Towards intelligent caching in NDN networks

Abstract

Can we develop intelligent strategies specific to NDN networks that outperform traditional approaches and optimize the performance of these NDN networks? This question guided the course of this dissertation, driven by the inherent limits of the present IP-based Internet paradigm and the rising shift toward data-centric architectures. The significance of this question derives from the crucial role caching plays in improving latency, bandwidth utilization, and scalability in NDN, and the inability of traditional caching techniques to adapt to dynamic user behavior. This research tested and confirmed several hypotheses: (1) reinforcement learning methods can dynamically outperform fixed cache replacement strategies, and (2) combining spatial and temporal learning components—specifically CNNs and LSTMs—improves the decision-making capability of RL-based caching models