Historical data transfer to agent in a multi-agent system for an action controlling performance of the multi-agent system

Patent metadata

Field	Value
Publication number	US-12425289-B2
Application number	US-202118282019-A
Country	US
Kind code	B2
Filing date	Mar 15, 2021
Priority date	Mar 15, 2021
Publication date	Sep 23, 2025
Grant date	Sep 23, 2025

Abstract

A computer-implemented method performed in a multi-agent system by a first network node is provided for transferring historical data from an operating agent to a second agent for an action controlling a performance of the multi-agent system. The method includes selecting at least one operating agent for transfer of historical data to the second agent. The historical data acquired from executions of an action by the at least one operating agent that at least partially fulfills an input parameter. The selecting is based on one or more criteria including (i) a performance of the at least one operating agent on the parameter or on a related parameter; (ii) an availability of the at least one operating agent; and (iii) an identity of an actuation target system for receipt of the action. The method further includes transferring the historical data to the second agent.

First claim

The invention claimed is:

1. A computer-implemented method performed in a multi-agent system by a first network node for transferring historical data from an operating agent to a second agent for an action controlling a performance of the multi-agent system, the method comprising:
selecting at least one operating agent from a plurality of operating agents for a transfer of historical data from the at least one operating agent in the system to the second agent, the historical data acquired from a plurality of executions of the action by the at least one operating agent that at least partially fulfills a parameter input to the at least one operating agent, and the selecting based on one or more criteria including (i) a performance of the at least one operating agent on the parameter or on the related parameter; (i) an availability of the at least one operating agent for the selection;
and (ii) an identity of an actuation target system for receipt of the action, wherein the multi-agent system comprises a mobile network;
performing a relatedness measurement on the parameter and the related parameter to obtain an index, the relatedness measurement comprising a cosine similarity or a Pearson correlation;
wherein the selecting comprises selection of the at least one operating agent when (i) the index is lower than a defined threshold, and (ii) the performance of the at least one operating agent meets an evaluation level; and
transferring the historical data from the selected at least one operating agent to the second agent.

2. The method of claim 1 , wherein the plurality of operating agents learn and make decisions, respectively on the action by training and executing neural networks.

3. The method of claim 1 , wherein the parameter and the related parameter each comprise a vector of quantifiable values for qualities of the multi-agent system that a decided action either operates within or targets to achieve directly or indirectly.

4. The method of any of claim 1 , further comprising:
performing a relatedness measurement on the parameter and the related parameter to obtain an index, the relatedness measurement comprising a cosine similarity or a Pearson correlation; and
wherein the selecting comprises selection of the at least one operating agent when (i) the index is lower than a defined threshold, and (ii) the performance of the at least one operating agent meets an evaluation level.

5. The method of claim 1 , wherein the availability of the at least one operating agent for selection comprises one of: the at least one operating agent is online and is available, and the at least one operating agent has predicted future availability.

6. The method of claim 1 , wherein the identity of the actuation target system comprises at least a portion of a system communicating with the second agent.

7. The method of claim 1 , further comprising:
providing identifying information about the selected at least one operating agent to the second agent, wherein the transferring is based on a reinforcement learning process of the second agent that balances exploration with exploitation by using an epsilon-greedy policy to retrieve an action from the selected at least one operating agent until performance of the second agent is verified by the at least one operating agent;
receiving a request from the second agent to authorize the second agent subsequent to the verification of the second agent; and
responsive to the request, authorizing the second agent.

8. The method of claim 1 , wherein the transferring comprises a transfer of weights from a neural network of the at least one operating agent to a neural network of the second agent.

9. The method of claim 1 , wherein the transferring comprises:
generating a set of training data comprising a selected action from the plurality of operating agents for each input; and
providing the set of training data to the second agent, the second agent set as a best performing agent from the selected at least one operating agent.

10. The method of claim 1 , further comprising:
subsequent to the transferring, isolating the second agent until training based on the transferred historical data is complete.

11. The method of claim 10 , wherein the isolating comprises providing a local copy of the inputs to an isolated copy of the second agent and isolated copy of the second agent operating on the local copy of the inputs to generate the action, and monitoring the isolated copy of the second agent for access security and lineage tracking.

12. The method of claim 1 , wherein the first network node and the second network node each comprise a network node in the multi-agent system, and the actuation target system comprises a portion of the mobile network.

13. The method of claim 12 , wherein the at least one parameter comprises a defined latency and a defined availability, wherein the at least one operating agent is an operating agent of a mission critical application assigned to a network slice, wherein the second agent is an agent of a recently deployed second mission critical application assigned to a second network slice, and the action is controlling resource allocation across network slice or the second network slice.

14. A computer-implemented method performed by a second network node in a multi-agent system for transferring historical data from an operating agent to a second agent for an action controlling a performance of the multi-agent system, the method comprising:
registering the second agent with a first network node in the multi-agent system, the registering indicating a presence or availability of the second agent for onboarding to the multi-agent system;
responsive to the registering, receiving from the first network node a selected at least one operating agent from a plurality of operating agents for a transfer of historical data from the at least one operating agent in the multi-agent system to the second agent, the historical data acquired from a plurality of executions of an action by the at least one operating agent that at least partially fulfills a parameter input to the at least one operating agent, and the selecting based on one or more criteria including (i) a performance of the at least one operating agent on the parameter or on a related parameter; (ii) an availability of the at least one operating agent for the selection;
and (iii) an identity of an actuation target system for receipt of the action, wherein the multi-agent system comprises a mobile network;
performing a relatedness measurement on the parameter and the related parameter to obtain an index, the relatedness measurement comprising a cosine similarity or a Pearson correlation;
wherein the selecting comprises selection of the at least one operating agent when (i) the index is lower than a defined threshold, and (ii) the performance of the at least one operating agent meets an evaluation level;
and receiving a transfer of the historical data from the selected at least one operating agent.

15. The method of claim 14 , wherein the parameter and the related parameter each comprise a vector of quantifiable values for qualities of the multi-agent system that a decided action either targets to achieve directly or indirectly or to operate within.

16. The method of claim 14 , wherein the related parameter is based on the first network node performing a relatedness measurement on the parameter and the related parameter to obtain an index, the relatedness measurement comprising a cosine similarity or other measure of similarity; and wherein the selected a least one operating agent is selected by the first

Assignees

• Ericsson Telefon Ab L M

Inventors

• Karapantelakis Athanasios
• Teslenko Maxim
• Baktir Ahmet Cihat
• Orlic Marin
• Mokrushin Leonid
• Daoutis Marios
• Vulgarakis Feljan Aneta
• Nikou Alexandros
• Eleftheriadis Lackis

Classifications

• H04L41/5009

  Determining service level performance parameters or violations of service level contracts, e.g. violations of agreed response time or mean time between failures [MTBF] · CPC title

• H04L41/16

  using machine learning or artificial intelligence · CPC title

• G06N3/092

  Reinforcement learning · CPC title

• G06N3/096

  Transfer learning · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title