On Reinforcement Learning in Stabilizability of Probabilistic Boolean Control Networks

This paper examines the stabilizability probability of probabilistic Boolean control networks (PBCNs) by utilizing reinforcement learning. The bounds of the reachability probability are formulated based on the algebraic state space representation (ASSR) of PBCNs, which leads to the stabilizability criterion. Furthermore, the equivalence between the stabilizability probability and the optimal state-value function in the form of an iteration equation is proved. The Q-learning technique is implemented to estimate the stabilizability probability and obtain the corresponding optimal control sequences. Theoretical results are demonstrated through an apoptosis network to elaborate on the findings.