Learning Dynamic Connectivities and Signal Recovery over Graphs
Original version
Ramezani-Mayiami, M. (2023). Learning Dynamic Connectivities and Signal Recovery over Graphs [Doctoral dissertation]. University of Agder.Abstract
The ever increasing rate of acquisition of real-world data sets including customer consumption data, social networks activities, financial data, temperature data from different regions, and brain-computer interface measurements results in rapidly growing data volumes. Collecting, transmitting, storing, retrieving, and processing these huge data volumes are challenging because of the need for high computational resources and data storage capacity. But the most important task, and the reason why the data was collected in the first place, is data analysis: Finding correlations, patterns and connections, aggregating to higher levels, and finally extracting useful information and knowledge. Recently, the Graph Signal Processing (GSP) framework has simplified the analysis of large data volumes by the use of graph theory, where graph vertices represent the components of the data network of interest. Thus, different applications are involved with graphs capturing the underlying topology among different entities of the network. Most of the research projects in this framework try to manipulate the ”classical” signal processing concepts and make a ”graphical” version by migrating from one sole entity to the network of entities. The results were promising but there are still some spaces for this research framework for more real-world scenarios and applications.