**5. Conclusions**

This chapter introduces a dynamic control decision-making algorithm, inspired by Lyapunov optimization theory under the situation where the tradeoff between utility/performance and delays exists. Thus, the dynamic decision-making algorithms aim at time-average utility maximization (or penalty minimization) in real-time deep learning platforms. As discussed, the Lyapunov optimization-based algorithms are scalable, hardware/system-independent, self-configurable, and lowcomplexity. Thus, it can be used in various emerging applications such as video streaming, wireless networks, security applications, and smart grid applications.
