence.

19. The one or more computer-readable hardware storage devices of claim 16 having further computer-executable instructions that cause the computing device to, based on a determination to act without obtaining the particular evidence, initiating engagement with an electronic assistant using the output device.

20. The one or more computer-readable hardware storage devices of claim 16, wherein constructing the one or more belief inference projection models comprises processing at least part of the existing evidence into percepts.