Machine learning worker node architecture

Patent metadata

Field	Value
Publication number	US-11574235-B2
Application number	US-201816135630-A
Country	US
Kind code	B2
Filing date	Sep 19, 2018
Priority date	Sep 19, 2018
Publication date	Feb 7, 2023
Grant date	Feb 7, 2023

Abstract

A database contains a corpus of incident reports, a machine learning (ML) model trained to calculate paragraph vectors of the incident reports, and a look-up set table that contains a list of paragraph vectors respectively associated with sets of the incident reports. A plurality of ML worker nodes each store the look-up set table and are configured to execute the ML model. An update thread is configured to: determine that the look-up set table has expired; update the look-up set table by: (i) adding a first set of incident reports received since a most recent update of the look-up set table, and (ii) removing a second set of incident reports containing timestamps that are no longer within a sliding time window; store, in the database, the look-up set table as updated; and transmit, to the ML worker nodes, respective indications that the look-up set table has been updated.

First claim

What is claimed is:

1. A system comprising:
a database containing a corpus of incident reports, a machine learning (ML) model trained to calculate paragraph vectors of the incident reports, and a look-up set table related to the ML model, wherein the paragraph vectors map one or more text fields of the incident reports into a semantically encoded vector space, wherein the look-up set table contains a list of paragraph vectors and respective associations between each paragraph vector in the list and sets of the incident reports, wherein each paragraph vector in the look-up set table is unique and is associated with a set of incident report IDs identifying incident reports that contain the text field for which the paragraph vector has been calculated, and wherein the incident reports referenced by the look-up set table each contain a timestamp that is within a sliding time window of a pre-determined length;
a plurality of ML worker nodes, each storing the look-up set table and configured to execute the ML model to calculate a paragraph vector for a text field of a new incident report, and look for paragraph vectors in the look-up set table that are similar to the paragraph vector for the text field of the new incident report; and
program instructions that, when executed, are configured to cause an update thread to perform operations including:
determining that the look-up set table has expired;
updating the look-up set table by: (i) adding a first set of incident report IDs for incident reports received since a most recent update of the look-up set table, and (ii) removing a second set of incident report IDs for incident reports containing timestamps that are no longer within the sliding time window;
storing, in the database, the look-up set table as updated; and
transmitting, to one or more of the plurality of ML worker nodes, respective indications that the look-up set table has been updated, wherein reception of the respective indications causes the one or more ML worker nodes that were notified to retrieve, from the database, the look-up set table as updated.

2. The system of claim 1 , wherein each of the plurality of ML worker nodes stores the ML model.

3. The system of claim 1 , wherein determining that the look-up set table has expired comprises determining that a periodic timer associated with the look-up set table has fired.

4. The system of claim 3 , wherein the sliding time window is longer than a period of the periodic timer.

5. The system of claim 1 , wherein the program instructions are stored within a particular ML worker node of the plurality of ML worker nodes, wherein the update thread is executed by the particular ML worker node, and wherein transmitting, to the one or more ML worker nodes, the respective indications that the look-up set table has been updated comprises:
transmitting, to all of the plurality of ML worker nodes except for the particular ML worker node, the respective indications that the look-up set table has been updated.

6. The system of claim 1 , wherein each of the plurality of ML worker nodes is a physically distinct computing device.

7. The system of claim 1 , wherein the timestamps record when the incident reports were opened.

8. The system of claim 1 , wherein the timestamps record when the incident reports were closed.

9. The system of claim 1 , wherein the plurality of ML worker nodes is further configured to:
receive new incident reports;
calculate respective paragraph vectors of the new incident reports; and
use the look-up set table as updated to determine similarities between the respective paragraph vectors of the new incident reports and the list of paragraph vectors.

10. A computer-implemented method comprising:
determining, by an update thread executing on a computing device, that a look-up set table has expired, wherein the look-up set table is related to a machine learning (ML) model trained to calculate paragraph vectors of incident reports, wherein the paragraph vectors map one or more text fields of the incident reports into a semantically encoded vector space, wherein the look-up set table contains a list of paragraph vectors and respective associations between each paragraph vector in the list and sets of the incident reports, wherein each paragraph vector in the look-up set table is unique and is associated with a set of incident report IDs identifying incident reports that contain the text field for which the paragraph vector has been calculated, wherein the incident reports referenced by the look-up set table each contain a timestamp that is within a sliding time window of a pre-determined length, and wherein the look-up set table is stored by each of a plurality of ML worker nodes, each being configured to calculate a paragraph vector for a text field of a new incident report using the ML model and to look for paragraph vectors in the look-up set table that are similar to the paragraph vector for the text field of the new incident report;
updating, by the update thread, the look-up set table by: (i) adding a first set of incident report IDs for incident reports received since a most recent update of the look-up set table, and (ii) removing a second set of incident report IDs for incident reports containing timestamps that are no longer within the sliding time window;
storing, by the update thread and in a database, the look-up set table as updated; and
transmitting, by the update thread and to one or more of the plurality of one or more ML worker nodes, respective indications that the look-up set table has been updated, wherein reception of the respective indications causes the one or more ML worker nodes that were notified to retrieve, from the database, the look-up set table as updated.

11. The computer-implemented method of claim 10 , wherein each of the plurality of ML worker nodes stores the ML model.

12. The computer-implemented method of claim 10 , wherein determining that the look-up set table has expired comprises determining that a periodic timer associated with the look-up set table has fired.

13. The computer-implemented method of claim 12 , wherein the sliding time window is longer than a period of the periodic timer.

14. The computer-implemented method of claim 10 , wherein the update thread is executed by a particular ML worker node of the plurality of ML worker nodes, and wherein transmitting, to the one or more ML worker nodes, the respective indications that the look-up set table has been updated comprises:
transmitting, to all of the plurality of ML worker nodes except for the particular ML worker node, the respective indications that the look-up set table has been updated.

15. The computer-implemented method of claim 10 , wherein the timestamps record when the incident reports were opened.

16. The computer-implemented method of claim 10 , wherein the timestamps record when the incident reports were closed.

17. An article of manufacture including a non-transitory computer-readable medium, having stored thereon program instructions that, upon execution by a computing system, cause the computing system to perform operations comprising:
determining, by an update thread executing on the computing system, that a look-up set table has expired, wherein the look-up set table is related to a machine learning (ML) model trained to calculate paragraph vectors of incident reports, wherein the paragraph vectors map one or more text fields of the incident reports into a semantically encoded vector space, wherein the look-up set table contains a list of paragraph vectors and respective associations betwe

Assignees

• Servicenow Inc

Inventors

• Jayaraman Baskar
• Thakur Aniruddha Madhusudan
• Feng Tao
• Govindarajan Kannan

Classifications

• G06F16/316

  Indexing structures · CPC title

• G06F16/9017

  using directory or table look-up (use of a directory or look-up table in file systems G06F16/13) · CPC title

• G06N20/00 Primary

  Machine learning · CPC title

• G06F16/2272

  Management thereof · CPC title

• G06F40/30 Primary

  Semantic analysis · CPC title