Risk-Sensitive Rate Correcting for Dynamic Heterogeneous Networks: Autonomy and Resilience
Khanh Dai Pham · 2020
This work is proposing a unified framework for dynamic and heterogeneous networks from a control engineering standpoint. Adaptive learning and risk-sensitive policies illustrate ways of dealing with: i) state-space realization to account for heterogeneous flow rates, bottleneck links with rate allocating routers or switches, and coupling between queue regulation, stochastic and time lag effects; ii) centralized rate-based stabilization with explicit congestion indication feedback via partially noisy observations; iii) performance risk mitigation enabled by Minimal-Cost-Variance control theory; and iv) distributed adaptation based on companion window flow control mechanisms at sources. The expected results will help to reveal new insights on systematic designs and rigorous analysis of adaptive and scalable learning and management schemes for coordinated resilience across all layers and elements in networked systems.