Stream processing with multiple connections between local and central modelers

• US 10,579,937 B2
• Filed: 05/12/2015
• Issued: 03/03/2020
• Est. Priority Date: 09/08/2014
• Status: Active Grant

• Current Assignee: Pivotal Software, Inc. (Broadcom, Inc.)
• Original Assignee: Pivotal Software, Inc. (Broadcom, Inc.)
• Inventors: Brand, Michael, Adams, Lyndon John, Brown, David Russell, Ng, Kee Siong
• Primary Examiner(s): Rifkin, Ben M
• Application Number:
  US14/710,418
• Publication Number:
  US 20160071014A1
• Time in Patent Office: 1,757 Days
• Field of Search: None
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 5/02 Knowledge representation; Symbolic representation
• CPC Class Codes: G06N 5/04 Inference or reasoning models
• CPC Class Codes: H04L 41/16 using machine learning or artificial intelligence
• CPC Class Codes: H04L 45/08 Learning-based routing, e.g. using neural networks or artificial intelligence

Abstract

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for stream processing. One method includes receiving an event stream of first events by a first plurality of first local modelers of a stream processing system. Each local modeler processes a portion of received events of the event stream according to a first set of operations, the operations including aggregating information associated with each event to generate first aggregated information. A second plurality of second local modelers similarly generates second aggregated information from an event stream of second events. First and second local modelers provide, to a first central modeler, first and second aggregated information. A set of parameters of a respective machine learning model is determined by the first central modeler using the received aggregated information.

Claims

1. A method performed by a distributed stream processing system comprising a plurality of computers to implement a stream processing pipeline having a plurality of stages, the method comprising:

implementing, by a plurality of first local modelers executing on one or more first computers of the stream processing system, a first stage of the stream processing pipeline, including;

processing, by each first local modeler, events in an event stream received by the first local modeler according to a first transformation of the first stage of the stream processing pipeline using first machine learning model parameters of a first machine learning model,

providing, by each first local modeler of one or more of the first local modelers to a first central modeler executing on one or more computers of the stream processing system that are separate from each of the one or more first computers, a respective first representative value computed from events received by the first local modeler,

computing, by the first central modeler, updated machine learning model parameters of the first machine learning model using the respective first representative values computed by the one or more first local modelers, and

providing, by the first central modeler to one or more of the first local modelers, the computed updated machine learning model parameters of the first machine learning model, thereby causing the first local modelers to process events according to the first transformation of the first stage of the stream processing pipeline using the updated machine learning model parameters for the first machine learning model;

implementing, by a plurality of second local modelers executing on one or more second computers of the stream processing system that are separate from each of the one or more first computers of the first local modelers, a second stage of the stream processing pipeline, including;

processing, by each second local modeler, one or more processed events generated by one or more first local modelers received by the second local modeler according to a second transformation of the second stage of the stream processing pipeline using second machine learning model parameters of a second machine learning model,

providing, by each second local modeler of one or more of the second local modelers to a second central modeler executing on one or more computers of the stream processing system that are separate from each of the one or more second computers, a respective second representative value computed from the processed events received by the second local modeler,

computing, by the second central modeler, updated machine learning model parameters of a second machine learning model using the respective second representative values computed by each respective second local modeler from the processed events generated by the one or more second local modelers, and

providing, by the second central modeler to one or more of the second local modelers, the computed updated machine learning model parameters of the second machine learning model, thereby causing the second local modelers to process events according to the second transformation of the second stage of the stream processing pipeline using the updated machine learning model parameters for the second machine learning model.

2. The method of claim 1, wherein:

providing the respective first representative values to the first central modeler comprises providing the respective first representative values asynchronously; and

each first local modeler receives a respective portion of the events from a router that executes on the stream processing system.

3. The method of claim 1, wherein providing the respective first representative values comprises, by each of the first local modelers:

determining that the first local modeler has processed a threshold number of events and then, as a consequence, providing the first representative value generated by the local modeler to the first central modeler.

4. The method of claim 1, comprising:

receiving, by the first central modeler, data specifying a learning program for determining parameters of the first machine learning model from one or more representative values computed by the first local modelers; and

determining, by the first central modeler, the parameters of the first machine learning model by executing the learning program on received representative values computed by the first local modelers.

5. The method of claim 1,

wherein each respective first representative computed by a respective first local modeler is an average value for events processed by the first local modeler, and

wherein the first machine learning model parameters computed by the first central modeler represent an average value of events processed by all of the plurality of first local modelers.

6. The method of claim 5, wherein the average value of events is part of an average face vector representing an average of facial properties processed by all of the plurality of first local modelers.

7. The method of claim 6, wherein the second machine learning model parameters of the second machine learning model represent one or more eigenfaces, each eigenface being a vector in a determined space of representative facial properties processed by the plurality of first local modelers.

8. A distributed stream processing system comprising a plurality of computers to implement a stream processing pipeline having a plurality of stages comprising:

one or more computers and one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising;

implementing, by a plurality of first local modelers executing on one or more first computers of the stream processing system, a first stage of the stream processing pipeline, including;

processing, by each first local modeler, events in an event stream received by the first local modeler according to a first transformation of the first stage of the stream processing pipeline using first machine learning model parameters of a first machine learning model,

providing, by each first local modeler of one or more of the first local modelers to a first central modeler executing on one or more computers of the stream processing system that are separate from each of the one or more first computers, a respective first representative value computed from events received by the first local modeler,

computing, by the first central modeler, updated machine learning model parameters of the first machine learning model using the respective first representative values computed by the one or more first local modelers, and

providing, by the first central modeler to one or more of the first local modelers, the computed updated machine learning model parameters of the first machine learning model, thereby causing the first local modelers to process events according to the first transformation of the first stage of the stream processing pipeline using the updated machine learning model parameters for the first machine learning model;

implementing, by a plurality of second local modelers executing on one or more second computers of the stream processing system that are separate from each of the one or more first computers of the first local modelers, a second stage of the stream processing pipeline, including;

processing, by each second local modeler, one or more processed events generated by one or more first local modelers received by the second local modeler according to a second transformation of the second stage of the stream processing pipeline using second machine learning model parameters of a second machine learning model,

providing, by each second local modeler of one or more of the second local modelers to a second central modeler executing on one or more computers of the stream processing system that are separate from each of the one or more second computers, a respective second representative value computed from the processed evens received by the second local modeler,

computing, by the second central modeler, updated machine learning model parameters of a second machine learning model using the respective second representative values computed by each respective second local modeler from the processed events generated by the one or more second local modelers, and

providing, by the second central modeler to one or more of the second local modelers, the computed updated machine learning model parameters of the second machine learning model, thereby causing the second local modelers to process events according to the second transformation of the second stage of the stream processing pipeline using the updated machine learning model parameters for the second machine learning model.

9. The system of claim 8, wherein:

providing the respective first representative values to the first central modeler comprises providing the respective first representative values asynchronously; and

each first local modeler receives a respective portion of the events from a router that executes on the stream processing system.

10. The system of claim 8, wherein providing the respective first representative values comprises, by each of the first local modelers:

determining that the first local modeler has processed a threshold number of events and then, as a consequence, providing the first representative value generated by the local modeler to the first central modeler.

11. The system of claim 8, wherein the operations further comprise:

receiving, by the first central modeler, data specifying a learning program for determining parameters of the first machine learning model from one or more representative values computed by the first local modelers; and

determining, by the first central modeler, the parameters of the first machine learning model by executing the learning program on received representative values computed by the first local modelers.

12. The system of claim 8,

wherein each respective first representative computed by a respective first local modeler is an average value for events processed by the first local modeler, and

wherein the first machine learning model parameters computed by the first central modeler represent an average value of events processed by all of the plurality of first local modelers.

13. The system of claim 12, wherein the average value of events is part of an average face vector representing an average of facial properties processed by all of the plurality of first local modelers.

14. The system of claim 13, wherein the second machine learning model parameters of the second machine learning model represent one or more eigenfaces, each eigenface being a vector in a determined space of representative facial properties processed by the plurality of first local modelers.

15. A computer program product performed by a distributed stream processing system comprising a plurality of computers to implement a stream processing pipeline having a plurality of stages, encoded on one or more non-transitory computer storage media, comprising instructions that when executed by one or more computers cause the one or more computers to perform operations comprising:

implementing, by a plurality of first local modelers executing on one or more first computers of the stream processing system, a first stage of the stream processing pipeline, including;

processing, by each first local modeler, events in an event stream received by the first local modeler according to a first transformation of the first stage of the stream processing pipeline using first machine learning model parameters of a first machine learning model,

providing, by each first local modeler of one or more of the first local modelers to a first central modeler executing on one or more computers of the stream processing system that are separate from each of the one or more first computers, a respective first representative value computed from events received by the first local modeler,

computing, by the first central modeler, updated machine learning model parameters of the first machine learning model using the respective first representative values computed by the one or more first local modelers, and

providing, by the first central modeler to one or more of the first local modelers, the computed updated machine learning model parameters of the first machine learning model, thereby causing the first local modelers to process events according to the first transformation of the first stage of the stream processing pipeline using the updated machine learning model parameters for the first machine learning model;

implementing, by a plurality of second local modelers executing on one or more second computers of the stream processing system that are separate from each of the one or more first computers of the first local modelers, a second stage of the stream processing pipeline, including;

processing, by each second local modeler, one or more processed events generated by one or more first local modelers received by the second local modeler according to a second transformation of the second stage of the stream processing pipeline using second machine learning model parameters of a second machine learning model,

providing, by each second local modeler of one or more of the second local modelers to a second central modeler executing on one or more computers of the stream processing system that are separate from each of the one or more second computers, a respective second representative value computed from the processed events received by the second local modeler,

computing, by the second central modeler, updated machine learning model parameters of a second machine learning model using the respective second representative values computed by each respective second local modeler from the processed events generated by the one or more second local modelers, and

providing, by the second central modeler to one or more of the second local modelers, the computed updated machine learning model parameters of the second machine learning model, thereby causing the second local modelers to process events according to the second transformation of the second stage of the stream processing pipeline using the updated machine learning model parameters for the second machine learning model.

16. The computer program product of claim 15, wherein:

providing the respective first representative values to the first central modeler comprises providing the respective first representative values asynchronously; and

each first local modeler receives a respective portion of the events from a router that executes on the stream processing system.

17. The computer program product of claim 15, wherein providing the respective first representative values comprises, by each of the first local modelers:

determining that the first local modeler has processed a threshold number of events and then, as a consequence, providing the first representative value generated by the local modeler to the first central modeler.

18. The computer program product of claim 15, wherein the operations further comprise:

receiving, by the first central modeler, data specifying a learning program for determining parameters of the first machine learning model from one or more representative values computed by the first local modelers; and

determining, by the first central modeler, the parameters of the first machine learning model by executing the learning program on received representative values computed by the first local modelers.

19. The computer program product of claim 15,

wherein each respective first representative computed by a respective first local modeler is an average value for events processed by the first local modeler, and

wherein the first machine learning model parameters computed by the first central modeler represent an average value of events processed by all of the plurality of first local modelers.

20. The computer program product of claim 19, wherein the average value of events is part of an average face vector representing an average of facial properties processed by all of the plurality of first local modelers.

21. The computer program product of claim 20, wherein the second machine learning model parameters of the second machine learning model represent one or more eigenfaces, each eigenface being a vector in a determined space of representative facial properties processed by the plurality of first local modelers.