Abstract

To minimize the effect on the network as much as possible, the nodes should coordinate their responses across a wide number of nodes based on incorrect indications of penetration. Most sophisticated assaults may take months to be developed and show no visible signs of development before being carried out. The sequential choice issue is difficult to solve due to the vast observation and action areas and the lengthy time horizon involved, as illustrated in Table 1. In this paper, we propose approaches for scaling deep reinforcement learning in order to address the challenge of orchestration of industrial control networks for cyber security. This neural architecture utilizes the principles of attention and size complexity and is only reliant on the size of the protected network. An early exploration training program is implemented to relieve the difficulties early exploration brings. The findings of the trials suggest that the suggested techniques have a better chance of converging on effective policies than baseline methods, as far as learning sample complexity and complexity of policies are concerned.