An Intelligent Memory Framework for Resource Constrained IoT Systems

Internet-of-Things systems are being widely used in diverse applications ranging from healthcare and retail to surveillance and smart agriculture. However the memory in these IoT applications can become a severe bottleneck particularly for resource-constrained edge systems. In this work we will discuss different existing Artificial Intelligence (AI) driven techniques that can lead to improved memory performance. Particularly we will discuss BINGO, a recently proposed reinforcement learning guided framework that can boost memory performance for application-specific edge systems by mimicking human brain properties such as data-awareness, forgetfulness, preferential access, and plasticity. We will highlight, through quantitative analysis, the effectiveness of the BINGO framework for a context-switching application that prioritizes different data over time. Finally, we will present different pathways for future research directions in this emerging area of research.