Telefonaktiebolaget Lm Ericsson (publ)

Tobias Ley

Martha Vlachou-Konchylaki

Michael Meyer

Daniele Ceccarelli

Vincent Huang

Uwe Horn

Massimiliano Maggiari

Andreas Schieder

Wenfeng Hu

Roberta Daidone

Wei Huang

A pre-training apparatus and method for reinforcement learning based on a Generative Adversarial Network (GAN) is provided. GAN includes a generator and a discriminator. The method comprising receiving training data from a real environment where the training data includes a data slice corresponding to a first state-reward pair and a first state-action pair, training the GAN using the training data, training a relations network to extract a latent relationship of the first state-action pair with the first state-reward pair in a reinforcement learning context, causing the generator trained with training data to generate first synthetic data, processing a portion of the first synthetic data in the relations network to generate a resulting data slice, merging the second state-action pair portion of the first synthetic data with the second state-reward pair from the relations network to generate second synthetic data to update a policy for interaction with the real environment.

Pre-training system for self-learning agent in virtualized environment

A method in a telecommunications network comprises acquiring values of one or more parameters relating to traffic flow between a first group of nodes in the network. The method further comprises using a first reinforcement learning agent to dynamically adjust a first routing metric used to route traffic through the first group of nodes, based on the values of the one or more parameters, so as to alter the traffic flow through the first group of nodes.

Methods in a telecommunications network

A method for training a Software Agent to control a communication network is disclosed. The method comprises initialising a training network slice within the communication network, instantiating within the training network slice a cloned version of at least one Virtualised Network Function (VNF) comprised within a production network slice, mirroring traffic incoming to the VNFs of the production network slice and forwarding the mirrored traffic to the training network slice. The method further comprises causing a training instance of the Software Agent to execute a Reinforcement Learning algorithm on the training network slice, and transferring knowledge acquired by the training instance of the Software Agent to a production instance of the Software Agent. Also disclosed are an apparatus and a computer program configured to carry out methods for training a Software Agent to control a communication network.

Training a software agent to control a communication network

In a communication system (SYS) in which a packet data transfer is performed between the network side (NS) and the subscriber terminal side (SS) a physical connection (LC) is maintained during the data packet transfer. An active period detector (AP-DET) detects an active period of data packet generation by evaluating the inter-arrival time (TDIFF) between successively arriving data packets (DP). If the inter-arrival time (TDIFF) falls within a predetermined range a physical connection maintaining device (LC-MAIN) maintains the physical connection (LC) even if a transmitter queue (TR-QUE) temporarily becomes empty. Therefore, within the active period (AP) the transmitter (NS-TR; SS-TR) indicates a non-empty queue. The invention finds particular application in the GPRS/(E)GPRS/GSM environment.

Subscriber terminal, network controller and communication system for performing packet data transfer with reduced delay

Growing community of inventors

Täby, Sweden

Tobias Ley