Q-learning based asynchronous Boolean control networks stabilization with data loss.

Area of Science:

• Control Theory
• Network Science
• Computational Biology

Background:

• Real-world networks frequently experience data loss and asynchronous updates, posing challenges for system stability.
• Asynchronous Boolean networks (ABNs) are critical models for biological systems, but their analysis is complicated by data loss and asynchronicity.

Purpose of the Study:

• To investigate and achieve the stabilization of asynchronous Boolean networks (ABNs) in the presence of data loss and asynchronous updates.
• To develop a robust control strategy combining Q-learning and flip control for ABN stabilization.

Main Methods:

• Algebraic formulation of ABNs using the semi-tensor product (STP) tool.
• Construction and equivalence proof of an augmented system for theoretical analysis.
• Development of a stabilization criterion for ABNs under flip control.
• Application of Q-learning to design optimal flip sequences for stabilization.

Main Results:

• An algebraic framework was established for analyzing ABNs with data loss.
• A stabilization criterion was derived for ABNs subjected to flip control.
• Q-learning effectively designed flip sequences to achieve system stabilization.
• The proposed control strategies were validated through two biological network examples.

Conclusions:

• The combination of Q-learning and flip control provides an effective approach for stabilizing asynchronous Boolean networks with data loss.
• The developed methods offer a robust solution for managing data loss and asynchronous updates in complex network systems.
• The findings have significant implications for the control and analysis of biological networks and other real-world systems.