A Performance and Power Characterization study of Memoization and Tabulation methods in Graph Neural Networks by assessing Dynamic Programming Workloads

Machine learning solutions for daily life using the ubiquitous computing environment require new efforts to improve software designs' energy efficiency without affecting performance. Recent advances in graph neural networks have become a game-changing method in machine learning in Ubiq-uitous systems. Graph neural network considers being good in certain Ubiquitous systems learning use-cases to solve reasoning jobs. At times, graph neural networks are related to align with dynamic programming approach, a frequently used method to solve problems observable from nature. Dynamic programming is a flexible way to handle several sequential choices in uncertain conditions. This method aims to quantify acceptable rules for resolution that decide the best possible decision. This paper presents a research direction to explore the possibility of adopting Memoization and Tabulation methods used in Dynamic Programming Workloads for optimizing per-formance and power consumption of ubiquitous computation in graph neural networks. As a proof of concept, we perform experiments on selected dynamic programming problems where the computation resembles graph neural network. The findings shows the practical feasibility of adopting dynamic program-ming improvement methods with algorithmic optimization in graph neural networks.