Efficiency of training for ranking systems

• US 20070239632A1
• Filed: 03/17/2006
• Published: 10/11/2007
• Est. Priority Date: 03/17/2006
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Burges, Christopher, Ragno, Robert
• Granted Patent: US 7,617,164 B2
• Time in Patent Office: Days
• US Class Current: 706/15
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: Y10S 707/99935 Query augmenting and refining, e.g. inexact access

Abstract

The subject disclosure pertains to systems and methods for facilitating training of machine learning systems utilizing pairwise training. The number of computations required during pairwise training is reduced by grouping the computations. First, a score is generated for each retrieved data item. During processing of the data item pairs, the scores of the data items in the pair are retrieved and used to generate a gradient for each data item. Once all of the pairs have been processed, the gradients for each data item are aggregated and the aggregated gradients are used to update the machine learning system.

Claims

1. Method for facilitating training of a learning machine utilizing a pairwise algorithm, comprising:

generating a score for each of a plurality of data items;

maintaining the scores for the plurality of data items;

generating at least one gradient for each of the plurality of data items based at least in part upon the maintained scores of pairs of data items; and

updating the learning machine based at least in part upon the at least one gradient of each of the plurality of data items.

2. The method of claim 1, generating gradients for the plurality of data items, further comprises:

selecting a pair of data items; and

obtaining the maintained scores for the pair of data items, the at least one gradient for each data item in the pair is based at least in part upon the maintained scores of the pair of data items.

3. The method of claim 2, further comprising:

aggregating the at least one gradient for each of the plurality of data items.

4. The method of claim 2, further comprising:

obtaining internal parameters of the learning machine for each of the plurality of data items, the internal parameters are utilized to update the learning machine.

5. The method of claim 4, further comprising:

maintaining the internal parameters for each of the plurality of data items; and

retrieving the maintained internal parameters.

6. The method of claim 1, the plurality of data items includes at least one of a text file, a web page, an image, audio data, video data and a word processing file.

7. The method of claim 1, the learning machine is a neural network.

8. The method of claim 7, generating a score further comprises performing forward propagation, and updating the learning machine further comprises performing backward propagation.

9. The method of claim 8, further comprising:

obtaining internal parameters of the neural network during the forward propagation for score generation;

maintaining the internal parameters; and

utilizing the internal parameters during the backward propagation.

10. The method of claim 8, updating the learning machine further comprises:

performing a second forward propagation to obtain internal parameters of the neural network; and

utilizing the internal parameters in the backward propagation.

11. A system for facilitating training of a learning machine utilizing a pairwise algorithm, comprising:

a scorer component that generates a score for each of a plurality of data items;

a memory component that maintains the score for each of the plurality of data items;

a pair processor component that generates a gradient for each data item in a pair of data items based at least in part upon the scores for the data items in the pair; and

an update component that updates the learning machine based at least in part upon all of the gradients for each data item of the plurality of data items.

12. The system of claim 11, the memory component maintains an aggregated gradient for each of the plurality of data items and the update component obtains the aggregated gradient for each data item.

13. The system of claim 11, further comprising a parameter component for obtaining internal parameters of the learning machine for each of the plurality of data items, the update component updates the learning machine based at least in part upon the internal parameters.

14. The system of claim 11, the learning system is a neural network.

15. The system of claim 14, the scorer component performs forward propagation and the update component performs backward propagation.

16. The system of claim 15, the scorer component obtains internal parameters of the neural network during forward propagation, the memory component maintains the internal parameters and the update component utilizes the internal parameters during the backward propagation.

17. The system of claim 15, further comprising:

a parameter component that performs a second forward propagation to obtain internal parameters of the neural network during update, and the update component utilizes the internal parameters during backward propagation.

18. A system for training a learning machine using pairwise training, comprising:

means for generating a score for each of a plurality of data items;

means for maintaining the scores for the plurality of data items;

means for generating gradients for the plurality of data items based at least in part upon the scores of pairs of data items; and

means for updating the learning machine based at least in part upon the gradients of each of the plurality of data items.

19. The system of claim 18, means for generating gradients, further comprises

means for selecting a pair of data items;

means for obtaining the scores for the pair of data items; and

means for generating a gradient for each data item in the pair, the gradient based at least in part upon the retrieved scores of the pair of data items.

20. The system of claim 18, further comprising:

means for obtaining internal parameters of the learning machine for each of the plurality of data item, the internal parameters utilized in updating the learning machine; and

means for maintaining internal parameters.