Intelligentes Agentensystem

Details

Veröffentlichungs-Nr.

EP4764961

Anmeldetag

18. Dezember 2024

Veröffentlichung

24. Juni 2026

Rechtsraum

EP

IPC

G06N3/092, G06N20/00

Abstract

A computer-implemented method of training an intelligent agent system for influencing the operation of a physical system to achieve an objective relating to an intended state of the physical system is described. The method comprises performing a respective reinforcement learning process for an intelligent agent, the reinforcement learning process being conducted within an environment that provides a representation of the physical system, the agent being trained to achieve the objective within the environment by carrying out one or more actions selected from a respective set of available actions, the training being based on a reward that is determined by the reinforcement learning process for each of the training steps. The reinforcement learning process comprises providing, to the agent, a set of observations of the environment of that agent, the agent being configured to select the one or more actions in dependence on the set of observations to achieve the objective, the set of observations being provided in a data structure in which the set of observations are arranged in an order which is dependent on a starting position of the agent and/or a position relating to the objective, within the environment.

Anmelder

Firma

British Telecommunications Public Limited Company

Land

🇬🇧 Großbritannien

Fachgebiete