Machine learning with distributed training

Patent metadata

Field	Value
Publication number	US-10380504-B2
Application number	US-201715849356-A
Country	US
Kind code	B2
Filing date	Dec 20, 2017
Priority date	May 5, 2017
Publication date	Aug 13, 2019
Grant date	Aug 13, 2019

Abstract

A network system may include a plurality of trainer devices and a computing system disposed within a remote network management platform. The computing system may be configured to: receive, from a client device of a managed network, information indicating (i) training data that is to be used as basis for generating a machine learning (ML) model and (ii) a target variable to be predicted using the ML model; transmit an ML training request for reception by one of the plurality of trainer devices; provide the training data to a particular trainer device executing a particular ML trainer process that is serving the ML training request; receive, from the particular trainer device, the ML model that is generated based on the provided training data and according to the particular ML trainer process; predict the target variable using the ML model; and transmit, to the client device, information indicating the target variable.

First claim

What is claimed is:

1. A remote network management platform comprising:
a plurality of computational instances dedicated to respective managed networks;
a plurality of machine learning (ML) trainer devices, configured to execute ML trainer processes; and
a scheduler device configured to:
receive an ML training request from a particular computational instance of the plurality of computational instances;
assign the ML training request to a particular ML trainer process of the ML trainer processes and a particular ML trainer device of the plurality of ML trainer devices, wherein the ML training request identifies: training data to be used as basis for generating an ML model, and a target variable to be predicted using the ML model; and
provide an identifier of the particular computational instance to the particular ML trainer device, wherein the identifier enables direct communication between the particular ML trainer device and the particular computational instance;

wherein the assigning of the ML training request causes the particular ML trainer device of the plurality of ML trainer devices to: receive the training data from the particular computational instance, train the ML model using the particular ML trainer process in accordance with the received training data and the target variable, and provide the ML model as trained to the particular computational instance for local use on the computational instance.

2. The remote network management platform of claim 1 , wherein the ML training request is based on information received by the particular computational instance from a client device of a managed network, and wherein the information received from the client device indicates the training data and the target variable.

3. The remote network management platform of claim 2 , wherein the information received from the client device specifies a training time, and wherein the scheduler device assigning the ML training request to the particular ML trainer process comprises the scheduler device assigning the particular ML trainer process to serve the ML training request at the specified training time.

4. The remote network management platform of claim 1 , wherein the particular ML trainer device is configured to execute the particular ML trainer process, wherein the scheduler device is further configured to:
make a determination that a location of the particular ML trainer device is within a threshold distance of a location of the particular computational instance, and
wherein the scheduler device assigns the ML training request to the particular ML trainer process based at least on the determination.

5. The remote network management platform of claim 1 , wherein the scheduler device is further configured to:
make a determination that the particular ML trainer process is available to serve the ML training request, and
wherein the scheduler device assigns the ML training request to the particular ML trainer process based at least on the determination.

6. The remote network management platform of claim 1 , wherein the particular ML trainer device is configured to execute the particular ML trainer process, and wherein the scheduler device is further configured to:
determine a performance metric indicating performance of the particular ML trainer device, and
wherein the scheduler device assigns the ML training request to the particular ML trainer process based at least on the determined performance metric meeting a criteria.

7. The remote network management platform of claim 1 , wherein the particular ML trainer device is configured to:
while training the ML model, transmit, to the particular computational instance, a message indicating a status of the ML training request.

8. The remote network management platform of claim 7 , wherein the particular computational instance is configured to:
while the particular ML trainer device is training the ML model, transmit, to a client device of a respectively managed network, the message indicating the status of the ML training request.

9. The remote network management platform of claim 1 , wherein the particular computational instance is configured to:
predict the target variable using the ML model; and
transmit, to a client device of a respectively managed network, information indicating the target variable as predicted.

10. The remote network management platform of claim 9 , wherein a web browser is operated by the client device, and wherein transmitting, to the client device, information indicating the target variable comprises causing the web browser to display the information indicating the target variable.

11. The remote network management platform of claim 1 , wherein the particular ML trainer device comprises a temporary data storage device, and wherein the particular ML trainer device is configured to:
store the training data at the temporary data storage device while training the ML model using the particular ML trainer process; and
delete the training data from the temporary data storage device after completing the training of the ML model using the particular ML trainer process.

12. The remote network management platform of claim 1 , wherein the particular ML trainer device is configured to:
receive a randomly generated bitstring associated with the ML training request, wherein the randomly generated bitstring is originally transmitted by the particular computational instance for reception by one of the plurality of ML trainer devices;
transmit the randomly generated bitstring to the particular computational instance along with a request for the training data; and
receive the requested training data from the particular computational instance upon the particular computational instance verifying that the randomly generated bitstring transmitted by the particular ML trainer device is identical to the randomly generated bitstring originally transmitted by the particular computational instance.

13. The remote network management platform of claim 1 , wherein the scheduler device is further configured to:
receive a second ML training request from a second computational instance of the plurality of computational instances; and
assign the second ML training request to a second ML trainer process of the ML trainer processes, wherein the second ML training request identifies: second training data to be used as basis for generating a second ML model, and a second target variable to be predicted using the second ML model,
wherein the assigning of the second ML training request causes a second ML trainer device of the plurality of ML trainer devices to: receive the second training data from the second computational instance, train the second ML model using the second ML trainer process in accordance with the received second training data and the second target variable, and provide the second ML model as trained to the second computational instance.

14. The remote network management platform of claim 13 , wherein the ML trainer device and the second ML trainer device are the same ML trainer device, and wherein the second ML trainer process is different from the ML trainer process.

15. The remote network management platform of claim 13 , wherein the ML trainer device and the second ML trainer device are the same ML trainer device, and wherein the ML trainer process and second ML trainer process are the same ML trainer process.

16. A method comprising:
receiving, by a scheduler device of a remote network management platform, a machine learning (ML) training request from a particular computational instance, wherein the particular computation

Assignees

• Servicenow Inc

Inventors

• Bendre Nikhil
• Ros Fernando
• Govindarajan Kannan
• Jayaraman Baskar
• Thakur Aniruddha
• Palapudi Sriram
• Karakusoglu Firat

Classifications

• G06F9/46

  Multiprogramming arrangements · CPC title

• H04L67/00

  Network arrangements or protocols for supporting network services or applications (user-to-user messaging H04L51/00; network arrangements, protocols or services for supporting real-time applications in data packet communications networks H04L65/00) · CPC title

• H04L41/12

  Discovery or management of network topologies · CPC title

• H04L67/32

  Electricity · mapped topic

• G06F17/27

  Physics · mapped topic