Mohammad G. Khoshkholgh

Mohammad G. Khoshkholgh, Halim Yanikomeroglu

We address how to exploit power control data, gathered from a monitored environment, for performing power control in an unexplored environment. We adopt offline deep reinforcement learning, whereby the agent learns the policy to produce the transmission powers solely by using the data. Experiments demonstrate that despite discrepancies between the monitored and unexplored environments, the agent successfully learns the power control very quickly, even if the objective functions in the monitored and unexplored environments are dissimilar. About one third of the collected data is sufficient to be of high-quality and the rest can be from any sub-optimal algorithm.

Mohammad G. Khoshkholgh, Halim Yanikomeroglu

Policy gradient reinforcement learning techniques enable an agent to directly learn an optimal action policy through the interactions with the environment. Nevertheless, despite its advantages, it sometimes suffers from slow convergence speed. Inspired by human decision making approach, we work toward enhancing its convergence speed by augmenting the agent to memorize and use the recently learned policies. We apply our method to the trust-region policy optimization (TRPO), primarily developed for locomotion tasks, and propose faded-experience (FE) TRPO. To substantiate its effectiveness, we adopt it to learn continuous power control in an interference channel when only noisy location information of devices is available. Results indicate that with FE-TRPO it is possible to almost double the learning speed compared to TRPO. Importantly, our method neither increases the learning complexity nor imposes performance loss.