Graph clustering detection as classification and evaluation using deep learning and force directed algorithm

Project ID
FG-570
Project Categories
Computer Science
Abstract

Graph clustering has many applications in real life and retrieved clusters are often assessed based on quality and accuracy. Measuring accuracy of graph clustering is difficult for real life examples as it requires ground truth clusters to be known whereas quality of a cluster is assessed using nodes and edges of a graph. There are many measures in the literature which suffers from one or more limitations. Thus it will be advantageous to have a new measure to evaluate graph clustering (community detection). Recently, deep learning has become very popular which has been applied successfully to many research domains. In this project, we have proposed a classification and evaluation technique for graph clustering with the help of deep learning and force-directed graph layout. We want to perform an extensive set of experiments to assess the effectiveness of our method using benchmark datasets generator. 

Use of FutureSystems
I will use GPU cluster for running experiments which are related to deep learning. My code is written in Pytorch which can take advantage of CUDA. A module in simulated data generation requires force directed algorithm which can run on multiple cores.
Scale of Use
I am intended to use 2 nodes in GPU cluster for experiment related to deep learning. I hope to use it carefully and sincerely so that others are not affected.