Samiur Rahman Khan

Samiur Rahman Khan

Machine Learning Researcher  ·  London, UK

I am a researcher working on machine learning, combinatorial optimisation, and explainable AI, with peer-reviewed publications cited 27 times and a top-ranked MSc thesis in biomedical gait pattern recognition. My research philosophy centres on root-cause analysis and unbiased analytics of the subject matter, with visual distribution used to preserve authenticity. I am currently targeting PhD research in efficient ML inference, neuromorphic computing, and high-performance computing.

Current: Submitted T=8 or Not T=8: Pareto-Optimal Timestep Selection for Energy-Efficient SNN Anomaly Detection to ICONS 2026. Applying for PhD positions in Germany, Sweden, the Netherlands, and Austria.
samiurk70@gmail.com github.com/samiurk70 linkedin.com/in/samiurk70 CV (PDF)
Research interests

What I work on

My research sits at the intersection of efficient machine learning and neuromorphic computing, with a methodological commitment to rigorous, reproducible evaluation across multiple seeds and datasets. Three threads run through recent work.

Energy-efficient inference
Spiking neural networks for anomaly detection, timestep-sensitivity Pareto analysis, and theoretical energy projections on neuromorphic substrates such as Intel Loihi. Empirical work on NSL-KDD, UNSW-NB15, thyroid, and cardiotocography benchmarks.
Neural combinatorial optimisation
Generalisation behaviour of attention-based solvers for the Travelling Salesman Problem at scales from TSP-20 to TSP-200, with progressive instance scaling and graph embedding normalisation as fine-tuning strategies.
LLM compression and explainability
Quantisation-aware knowledge distillation for reasoning benchmarks (HellaSwag, ARC, WinoGrande), and SHAP and LIME applied to ML systems in regulated domains.
Publications

Peer-reviewed research

Google Scholar: 27 citations h-index: 2 i10-index: 2
T=8 or Not T=8: Pareto-Optimal Timestep Selection for Energy-Efficient SNN Anomaly Detection
Khan, S. · ICONS 2026 (under review) submitted

Timestep-sensitivity Pareto frontier analysis for spiking autoencoders. Multi-seed evaluation across five seeds and four benchmark datasets, with honest energy labelling distinguishing measured GPU energy from theoretical neuromorphic projections.

Novel Identity Check Using W3C Standards and Hybrid Blockchain for Paperless Verification
Khan, S., Al-Amin, M. · IJIEEB Vol. 15 No. 4, 2023

DOI: 10.5815/ijieeb.2023.04.02

Towards a Secured Smart IoT Using Lightweight Blockchain for Pharmacy Products
Sohan, M., Khan, S. et al. · arXiv, 2022 13 citations
A Pragmatical Study on Blockchain-Empowered Decentralized Application Development Platform
Khan, S. et al. · ICCA 2020 (ACM) 14 citations

DOI: 10.1145/3377049.3377136

Selected projects

Research and engineering

Timestep-Sensitivity Pareto Analysis for SNNs

github
Python · PyTorch · SpikingJelly · FLOPs analysis · multi-seed evaluation

Demonstrated that SNN-based inference matches deep autoencoder F1-scores within 1.3% on real benchmark datasets (Thyroid, NSL-KDD) while achieving a 35 to 87x reduction in energy consumption. Includes T-sensitivity analysis across T ∈ {4, 8, 16, 32} and a Pareto-optimal selection methodology.

ReviewerMatch — Semantic Researcher Discovery

github
FastAPI · sentence-transformers · FAISS · Railway · Vercel · PostgreSQL

Production ML system indexing active ML and CS researchers from OpenAlex into a 384-dimensional semantic space. Returns ranked shortlists for peer review, collaboration, and supervisor search, reranked by similarity, h-index, and recency. Deployed full-stack with edge proxying and pre-computed embeddings.

Live demo →

GrantMatch — Funding Opportunity Retrieval

github
FastAPI · ML reranker · semantic search · 24,699 grants indexed

Parent project for ReviewerMatch. A neural information retrieval system for UK and EU research funding, ingesting from UKRI, Innovate UK, GOV.UK, and European Commission CORDIS. Matches project descriptions to active grants through semantic similarity with TRL, sector, and organisation-type filters.

Live demo →

Human Gait Analysis — Biomedical ML

github
Python · MySQL · Random Forest · Dynamic Time Warping · SHAP · LIME

Pattern-recognition pipeline on real pressure and temperature data from 312 patients, identifying gait abnormalities across four metatarsal regions. Abnormality index achieved 22% detection accuracy for patients requiring medical attention, against a 10% baseline. Awarded Top 5 MSc Data Science thesis at Middlesex University, 2025, graded Distinction.

Neural Combinatorial Optimisation Study

github
Python · PyTorch Geometric · TSPLib · GNN · Attention Model

Reproduced and extended Kool et al. across TSP-20 to TSP-200, quantifying non-linear generalisation degradation above a critical scale threshold. Proposed progressive instance scaling and graph embedding normalisation, closing the optimality gap by 2% without full retraining.

Quantisation-Aware Knowledge Distillation for LLM Reasoning

github
Python · HuggingFace · BitsAndBytes · GPTQ · DistilBERT

Joint QA-KD retained higher reasoning accuracy at 4-bit precision than sequential compression, with 1.67% gains on out-of-distribution benchmarks (HellaSwag, ARC, WinoGrande).

Experience

Professional trajectory

AI / ML Research Engineer
Aug 2025 — Present
Self-directed applied research · London, UK
  • Designed and benchmarked an SNN architecture in SpikingJelly achieving anomaly-detection F1 within 1.3% of ANN baselines, with 35 to 87x energy reduction.
  • Reproduced and extended the attention model for TSP across scales 20 to 200, proposing fine-tuning strategies that closed the optimality gap by 2% without retraining.
  • Investigated quantisation-aware knowledge distillation for LLM compression, with joint QA-KD outperforming sequential compression by 1.67% on reasoning benchmarks.
  • Conducting XAI research with Middlesex University academic staff using SHAP and LIME for model transparency in regulated domains.
Lecturer, Computer Science
Dec 2022 — Dec 2023
American International University Bangladesh · Dhaka
  • Delivered undergraduate modules in discrete mathematics, computer networks, and Cisco-certified courses to cohorts of 60 plus students.
  • Designed and launched a new PGD curriculum covering ICT, data science, and cybersecurity, adopted as a core departmental offering.
  • Supervised 8 final-year undergraduate research theses, with 2 students progressing to publication.
Technical Business Analyst
Feb 2022 — Aug 2022
Deepchain Labs · Dhaka
  • Led data-driven marketing analytics across five enterprise clients using HubSpot, Google Analytics, and Tableau, informing Web3 go-to-market strategy.
  • Conducted Python and SQL market research, reducing requirement-gathering cycles by approximately 30%.
Education

Academic background

MSc Data Science
Jan 2024 — Feb 2025
Middlesex University London

Awarded Distinction. Top 5 MSc thesis project across the Data Science cohort, 2025. Modules: Machine Learning, Visual Data Analysis, Applied Data Analytics, Ethics in Data Science.

MSc Computer Network and Architecture
Jan 2021 — Sep 2022
American International University Bangladesh

Summa Cum Laude, CGPA 3.98 / 4.00. Modules: Advanced Computer Networks, Cisco DevOps, Cyberops, Parallel Computation, System Security.

BSc Computer Science
Jan 2016 — Dec 2019
American International University Bangladesh

Best BSc Thesis Award at ICCA conference, ACM publications. Modules: Discrete Mathematics, Computer Graphics, Web Technology, Embedded Systems, CCNA, Linux.

Technical stack

Tools and methods

Languages
Python, R, C / C++, SQL
Machine learning
PyTorch, TensorFlow, Scikit-learn, HuggingFace, SpikingJelly, NumPy, Pandas, SHAP, LIME
Research areas
Neuromorphic ML, spiking neural networks, anomaly detection, knowledge distillation, LLM compression, graph neural networks, RAG, transformer architectures, combinatorial optimisation, explainable AI
Infrastructure
MySQL, MongoDB, PostgreSQL, Apache Spark, Hadoop, AWS Redshift, Google Cloud, Docker, Git, CI / CD, MLflow, FastAPI, REST APIs
Analytics
Power BI, Tableau, Excel, Google Data Studio
Contact

Get in touch

I am actively applying for PhD positions for 2026 entry, with a particular interest in groups working on neuromorphic computing, efficient ML inference, and high-performance computing in Germany, Sweden, the Netherlands, and Austria. I am equally open to applied research roles.

The best way to reach me is by email. Reference letters, unpublished drafts, and the ICONS 2026 submission are available on request.

samiurk70@gmail.com GitHub LinkedIn CV