Executive Summary : | The centrality of a node in a network is a crucial measure of its importance, used for various applications such as identifying key players in social networks, finding critical nodes in transportation networks, or detecting essential genes in biological networks. The positive notion of importance denotes dominance, power, control, influence, while the negative notion denotes criticalness, proneness to failure, load or tension on nodes, and exposure to infection. Centrality scores depend on the network structure, and adding missing links can increase or decrease centrality scores. This problem arises in flow networks where the centrality of a node denotes the load on the node, which can cause it to fail. The objective is to identify and add missing links in the network to optimize the load on the considered node, reducing the chance of failure. The brute force method of adding links is time-consuming, so deep learning frameworks are proposed to solve this problem. The project aims to solve an optimization variant of the link prediction problem, first deterministically solving it for three centrality measures (closeness, betweenness, and stress centrality) on particular classes of networks and later using deep learning for general networks. Machine learning and deep learning frameworks have been successfully applied to link prediction in various domains, including social networks, recommendation systems, and biological networks. Solving the problem has critical applications in many fields, such as maximizing information spread in social networks by adding new connections between existing users and minimizing average traffic flow across the network by adding more routes. |