Learning model selection in a distributed network

Patent metadata

Field	Value
Publication number	US-9413779-B2
Application number	US-201414164443-A
Country	US
Kind code	B2
Filing date	Jan 27, 2014
Priority date	Jan 6, 2014
Publication date	Aug 9, 2016
Grant date	Aug 9, 2016

Abstract

In one embodiment, local model parameters are generated by training a machine learning model at a device in a computer network using a local data set. One or more other devices in the network are identified that have trained machine learning models using remote data sets that are similar to the local data set. The local model parameters are provided to the one or more other devices to cause the one or more other devices to generate performance metrics using the provided model parameters. Performance metrics for model parameters are received from the one or more other devices and a global set of model parameters is selected for the device and the one or more other devices using the received performance metrics.

First claim

What is claimed is:

1. A method comprising:
generating local model parameters by training a machine learning model at a device in a computer network using a local data set;
identifying, at the device, one or more other devices in the network that have trained machine learning models using remote data sets that are similar to the local data set;
transmitting, by the device, the local model parameters to the one or more other devices to cause the one or more other devices to generate performance metrics using the transmitted model parameters:
receiving, at the device, the performance metrics from the one or more other devices;
receiving, at the device, model parameters from the one or more other devices that were generated by the one or more other devices training one or more other machine learning models;
using, by the device, the received model parameters with the local data set to generate local performance metrics;
comparing, by the device, the local performance metrics with the performance metrics received from the one or more other devices to select a global set of model parameters;
selecting, by the device, the global set of model parameters for the device and the one or more other devices based on the comparison between the local performance metrics and the received performance metrics; and
selecting, by the device, the model parameters having the highest average performance metrics as the global set of model parameters.

2. The method as in claim 1 , wherein the machine learning model trained by the device is an artificial neural network (ANN) and the trained parameters comprise weighted links between neurons in the ANN.

3. The method as in claim 1 , wherein identifying the one or more other devices comprises: sending an identification request to a network management service (NMS).

4. The method as in claim 1 , wherein identifying the one or more other devices comprises: multicasting identification requests to the one or more other devices.

5. The method as in claim 1 , wherein the highest average performance metrics are the highest weighted average performance metrics.

6. The method as in claim 1 , further comprising: selecting the model parameters that performed best on the highest number of devices as the global set of model parameters.

7. The method as in claim 1 , wherein the machine learning model is configured to determine a model output selected from a group comprising: a detected network attack, a detected network anomaly, or an estimated network parameter.

8. The method as in claim 1 , further comprising:
receiving an identification request from a remote network device that comprises data regarding the remote network device and devices managed by the remote network device;
determining that the remote network device uses a data set that is similar to the local data set based on the identification request; and
notifying the remote network device that the remote network devices uses a similar data set.

9. The method as in claim 8 , wherein the data regarding the remote network device is selected from the group comprising: hardware resources of the remote network device and traffic levels of the remote network device.

10. The method as in claim 8 , wherein the data regarding the devices managed by the remote network device is selected from the group comprising: the number of managed devices, the density of managed devices, or the types of applications executed by the managed devices.

11. An apparatus, comprising:
one or more network interfaces to communicate in a computer network;
a processor coupled to the network interfaces and configured to execute one or more processes; and
a memory configured to store a process executable by the processor, the process when executed operable to:
generate local model parameters by training a machine learning model using a local data set;
identify one or more other devices that have trained machine learning models using remote data sets that are similar to the local data set;
transmit the local model parameters to the one or more other devices to cause the one or more other devices to generate performance metrics using the transmitted model parameters;
receive the performance metrics from the one or more other devices;
receive model parameters from the one or more other devices that were generated by the one or more other devices training one or more other machine learning models:
use the received model parameters with the local data set to generate local performance metrics;
compare the local performance metrics with the performance metrics received from the one or more other devices to select a global set of model parameters;
select the global set of model parameters for the device and the one or more other devices based on the comparison between the local performance metrics and the received performance metrics; and
select the model parameters having the highest average performance metrics as the global set of model parameters.

12. The apparatus as in claim 11 , wherein the machine learning model trained by the device is an artificial neural network (ANN) and the trained parameters comprise weighted links between neurons in the ANN.

13. The apparatus as in claim 11 , wherein the machine learning model is configured to determine a model output selected from a group comprising: a detected network attack, a detected network anomaly, or an estimated network parameter.

14. The apparatus as in claim 11 , wherein the process when executed is further operable to: multicast identification requests to the one or more other devices.

15. The apparatus as in claim 11 , wherein the process when executed is further operable to:
generate performance metrics using the local data set with the local model parameters; and
transmitting the generated performance metrics using the local model parameters to the one or more other devices.

16. A tangible, non-transitory, computer-readable media having software encoded thereon, the software when executed by a processor operable to:
generate local model parameters by training a machine learning model using a local data set;
identify, for the device, one or more other devices that have trained machine learning models using remote data sets that are similar to the local data set;
transmit, for the device, the local model parameters to the one or more other devices to cause the one or more other devices to generate performance metrics using the transmitted model parameters;
receive, at the device, the performance metrics from the one or more other devices;
receive model parameters from the one or more other devices that were generated by the one or more other devices training one or more other machine learning models;
use the received model parameters with the local data set to generate local performance metrics
compare the local performance metrics with the performance metrics received from the one or more other devices to select a global set of model parameters:
select the global set of model parameters for the device and the one or more other devices based on the comparison between the local performance metrics and the received performance metrics: and
select the model parameters having the highest average performance metrics as the global set of model parameters.

Assignees

• Cisco Tech Inc

Inventors

• Vasseur Jean-Philippe
• Di Pietro Andrea
• Cruz Mota Javier

Classifications

• G06N3/08

  Learning methods · CPC title

• G06N3/09

  Supervised learning · CPC title

• H04L45/48

  Routing tree calculation · CPC title

• H04L41/16

  using machine learning or artificial intelligence · CPC title

• H04L47/41

  by acting on aggregated flows or links · CPC title