SYSTEMS, DEVICES, AND/OR METHODS FOR MANAGING SAMPLE SELECTION BIAS

• US 20130041650A1
• Filed: 10/15/2012
• Published: 02/14/2013
• Est. Priority Date: 12/23/2008
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: AT&T Intellectual Property I LP (AT&T, Inc.)
• Inventors: Phillips, Steven
• Granted Patent: US 8,769,094 B2
• Time in Patent Office: Days
• US Class Current: 704/2
• CPC Class Codes: G06F 17/18 for evaluating statistical data , e.g. average values, frequency distributions, probability functions, regression analysis forecasting specially adapted for a specific administrative, business or logistic context G06Q10/04

Abstract

Certain exemplary embodiments can provide a method that can include, via a special purpose processor, automatically determining an unbiased estimate of a distribution from occurrence data having an occurrence data sample selection bias substantially equivalent to a background data sample selection bias, the occurrence data related to background data, the background data chosen with the background data sample selection bias, the occurrence data representing a physically-measurable variable of one or more physical and tangible objects or substances.

Claims

1. A system for managing sample selection bias, comprising;

a memory that stores instructions; and

a processor that executes the instructions to perform operations, the operations comprising;

randomly selecting sentences in a source language to obtain background data, wherein the background data has a background data sample selection bias that is substantially equivalent to an occurrence data sample selection bias of occurrence data;

sampling sentences in the source language that contain a predetermined word to obtain the occurrence data; and

determining an unbiased estimate of a distribution utilized for language translation from the occurrence data, wherein the occurrence data is related to the background data.

2. The system of claim 1, wherein the operations further comprise selecting the background data such that the background data and the occurrence data have equivalent environmental biases.

3. The system of claim 1, wherein the operations further comprising determining a set of constraints that the distribution satisfies.

4. The system of claim 3, wherein the operations further comprise determining a distribution of maximum entropy that is subject to the set of constraints.

5. The system of claim 1, wherein the operations further comprise determining parameters that minimize a penalty function and maximize a likelihood of the background data.

6. The system of claim 1, wherein the operations further comprise determining the unbiased estimate by utilizing a regression technique.

7. The system of claim 1, wherein the unbiased estimate of the distribution is further utilized for speech interpretation.

8. The system of claim 1, wherein the occurrence data represents a physically-measurable variable of a physical and tangible object, and a physical and tangible substance.

9. The system of claim 1, wherein the predetermined word is contained in translations of the sentences sampled in the source language.

10. The system of claim 1, wherein the operations further comprise rendering a symbol that represents the unbiased estimate of the distribution via an interface.

11. A method for managing sample selection bias, comprising;

randomly selecting sentences in a source language to obtain background data, wherein the background data has a background data sample selection bias that is substantially equivalent to an occurrence data sample selection bias of occurrence data;

sampling sentences in the source language that contain a predetermined word to obtain the occurrence data; and

determining an unbiased estimate of a distribution utilized for speech interpretation from the occurrence data by utilizing instructions stored in memory and executed by a processor, wherein the occurrence data is related to the background data.

12. The method of claim 11, further comprising rendering a symbol that represents the unbiased estimate of the distribution via an interface.

13. The method of claim 11, further comprising selecting the background data such that the background data and the occurrence data have environmental biases that are the same.

14. The method of claim 11, wherein the unbiased estimate of the distribution is further utilized for language translation.

15. The method of claim 11, further comprising determining parameters that minimize a penalty function and maximize a likelihood of the background data.

16. The method of claim 11, wherein the predetermined word is contained in translations of the sentences sampled in the source language.

17. The method of claim 11, further comprising determining a set of constraints that the distribution satisfies.

18. The method of claim 17, further comprising determining a distribution of maximum entropy that is subject to the set of constraints.

19. The method of claim 11, further comprising randomly selecting the sentences and sampling the sentences at a selected node of a network.

20. A tangible computer-readable medium comprising instructions, which when executed by a processor, cause the processor to perform operations comprising:

randomly selecting sentences in a source language to obtain background data, wherein the background data has a background data sample selection bias that is substantially equivalent to an occurrence data sample selection bias of occurrence data;

sampling sentences in the source language that contain a predetermined word to obtain the occurrence data; and

determining an unbiased estimate of a distribution from the occurrence data, wherein the occurrence data is related to the background data.