Reward weight adjustment device, and reward weight adjustment method

Abstract

To provide a technique capable of adjusting KPI weights for tasks with no training data.SOLUTION: The reward weight adjustment device acquires multiple pieces of weight information which is a parameter set included in a predetermined reward function and environmental information corresponding to weight information and a piece of training data that is a sequence of actions including multiple time steps of actions provided by users in environmental information, and create a training dataset which is a set of a combination of weight information, environmental information, and training data. The reward weight adjustment device calculates the weighting information corresponding to the unknown environmental information based on the multiple training datasets and presents the estimated value to the user.SELECTED DRAWING: Figure 1

Images