Publications — Md Nasiat Hasan Fahim

Publications

Peer-reviewed work is listed separately from manuscripts that have not yet completed peer review.

Peer-reviewed publications

Published

IEEE QPAIN
2026

Protein-Based Fish Species Identification: Dataset, Models, and Insights from Native Bangladeshi Fish

N. H. Fahim, A. U. Muhib and M. S. Rahman. 2026 IEEE 2nd International Conference on Quantum Photonics, Artificial Intelligence, and Networking (QPAIN), Chittagong, Bangladesh, 2026, pp. 1–6.

Contribution: Curated a study dataset of 2,845 protein sequences spanning nine native Bangladeshi fish species and compared seven architectures under a shared evaluation.
Result: A motif-aware CNN-Transformer with type-aware positional encoding reached 79.80% accuracy with CPU-capable inference, running 5× faster and using a model 42× smaller than the ProtBERT baseline.
Why it matters: The work examines species identification in a resource-constrained setting where deployment cost is part of the research problem.

DOI: 10.1109/QPAIN69676.2026.11546620

Published

IEEE ICCIT
2025

Real-Time Earthquake Magnitude Classification from Initial P-Waves: Models, Dataset, and Comparative Analysis for South Asia

M. N. Hasan Fahim, M. A. Ullah Muhib, R. A. Tanvir and A. Al Noman. 2025 28th International Conference on Computer and Information Technology (ICCIT), Cox’s Bazar, Bangladesh, 2025, pp. 5175–5180.

Contribution: Curated 7,318 South Asian earthquake events for single-station early-warning experiments and compared six model families using seven-second post-P-wave windows.
Result: The Transformer reached 76.23% accuracy, 81.56% adaptive accuracy, and 4.8 ms inference latency under the study protocol.
Why it matters: The seven-second single-station setup reflects the limited information available in practical early-warning settings.

DOI: 10.1109/ICCIT68739.2025.11489542

Manuscripts

Submitted, not yet peer-reviewed

Submitted

PLOS ONE
2026

Parameter-Efficient Preference Optimization and Supervised Fine-Tuning of ProGen2 for Thermophile Sequence Space

Md Nasiat Hasan Fahim, Md. Abid Ullah Muhib and Mohammad Shahidur Rahman. Submitted to PLOS ONE, 2026.

Contribution: Adapts ProGen2 toward thermophile-associated sequence space through LoRA-based supervised fine-tuning, DPO, and KTO under controlled compute and homology-controlled evaluation.
Result: The experiments reveal distinct likelihood-preference trade-offs across the adaptation objectives.

Submitted

SUST Journal of
Science and Technology
2026

Spatiotemporal Crime Dynamics and Recidivism Analysis in Urban Bangladesh: A Hybrid K-Means Approach

Ahsan Habib, Md. Mehedi Hasan, Rayhanul Amin Tanvir, Abu Sayeid Sawon, Md Nasiat Hasan Fahim and Nixon Deb Antu. Submitted to SUST Journal of Science and Technology, 2026.

Contribution: Combines newspaper reporting and police records to analyse Sylhet crime patterns using spatiotemporal clustering, recidivism analysis, forecasting, and 3D visualisation.
Result: The study identifies temporal hotspots and near-term patterns relevant to local public-safety analysis.