PA Pouria Arefijamal

PhD Student · Universidad de Navarra

Pouria
Arefijamal

I design systems that make machine learning practical where the resources are scarce — at the network edge, on constrained IoT devices, and across federated deployments that cannot ship their data anywhere.

My work spans federated learning, edge and cloud computing, and latency-aware scheduling, with recent papers in IEEE Access, IEEE Computer Architecture Letters, and IEEE ICC.

Publications Curriculum Vitae

About

Short version, and what I am after.

I am a PhD student at the Universidad de Navarra. Before that I completed my MSc at the Sharif University of Technology, where I worked on distributed learning for embedded and IoT platforms and on characterising the real cost of running machine-learning workloads on shared infrastructure.

A recurring theme runs through my papers: the gap between a model that works in a benchmark and a model that works on the device, link, or cluster you actually have. Closing that gap means treating I/O behaviour, network latency, device heterogeneity, and energy budgets as first-class parts of the learning problem rather than as deployment details discovered afterwards.

Concretely, that has meant robust distillation-based federated learning for resource-constrained IoT networks, priority-aware SDN orchestration for latency-critical medical IoT, and I/O-aware execution-time estimation for ML workloads. I am always glad to talk about collaborations in these areas.

Research

Four threads, one question.

  • 01

    Federated & Distributed Learning

    Knowledge distillation and robustness techniques that let heterogeneous, bandwidth-limited devices train together without centralising data.

  • 02

    Edge & Cloud Computing

    Placing computation where it pays off — balancing latency, energy, and cost across the device–edge–cloud continuum.

  • 03

    Task Scheduling & Orchestration

    Priority-aware, SDN-driven resource allocation for traffic that has hard deadlines, such as surgical and clinical IoMT.

  • 04

    ML Systems Performance

    Modelling the I/O and memory behaviour of ML workloads to predict execution time before you pay for the run.

Publications

Reverse chronological. Full BibTeX ↓

  1. 2026

    RIFLE: Robust Distillation-based FL for Deep Model Deployment on Resource-Constrained IoT Networks

    Pouria Arefijamal, Mahdi Ahmadlou, Bardia Safaei, Jörg Henkel

    IEEE International Conference on Communications (ICC), 2026

    DOI arXiv PDF

  2. 2025

    I/O-ETEM: An I/O-Aware Approach for Estimating Execution Time of Machine Learning Workloads

    Elham Adibi, Mohammadamin Ajdari, Pouria Arefijamal, Amirsaeed Ahmadi-Tonekaboni, Hossein Asadi

    IEEE Computer Architecture Letters, vol. 24, no. 2, pp. 325–328, 2025

    DOI

  3. 2025

    Priority-Aware SDN Orchestration for Surgical IoMT: A Joint Optimization of Hit Ratio and Latency With Dynamic Resource Reallocation

    Yalda Safaei, Pouria Arefijamal, Mahdi Siamaki, Bardia Safaei

    IEEE Access, vol. 13, pp. 113787–113808, 2025

    DOI Open Access PDF

Also indexed on OpenAlex and Google Scholar.

Contact

Open to collaborations and questions.

The fastest way to reach me is email. I read everything, and I reply to anything about federated learning, edge systems, or the papers above.