Sanjukta Bhowmick

S. M. Shovan, Arindam Khanda, Sanjukta Bhowmick et al.

Higher-order interactions beyond pairwise relationships in large complex networks are often modeled as hypergraphs. Analyzing hypergraph properties such as triad counts is essential, as hypergraphs can reveal intricate group interaction patterns that conventional graphs fail to capture. In real-world scenarios, these networks are often large and dynamic, introducing significant computational challenges. Due to the absence of specialized software packages and data structures, the analysis of large dynamic hypergraphs remains largely unexplored. Motivated by this gap, we propose ESCHER, a GPU-centric parallel data structure for Efficient and Scalable Hypergraph Evolution Representation, designed to manage large scale hypergraph dynamics efficiently. We also design a hypergraph triad-count update framework that minimizes redundant computation while fully leveraging the capabilities of ESCHER for dynamic operations. We validate the efficacy of our approach across multiple categories of hypergraph triad counting, including hyperedge-based, incident-vertex-based, and temporal triads. Empirical results on both large real-world and synthetic datasets demonstrate that our proposed method outperforms existing state-of-the-art methods, achieving speedups of up to 104.5x, 473.7x, and 112.5x for hyperedge-based, incident-vertex-based, and temporal triad types, respectively.

Turja Kundu, Sanjukta Bhowmick

Graph neural networks (GNNs) excel on homophilic graphs where connected nodes share labels, but struggle with heterophilic graphs where edges do not imply similarity. Moreover, iterative message passing limits scalability due to neighborhood expansion overhead. We introduce ATLAS (Adaptive Topology-based Learning at Scale), a propagation-free framework that encodes graph structure through multi-resolution community features rather than message passing. We first prove that community refinement involves a fundamental trade-off: finer partitions increase label-community mutual information but also increase entropy. We formalize when refinement improves normalized mutual information, explaining why intermediate granularities are often most predictive. ATLAS employs modularity-guided adaptive search to automatically identify informative community scales, which are one-hot encoded, projected into learnable embeddings, and concatenated with node attributes for MLP classification. This enables standard mini-batch training and adjacency-free inference after one-time preprocessing. Across 13 benchmarks including million-node graphs, ATLAS achieves competitive or superior accuracy, up to 20-point gains over GCN on heterophilic datasets and 12-point gains over MLPs on homophilic graphs. By treating topology as explicit features, ATLAS adapts intelligently: leveraging structure when informative, remaining robust when weakly aligned, and avoiding propagation when structure misleads, providing both scalable performance and interpretable structural insights.

Abhishek Santra, Kanthi Komar, Sanjukta Bhowmick et al.

Any large complex data analysis to infer or discover meaningful information/knowledge involves the following steps (in addition to data collection, cleaning, preparing the data for analysis such as attribute elimination): i) Modeling the data -- an approach for modeling and deriving a data representation for analysis using that approach, ii) translating analysis objectives into computations on the model generated; this can be as simple as a single computation (e.g., community detection) or may involve a sequence of operations (e.g., pair-wise community detection over multiple networks) using expressions based on the model, iii) computation of the expressions generated -- efficiency and scalability come into picture here, and iv) drill-down of results to interpret or understand them clearly. Beyond this, it is also meaningful to visualize results for easier understanding. Covid-19 visualization dashboard presented in this paper is an example of this. This paper covers all of the above steps of data analysis life cycle using a data representation that is gaining importance for multi-entity, multi-feature data sets - Multilayer Networks. We use several data sets to establish the effectiveness of modeling using MLNs and analyze them using the proposed decoupling approach. For coverage, we use different types of MLNs for modeling, and community and centrality computations for analysis. The data sets used - US commercial airlines, IMDb, DBLP, and Covid-19 data set. Our experimental analyses using the identified steps validate modeling, breadth of objectives that can be computed, and overall versatility of the life cycle approach. Correctness of results is verified, where possible, using independently available ground truth. We demonstrate drill-down that is afforded by this approach (due to structure and semantics preservation) for a better understanding and visualization of results.

Abhishek Santra, Sanjukta Bhowmick, Sharma Chakravarthy

Holistic analysis of many real-world problems are based on data collected from multiple sources contributing to some aspect of that problem. The word fusion has also been used in the literature for such problems involving disparate data types. Holistically understanding traffic patterns, causes of accidents, bombings, terrorist planning and many natural phenomenon such as storms, earthquakes fall into this category. Some may have real-time requirements and some may need to be analyzed after the fact (post-mortem or forensic analysis.) What is common for all these problems is that the amount and types of data associated with the event. Data may also be incomplete and trustworthiness of sources may also vary. Currently, manual and ad-hoc approaches are used in aggregating data in different ways for analyzing and understanding these problems. In this paper, we approach this problem in a novel way using multilayered networks. We identify features of a central event and propose a network layer for each feature. This approach allows us to study the effect of each feature independently and its impact on the event. We also establish that the proposed approach allows us to compose these features in arbitrary ways (without loss of information) to analyze their combined effect. Additionally, formulation of relationships (e.g., distance measure for a single feature instead of several at the same time) is simpler. Further, computations can be done once on each layer in this approach and reused for mixing and matching the features for aggregate impacts and "what if" scenarios to understand the problem holistically. This has been demonstrated by recreating the communities for the AND-Composed network by using the communities of the individual layers. We believe that techniques proposed here make an important contribution to the nascent yet fast growing area of data fusion.