Summary: In this paper, a model of a recursive neural network, able to process directed acyclic graphs with labeled edges, is introduced. Input to the classical neural networks is usually represented as vector of data, it means that in the presented information are missed some relations among parts of input. The applied and presented method assumes a graph’s based representation of objects that combines basic data and relation data. Such graph representations of data are then processed by the recursive model of neural network in order to determine the eventual presence and the position of objects inside the space. The proposed method is general and can be applied in different object detection systems. The developed model is tested on some games.
Reviewer: Igor Černák (Ružomberok)