Soroush Ziaeinejad

Soroush Ziaeinejad

Machine Learning Researcher · PhD in Computer Science

School of Computer Science, University of Windsor

Machine learning researcher and engineer. I completed my PhD in Computer Science in September 2026 with a dissertation on context-aware client selection in federated learning for heterogeneous, non-IID environments. My work spans information retrieval, ranking and recommendation, large language models, graph neural networks, and computer vision, with publications in Information Fusion (Elsevier) and ACM CIKM, and papers in press at ICMHI and FLLM.

📍 Toronto, Ontario, Canada 🎓 PhD, September 2026

Research Interests

Large Language Models Natural Language Processing Reinforcement Learning Federated Learning Deep Learning Graph Neural Networks Distributed AI Systems Privacy-Preserving AI

Experience

Research Assistant (PhD) · School of Computer Science, University of Windsor

Sep 2021 – Sep 2026
  • Developed a context-aware client-selection strategy for Federated Learning under non-IID data distributions.
  • Applied the strategy to federated medical image classification with ResNet-50 on 5 datasets, building the experimental pipeline in Python with PyTorch and Flower. Improved convergence speed, precision, and AUC. In press, ICMHI 2026.
  • Adapted the strategy for LLM instruction fine-tuning with PEFT/LoRA, using PyTorch, Hugging Face, and Flower. Improved topic-level convergence and performance on rare topics. In press, FLLM 2026.
  • First-authored a survey on federated learning in Information Fusion (Elsevier), introducing a taxonomy of existing methods and proposing context-aware client-selection strategies to address data heterogeneity.
  • Developed and released SEERa, an end-to-end open-source Python framework for predicting like-minded user communities in text-based social networks from users’ evolving interests, built as five layers covering data cleaning and feature extraction, topic modeling, user representation, temporal user graph embedding, and graph clustering. Evaluated as a news article recommender with NDCG, MRR, and F1@k. Published at ACM CIKM 2022.
  • Developed NeuCG (Neural Computational Geometry) — investigating computer vision and deep neural networks for solving computational geometry problems.

Machine Learning Engineer · Chamrosh Tech

2019 – 2021
  • Joined as the first ML engineer and built the foundational AI system for a Persian rug authentication platform, supporting fraud prevention and copyright protection (patent pending).
  • Designed and implemented an end-to-end computer vision pipeline to generate unique digital signatures from rug scans, training an image-to-image translation model to reconstruct rug design maps, generate signatures, and match rugs against registered records.
  • Addressed the domain shift between professional scans used for training and camera photographs received in production, training and integrating an encoder to map photographs into scan-like representations for the authentication pipeline.
  • Worked across the full ML development lifecycle, including data collection and cleaning, model architecture, training, and testing.

Research Assistant (MSc) · Image Processing & Pattern Recognition Lab, Amirkabir University of Technology

2016 – 2019
  • MSc Thesis — Multiple Event Detection in Surveillance Videos: combined sparse representation with deep CNNs for multi-class event detection (fire, car crashes, fights, robberies) in real-world surveillance video streams.
  • Investigated GAN-based anomaly detection with transfer learning to improve performance under limited labeled data.
  • Crowd Counting — built a computer-vision pipeline for public-safety monitoring of dense crowd scenes.
  • Artificial Poet — designed a neural language model for Persian poetic text generation.

Selected Projects

SEERa

Open-source Python framework for future community prediction in dynamic social networks using temporal Graph Neural Networks. Published at ACM CIKM 2022 and winner of the University of Windsor Demo Day.

PyG-TemporalGNNPython

Context-Aware Federated Learning

Client selection and collaboration strategies for FL under heterogeneous, non-IID data. Includes a curriculum-driven selection approach, a medical-imaging application (in press, ICMHI 2026), and federated LLM instruction fine-tuning (in press, FLLM 2026).

FlowerPyTorchPrivacy

NeuCG — Neural Computational Geometry

Research project investigating whether computer vision and deep neural networks can solve classical computational geometry problems.

Computer VisionDeep Learning

Rug Authentication System

End-to-end computer-vision pipeline generating unique digital signatures for handwoven Persian rugs via image-to-image translation. Patent pending.

Image-to-ImageOpenCV

Multiple Event Detection in Surveillance Video

Sparse representation combined with deep CNNs for detecting fire, car crashes, street fights, and robberies in real-world urban surveillance streams.

Sparse CodingCNNGAN

Artificial Poet

Neural language model for Persian poetic text generation — an early exploration of generative sequence modeling for low-resource creative text.

NLPLanguage Modeling

Publications

Journal
2026
S. Ziaeinejad, S. Samet. “Federated Learning with Context-Aware Client Collaboration: Challenges, Advances, and Open Problems.” Information Fusion, Elsevier. IF · Elsevier
Conference
2026
S. Ziaeinejad, S. Samet. “Context-Aware Client Selection in Federated LLMs under Non-IID Distributions.” FLLM. (In press) FLLM
2026
S. Ziaeinejad, S. Samet. “Context-Aware Client Selection in Federated Learning for Heterogeneous Medical Imaging.” ICMHI — International Conference on Medical and Health Informatics. (In press) ICMHI
2022
S. Ziaeinejad, S. Samet, H. Fani. “SEERa: A Framework for Community Prediction.” CIKM — Conference on Information and Knowledge Management, ACM, Atlanta, GA, USA. ACM CIKM
2019
A. Atghaei, S. Ziaeinejad, M. Rahmati. “Abnormal event detection in urban surveillance videos using GAN and transfer learning.” Smart Tehran Conference, Tehran, Iran.

Education

PhD, Computer Science · University of Windsor

Sep 2021 – Sep 2026

Windsor, Ontario, Canada · Advisor: Dr. Saeed Samet

  • Dissertation: Context-Aware Client Selection in Centralized Federated Learning.
  • Research domains: Federated Learning, Graph Neural Networks, NLP, Social Network Analysis, Information Retrieval.
  • Seminars: “An Introduction to SEERa — a Python-based Framework for Community Prediction in Social Networks”; “A Curriculum-Driven Approach to Client Selection in Federated Learning”.

MSc, Computer Science · Amirkabir University of Technology (AUT)

2016 – 2019

Tehran, Iran · Advisor: Dr. Mohammad Rahmati

  • Thesis: Multiple Event Detection in Surveillance Videos Using Sparse Representation.
  • Focus: Detection of safety-critical events (fire, car crash, street fight, robbery) in urban surveillance video using sparse representation and deep neural networks.
  • Seminar: Sparse Coding and Its Applications.

BSc, Computer Engineering · Iran University of Science and Technology (IUST)

2012 – 2016

Tehran, Iran · Advisor: Dr. Saeed Parsa

  • Thesis: Integration of Static and Dynamic Analyses of Java Code Faults.

Technical Skills

AI / ML
PyTorchTensorFlowKeras Hugging Facescikit-learnOpenCV
Federated Learning
Flower (flwr)TensorFlow Federated
Graph / GNN
PyTorch Geometric (PyG)PyG-TemporalDGL
NLP
HF TransformersLLM training & fine-tuning NLTKspaCyLDA / BERTopic
Programming
Python (primary)Java C / C++ / C#MATLAB
Cloud & Infra
DockerAWSAzureGit
Data Science
PandasNumPyMatplotlibSciPy
Databases
SQLPostgreSQLMongoDB

Teaching

University of Windsor — Graduate Teaching Assistant (PhD)
2021 – Present · 15 semesters
Internet Applications and Distributed SystemsFall 2025, Winter 2026, Summer 2026
Information Security for ITFall 2024
Deep Learning and Neural NetworksSummer 2024
Advanced Systems ProgrammingWinter 2024
Systems ProgrammingSummer 2023, Fall 2023
Computer Languages, Grammars, and Translators (Compilers)Winter 2023
Database Management SystemsFall 2022
Introduction to Algorithms and Programming IISummer 2022
Theoretical Foundations in Computer ScienceWinter 2022
Advanced Software Engineering TopicsFall 2021
Amirkabir University of Technology (AUT) — Graduate TA (MSc)
2016 – 2019
Digital Image ProcessingSpring 2019
Iran University of Science and Technology (IUST) — Volunteer TA (BSc)
2012 – 2016 · 5 courses
Advanced ProgrammingFall 2015
Languages Theory and AutomataSpring 2015, Fall 2015
Database LaboratorySpring 2015
Data Structures and AlgorithmsFall 2014, Spring 2015

Mentoring, Workshops & Service

Mentor — Agentic AI Challenge · JLR Hackathon, University of Windsor × Jaguar Land Rover

2024 – 2025
  • Mentored student teams in an industry–academia hackathon focused on Agentic AI system design.
  • Provided technical guidance on AI architectures, LLM integration, and agentic reasoning frameworks in collaboration with JLR engineers.

Research Project Mentor (RPM) — Technical Workshops · University of Windsor

Jan 2023 – May 2025

Recurring workshops delivered each semester to graduate and undergraduate students:

  • Introduction to Natural Language Processing (NLP)
  • Python Programming for Topic Modeling: NLP Techniques for Data Analysis
  • NLP for Recommender Systems: Personalized Content and Product Recommendations
  • NLP and Social Networks: Exploring Social Network Analysis (SNA)
  • Machine Learning and Neural Networks using PyTorch
  • Graph Neural Networks using PyTorch Geometric (PyG)

Academic Service

  • Mentor — leading a student team for the course topic “Project Management: Techniques and Tools”.
  • Organizer — Smart Tehran Conference, Tehran Municipality ICT Organization (2019).
  • Organizer — ACM Programming Contest, School of Computer Engineering, IUST (2015).

Awards & Honors

★Demo Day Winner — Best Research Project for SEERa, University of Windsor (2023).
★Ranked 9th (Top 1%) — National PhD Entrance Exam, Iran (2019).
★Ranked 5th among 60+ teams — Data Mining Contest, AUT × Alibaba flight booking system (2019).
★Ranked 71st (Top 0.25%) — National MSc Entrance Exam, Iran (2016).
★Ranked 1st among 25 teams — ACM Programming Contest, IUST (2014).

Contact

I'm open to research collaborations and to industry and academic opportunities in machine learning. Feel free to reach out.

Powered by Formspree — no server required.