Pranjal Naman

I'm a Ph.D. student at the Indian Institute of Science where I work on scalable systems for graph neural networks and abstractions and platforms for training/inference on large, dynamic, and distributed graphs.

Email  /  CV  /  Scholar  /  Github  /  LinkedIn  /  ORCID

News

Citations

Research

I'm interested in scalable systems for graph neural networks: out-of-core and distributed GNN inference on billion-scale graphs, incremental inference on streaming graphs, federated graph learning, and hardware-aware distributed training.

Traffic management in emerging megacities requires real-time analytics over thousands of CCTV video streams under latency, bandwidth, compute, and energy constraints.

Accelerates out-of-core GNN inference on billion-scale graphs beyond what fits in GPU or host memory.

An out-of-core inference engine that scales GNN inference to billion-node graphs beyond GPU and host memory limits.

Scalable algorithms for computing centrality measures over billion-scale, time-evolving financial transaction networks.

An edge-cloud system for real-time traffic analytics over city-scale camera networks.

Reduces communication overhead in federated GNN training by optimizing how remote node embeddings are exchanged across clients.

A scalable data management and anomaly detection system deployed on billion-scale transaction data at India's National Payments Corporation (NPCI).

An incremental inference framework that updates GNN predictions as graphs evolve, without full recomputation.

Propagates the effects of graph updates incrementally to keep GNN inference results fresh on large streaming graphs, avoiding full recomputation.

Rethinks distributed and out-of-core GNN training strategies to better exploit GPU resources at scale.

Techniques for mining temporal motifs over large-scale financial transaction networks.

Reduces communication cost in federated GNN training by optimizing the exchange of remote node embeddings across clients.

An early study of how hardware and neighbor-sampling configurations affect GNN inference performance.

Uses graph topology to guide aggregation in federated GNN training, improving convergence over topology-agnostic baselines.

A comparative evaluation of strategies for training GNNs in federated settings.

Models the performance characteristics of GNN training and inference across hardware configurations.

Examines vertex-centric versus alternative programming models for distributed GNN training.

Education & Experience

• Ph.D. Computational & Data Sciences, Indian Institute of Science (2022–Present)
• B.E. Instrumentation & Control Engineering, Netaji Subhas Institute of Technology, University of Delhi (2016–2020)
• Software Engineer, Soroco, Bengaluru, India (Sep 2020–Jul 2021)
• Software Engineer Intern, Nucleus Software, Noida, India (May 2019–Jul 2019)

Awards

• Winner, CCGrid TCSC SCALE Challenge, CCGrid 2026, Sydney, Australia
• SIGMOD/PODS India Fellowship to attend SIGMOD/PODS 2026, Bengaluru, India
• Best Student Research Poster and Lightning Talk, HiPC 2025, India
• ACM India-IARCS Travel Grant to present work at ICDCS 2025, Glasgow
• Young Researcher, Pingala Interactions in Computing, 2025
• HiPC Student Research Symposium Travel Grant, 2023, 2024, and 2025
• Euro-Par PhD Studentship to present work at Euro-Par 2024, Spain
• Full Tuition Waiver for top 5% students, all 4 years (2016–20), B.E. Instrumentation & Control Engg., NSIT Delhi

Academic Service

• Reviewer, 14th ACM IKDD CODS, 2026
• Lead Volunteer, ACM SIGMOD/PODS Conference, 2026, Bengaluru, India
• Reviewer, Applied Data Science Track, 13th ACM IKDD CODS, 2025
• Shadow Program Committee Member, ACM SIGMETRICS, 2026
• Volunteer, 2023 IEEE/ACM 23rd CCGridW

Teaching

Teaching Assistant, DS256: Scalable Systems for Data Science, Indian Institute of Science (Jan 2026–May 2026)

Design based on Jon Barron's personal website.