COMPUTER-IMPLEMENTED TRAINING OF A POLICY MODEL FOR SPECIFYING A CONFIGURABLE PARAMETER OF A TELECOMMUNICATIONS NETWORK, SUCH AS AN ANTENNA ELEVATION DEGREE OF A NETWORK NODE, BY SMOOTHED-LOSS INVERSE PROPENSITY

摘要

A computer implemented method for training a policy for operating a telecommunications network includes providing (602) a baseline dataset of performance indicator data for the telecommunications network, generating (612) a policy model that specifies actions to be taken on a configurable parameter of the telecommunications network given a context of the telecommunications network, generating (606) a loss model that estimates an expected loss experienced for execution in the telecommunications network of at least one action from a plurality of actions on the configurable parameter, training (608) the loss model to generate a trained loss model having a level of reduced noise, and performing (630) inverse propensity score learning on the policy model using the trained loss model to obtain a trained policy model. A method performed by a computer system for controlling an antenna elevation degree of an antenna of a network node in a telecommunications network is also provided.