About

I am a Ph.D. candidate in Computer Science at the New Jersey Institute of Technology (NJIT), with 4+ years of research experience in machine learning and artificial intelligence. My work spans federated learning, large language models, LLM fine-tuning, and combinatorial optimization algorithms. I have published at top-tier venues including ICLR and AAAI.

Education

New Jersey Institute of Technology (NJIT) — Newark, NJ
Ph.D. in Computer Science
Sep 2021 – Present
Khulna University of Engineering & Technology — Bangladesh
B.Sc. in Computer Science and Engineering
Mar 2016 – Mar 2020

Selected Publications

  1. [1]
    FedDAG: Clustered Federated Learning via Global Data and Gradient Integration for Heterogeneous Environments
    A. Pramanik, M. Kantarcioglu, V. Oria, S. Sharma
    ICLR 2026 International Conference on Learning Representations, 2026
    📄 PDF
  2. [2]
    Spark: Smart Building Fire Prediction and Risk Analysis
    A. Pramanik, M. Kumar, N. Panwar, L. V. Potts, J. Shah, S. Sharma
    IEEE NCA 2025 IEEE International Symposium on Network Computing and Applications, 2025
    📄 PDF
  3. [3]
    Equity Promotion in Public Transportation
    A. Pramanik, P. Xu, Y. Xu
    AAAI 2023 AAAI Conference on Artificial Intelligence, 2023
    📄 PDF

Selected Projects

NL2SQL Using LLMs

Text-to-SQL system resolving database and query ambiguity by decomposing SQL generation into subtasks. Deployed on GPU-based HPC with open-source LLMs (Qwen-7B, OSS-20B, OSS-120B) via Hugging Face Transformers; uses custom prompt engineering and model fine-tuning.

Federated Learning in Heterogeneous Environments

Clustered FL pipeline in PyTorch with non-IID data; incorporates SVD-based feature extraction, entropy-optimized MLP, dual-encoder with cross-cluster representation sharing, hierarchical clustering with adaptive thresholds, and a novel federated-aware evaluation metric.

Transportation Equity Optimization

Budget-allocation optimizer using LP-based rounding and bipartite matching to maximize equity across 5 demographic groups in Chicago; integrates ride-hailing and bus-line programs (17,875 households, 649 bus stops, 20 candidate routes).

Smart Building Fire Risk Analytics

Sensor-driven graph-diffusion framework combining static, dynamic (drone), and human sensor streams with weighted multimodal fusion; builds room adjacency graphs modeling structural and HVAC connectivity; applies k-means for real-time risk-zone classification.