Distributed scalable incrementally updated models in decisioning systems

• US 8,620,840 B2
• Filed: 07/18/2012
• Issued: 12/31/2013
• Est. Priority Date: 07/19/2011
• Status: Active Grant

• Current Assignee: NICE Systems Technologies UK Limited (Nice Ltd)
• Original Assignee: NICE Systems Technologies UK Limited (Nice Ltd)
• Inventors: Newnham, Leonard Michael, McFall, Jason Derek
• Primary Examiner(s): Chaki, Kakali
• Assistant Examiner(s): SITIRICHE, LUIS A
• Application Number:
  US13/552,519
• Publication Number:
  US 20130024405A1
• Time in Patent Office: 531 Days
• Field of Search: None
• US Class Current: 706/12
• CPC Class Codes: G06F 18/214 Generating training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06F 18/2411 based on the proximity to a decision surface, e.g. support vector machines
• CPC Class Codes: G06N 20/00 Machine learning

Abstract

In one embodiment, first weight information indicating a first set of delta values is obtained, where the first set of delta values includes a first delta value for each weight in a set of weights, the set of weights including a weight for each of a set of one or more parameters of a model. In addition, second weight information indicating a second set of delta values is obtained, where the second set of delta values includes a second delta value for each weight in the set of weights. Combined weight information including a combined set of delta values or a combined set of weights is generated based, at least in part, upon the first weight information and the second weight information.

Claims

1. At least one non-transitory computer-readable medium storing thereon computer-readable instructions for performing a method, comprising:

obtaining weight information indicating two or more sets of delta values, each set of delta values including a delta value for each weight in a set of weights, the set of weights including a weight for each of a set of one or more parameters of a model;

generating a combined set of delta values based, at least in part, upon each of the two or more sets of delta values; and

generating a combined set of weights or providing the combined set of delta values for use in generating the combined set of weights, the combined set of weights being generated based, at least in part, upon the set of weights and the combined set of delta values,

wherein the obtained weight information further comprises two or more sets of counts, where each of the two or more sets of counts includes a count for each weight in the set of weights, wherein the count indicates a number of times the weight has been changed by a corresponding decisioning module during a period of time;

wherein the combined set of delta values is generated based, at least in part, upon each of the two or more sets of delta values and the two or more sets of counts.

5. The non-transitory computer-readable medium as recited in claim 1, wherein each of the two or more sets of delta values is generated by a different one of two or more decisioning components, wherein generating the combined set of delta values comprises:

for each of the two or more decisioning components;

determining, for each of the one or more weights in the set of weights, a total number of times the weight has been updated during a period of time.

6. The non-transitory computer-readable medium as recited in claim 1, the method further comprising:

employing a voting mechanism to determine a relative contribution of two different strategies for generating the combined set of delta values.

7. The non-transitory computer-readable medium as recited in claim 1,wherein generating the combined set of delta values comprises:

applying one or more momentum terms to a previously generated set of delta values.

8. The non-transitory computer-readable medium as recited in claim 1, wherein each of the two or more sets of delta values is generated by a different one of two or more decisioning components, the method further comprising:

applying a traffic routing mechanism to distribute traffic among the two or more decisioning components.

2. At least one non-transitory computer-readable medium storing thereon computer-readable instructions for performing a method, comprising:

obtaining weight information indicating two or more sets of delta values, each set of delta values including a delta value for each weight in a set of weights, the set of weights including a weight for each of a set of one or more parameters of a model;

generating a combined set of delta values based, at least in part, upon each of the two or more sets of delta values; and

generating a combined set of weights or providing the combined set of delta values for use in generating the combined set of weights, the combined set of weights being generated based, at least in part, upon the set of weights and the combined set of delta values,

wherein generating the combined set of delta values comprises;

for each of one or more weights in the set of weights;

obtaining the delta value for the weight from each of the two or more sets of delta values such that two or more delta values for the weight are obtained;

identifying a largest positive value in the two or more delta values for the weight; and

identifying a largest negative value in the two or more delta values for the weight.

3. At least one non-transitory computer-readable medium storing thereon computer-readable instructions for performing a method, comprising:

obtaining weight information indicating two or more sets of delta values, each set of delta values including a delta value for each weight in a set of weights, the set of weights including a weight for each of a set of one or more parameters of a model;

generating a combined set of delta values based, at least in part, upon each of the two or more sets of delta values; and

generating a combined set of weights or providing the combined set of delta values for use in generating the combined set of weights, the combined set of weights being generated based, at least in part, upon the set of weights and the combined set of delta values,

wherein generating the combined set of delta values comprises;

for each of one or more weights in the set of weights;

obtaining the delta value for the weight from each of the two or more sets of delta values such that two or more delta values for the weight are obtained;

determining a percentile value of positive delta values in the two or more delta values for the weight; and

determining a percentile value of negative delta values in the two or more delta values for the weight.

4. At least one non-transitory computer-readable medium storing thereon computer-readable instructions for performing a method, comprising:

obtaining weight information indicating two or more sets of delta values, each set of delta values including a delta value for each weight in a set of weights, the set of weights including a weight for each of a set of one or more parameters of a model;

generating a combined set of delta values based, at least in part, upon each of the two or more sets of delta values; and

generating a combined set of weights or providing the combined set of delta values for use in generating the combined set of weights, the combined set of weights being generated based, at least in part, upon the set of weights and the combined set of delta values,

wherein each of the two or more sets of delta values is generated by a different one of two or more decisioning components, wherein generating the combined set of delta values comprises;

for each of one or more weights in the set of weights;

obtaining the delta value for the weight from each of the two or more sets of delta values such that two or more delta values for the weight are obtained;

determining, a mean delta value cross the two or more decisioning components based, at least in part, the two or more delta values for the weight.