Samir Poudel
Ph.D. Candidate in Computational & Data Science
Middle Tennessee State University · Murfreesboro, TN
Ph.D. Advisor: Dr. Kritagya Upadhyay
spoudel04@gmail.com · June 2026
Received August 2023 · Revised June 2026
Abstract. I am a Ph.D. candidate building privacy-preserving distributed AI systems across federated learning, blockchain security, and robust machine learning. My research addresses a core tension in modern AI: the most valuable data (medical records, financial histories, sensor streams) is exactly the data that cannot be shared. I build the tools that let models learn from it anyway, without the data ever leaving its owner.
Index Terms: Federated Learning, Blockchain Security, Robust Machine Learning, Privacy-Preserving AI, Distributed Systems, Smart Contract Analysis, Healthcare AI.
1.Background
Fig. 1. Samir Poudel, Murfreesboro, TN.
I grew up in Pokhara, Nepal, and trained as a software engineer before moving into research. I am currently a Ph.D. candidate in Computational & Data Science at Middle Tennessee State University (August 2023–present). Along the way I completed an M.S. in Computer Science from MTSU in December 2025. My undergraduate degree was a B.S. in Software Engineering from Pokhara University (2014–2018).
Before the Ph.D., I worked as a data analyst at Urja Tech in Pokhara, Nepal (2018–2022), building SQL databases, Tableau dashboards, and predictive models for business forecasting.
My research focus is secure distributed machine learning, specifically federated learning, blockchain security, and privacy-preserving AI. I test under realistic adversarial conditions, not just clean benchmarks.
At a glance
17 publications · 140+ citations
IEEE, ACM & indexed journals
Ph.D., Computational & Data Science
MTSU, Aug 2023–Present
M.S., Computer Science
MTSU, Aug 2023–Dec 2025
B.S., Software Engineering
Pokhara University, Nepal, 2014–2018
Based in
Murfreesboro, Tennessee · from Nepal 🇳🇵
2.Research
My work clusters into five threads, connected by shared methods and a common goal of making AI systems trustworthy under adversarial conditions. Two novel architectures central to this work are SPA (Subspace Projection Aggregation), a federated aggregation method robust to non-IID data and Byzantine clients, and HetLoRA-M, a heterogeneous LoRA fine-tuning framework for federated LLMs across clients with differing model capacities.
1. Federated Learning & Privacy-Preserving AI. I build federated learning frameworks that train models across distributed data owners without centralizing any sensitive data. The main failure modes I address are backdoor attacks, heterogeneous data distributions, and communication bottlenecks that block real-world deployment.
RBBD reduces backdoor attack success below 15% while keeping model utility intact
2. Blockchain Security & Smart Contract Analysis. Smart contracts hold billions of dollars in value but break under subtle code vulnerabilities. I build multimodal tools that fuse source code, opcode, and bytecode to catch vulnerabilities that single-modality approaches miss.
VulnFusion: multimodal transformer for smart contract vulnerability detection
3. Robust Machine Learning for Healthcare & Science. I apply machine learning to safety-critical domains where reliability directly affects patient outcomes. My work covers brain tumor segmentation, Alzheimer's classification, ICU outcome prediction, cervical cancer detection, and physiological stress analysis.
ICU length-of-stay prediction from MIMIC-III (28 citations, most cited work)
4. Distributed Systems & Big Data Engineering. Scalable AI needs scalable infrastructure. I benchmark distributed data pipelines, optimize parallelization for preprocessing bottlenecks, and build cloud-native ML systems with measurable throughput and latency gains.
MPI4Py preprocessing optimization, 38 citations, published in Machine Learning with Applications
5. Game Theory & Strategic AI. Federated learning and multi-agent systems are strategic environments. Participants can act selfishly, collude, or drop out to maximize their own gain. I apply mechanism design, Nash equilibria, and cooperative game theory to build distributed AI systems where honest participation is the rational choice.
Models FL participation as a strategic game under data heterogeneity and free-rider incentives
3.Selected Publications
Full list on Google Scholar. A selection follows, grouped by year.
Loading…
4.Software & Projects
Selected implementations and tools arising from or alongside my research.
Cera Dental Clinic Website ↗ [Next.js,Tailwind CSS,Framer Motion]
Full production website for a dental clinic in Pokhara, Nepal. Features glassmorphism navigation, Bento grid service catalog, Framer Motion animations, FAQ accordion, doctor profiles, patient reviews, and a mobile floating CTA bar.
Personal Finance Dashboard ↗ [Next.js,Supabase,TypeScript]
Comprehensive personal finance tracker with transactions, installments, monthly bills, bonus tracking, and credit card management. Backed by Supabase.
Academic Tracker ↗ [Next.js,Supabase,Groq AI]
Full-stack grade tracker with AI syllabus extraction, live grade calculator, and animated progress visualization per course.
I-140 Insights ↗ [FastAPI,React,PostgreSQL]
Tracks USCIS I-140 immigration cases across all service centers by enumerating receipt numbers, with automated scraping and analytics dashboard.
Edge-Case Backdoor Detection (RBBD) ↗ [Federated Learning,Security,PyTorch]
Representation-based defense framework for federated learning, reducing attack success rates below 15%.
Privacy-Preserving Environmental Monitoring ↗ [Federated Learning,Computer Vision,UNet]
Federated learning framework using satellite imagery for deforestation detection, achieving 0.85 IoU.
Federated Fine-Tuning of LLMs (FED-LLM) ↗ [Federated Learning,LLMs,PyTorch]
Privacy-preserving fine-tuning of large language models across distributed clients using federated learning. Evaluates communication efficiency, data heterogeneity, and model convergence without centralizing training data.
Cloud Fake News Detection [NLP,AWS,BERT]
Serverless AWS solution (BERT/LSTM) with 94.9% accuracy and low latency.
Notebook ↗ [Next.js · Supabase · AES encryption]
Paper-aesthetic research notebook with auto-save, note types, tags, folders, export, and per-note colour labels. Live below.
6.Experience
Research Appointment
Graduate Research Assistant, Middle Tennessee State University (MTSU)
Aug 2023 – Present · Murfreesboro, TN
- Year 1: Built autonomous vehicle simulation (CARLA/DQN), CNN for Alzheimer's classification, fNIRS mental workload detection, crypto price prediction, MPI4Py preprocessing optimization, and ICU length-of-stay prediction
- Year 2: Built VulnFusion and QuadraCode AI for smart contract vulnerability detection, multimodal cervical cancer detection, and stress analysis from chest-worn sensors
- Year 3: Built RBBD federated learning backdoor defense, benchmarked big data frameworks, and studied blockchain in gaming ecosystems
Table 1. — Conference Presentations & Invited Talks (11 total, 2023–2025)
| # | Year | Title | Venue | Location |
|---|---|---|---|---|
| 1 | 2025 | RBBD: A Representation-Based Framework for Edge-Case Backdoor Defense in Federated Learning | IEEE TPS / CIC 2025 | Pittsburgh, PA |
| 2 | Privacy Meets Conservation: Federated Learning's Revolution in Deforestation Detection | IEEE AIIoT 2025 | Seattle, WA | |
| 3 | A Big Data Optimization Comparison Using Apache Spark and Apache Airflow | IEEE CCWC 2025 | Las Vegas, NV | |
| 4 | Chain Your Loot: Implementing Blockchain Into Gaming Loot Box Markets | IEEE CCWC 2025 | Las Vegas, NV | |
| 5 | 2024 | QuadraCode AI: Smart Contract Vulnerability Detection with Multimodal Representation | ICCCN 2024 | — |
| 6 | Scalable Multimodal Machine Learning for Cervical Cancer Detection | IEEE AIIoT 2024 | Seattle, WA | |
| 7 | Comparative Study of Real-Time and Batch Processing Approaches in ML-Based Fraud Detection | ACR 2024 | Madrid, Spain | |
| 8 | 2023 | Using the CARLA Simulator to Train A Deep Q Self-Driving Car | IEEE Big Data 2023 | Sorrento, Italy |
| 9 | GAN-Based Data Augmentation for Chest X-ray Classification in Lung Disease Diagnosis | ACM Mid-Southeast 2023 | Gatlinburg, TN | |
| 10 | Empowering AI's Trust and Big Data Resources in Healthcare Industry with Blockchain | ACM Mid-Southeast 2023 | Gatlinburg, TN | |
| 11 | Improved Prediction of Classification of the Alzheimer's Disease with Convolutional Neural Network | IEEE SPMB 2023 | — |
Bold dividers separate presentation years. Rows shaded for readability.
7.Awards & Honors
Graduate Research Assistantship2023–Present · MTSU
Full funding award covering tuition and stipend for Ph.D. research in distributed AI.
Additional awards (scholarships, best-paper nominations, fellowships) to be listed here.
9.Cite This Page
If you reference this page in academic work, please use the following BibTeX entry:
For individual papers, see DOI links in § 3.