Multiplex Networks
Extending network measures and methods to Multiplex networks and Hypergraphs
Key contributors: Tarun Kumar
Domain: Multiplex Network Analysis, Hypergraph modeling
Description: Most traditional classification techniques, whether fully supervised or semi-supervised, treat data instances independently and train classifiers using their attribute values alone. But many a time real world data can be represented naturally in the form of networks and we often have access to multiple sets of attributes describing the same data, multiple relations between data instances and sometimes super-dyadic relations between them. This motivates us to use multilayer graphs for representing the information from multiple relations and attribute views, and hypergraphs, a generalization of simple graphs, for representing super-dyadic relations.
Suggested Readings:
Disease Module Identification
Key contributors: Beethika Tripathi
Domain: Multiplex Network Analysis, Deep Learning
Description: In the field of biology, complex network analysis have a special importance as exhaustive characterizing of genes or proteins through biological experiments is intractable. So leveraging the power of computation and available knowledge several hypotheses can be made to guide the experiments. The basic assumption follows "guilt by association" principle where genes or proteins that are co-localized or have similar topological roles are functionally correlated allowing us to infer properties of unknown genes. But inferencing from the biological network using traditional clustering algorithms for extracting biologically meaningful functional modules is tough due to the presence of noisy false positive interactions. Moreover, there are heterogeneous sources of information available so the challenge lies in developing integrative methods which can take advantage of the topology of the multiple heterogeneous networks available. These help in providing stronger confidence on the predictions made for unknown genes. We have proposed several methods and heuristics which gives significant improvement in performance.
Suggested Readings:
Representation learning for Heterogeneous Networks and Multiplex Networks
Key contributors: Ujjawal Soni
Domain: Network Representation Learning, Heterogeneous Networks, Multiplex Networks
Description: Many real-world networks consist of multiple types of entities (nodes) and multiple types of relationships (edges) between those entities and therefore are better modeled as heterogeneous networks or multiplex networks. With the recent advances in representation learning and deep learning, there are techniques like DeepWalk/node2vec to learn node embeddings in a homogeneous network, but because of the advance complexities due to existence of multiple types of nodes and edges, learning representations in heterogeneous or multiplex networks in a non-trivial task and is still in its nescent stage. In this work, we aim to tackle these challenges and learn representations which capture the semantic and structural properties of such complex networks.
Suggested Readings: