Portfolio 2026

Hossein Shahverdi

Compiling models, hardening endpoints, warming up the GPU.000

Hossein Shahverdi, M.Sc. Telecommunications Engineering

Small models for streaming signals. Clear explanations for clinical AI.

My research sits where signal processing meets deep learning: lightweight Transformers for sensor streams, generative models for medical imaging, and explainability that clinicians can act on.

Read the papers
Switch to engineering

Selected research

From Wi-Fi signals to MRI volumes: the projects behind the papers.

ThesisPublished

Lightweight Transformer for HAR

2023 – 2025

Activity recognition that fits on a phone.

A parameter-efficient Transformer for on-device inference on streaming smartphone IMU data, benchmarked against CNN and LSTM baselines on public and self-collected datasets. M.Sc. thesis, graded 20/20, presented at IHIET-FS 2025 in London.

  • Designed for on-device, streaming inference
  • Benchmarked against CNN and LSTM baselines
  • Evaluated on public and self-collected datasets
PyTorchTransformersTime seriesEdge AI

Links coming soon

Signal processingPublished

Seeing motion in Wi-Fi

2020 – 2023

Edge detection on channel state information.

Preprocessing pipelines that treat Wi-Fi CSI as an image and extract Canny edge features, sharpening activity boundaries before CNN classification. Published in Information (MDPI) and at ABC 2022.

  • Device-free sensing: no wearables required
  • Canny edge features improve CNN accuracy
  • Two journal papers and one conference paper
Wi-Fi CSICNNOpenCVSignal processing
Read the paper
Medical imagingPublished

3D Dual-CycleGAN

2023 – 2024

Synthesising MRI contrasts that were never acquired.

Co-developed a 3D Dual-CycleGAN for cross-contrast MRI synthesis, trained on multi-GPU PyTorch. Published in the Egyptian Journal of Radiology and Nuclear Medicine.

  • Volumetric (3D) image-to-image translation
  • Cycle consistency in both directions
  • Multi-GPU training pipeline
GANs3D CNNPyTorchMRI

Links coming soon

Explainable AIPreprint

UPhAIR

2025 – 2026

AI explanations a clinician can read.

A hybrid post-hoc explanation pipeline that couples radiomics with LLM-generated clinical reports for glioma IDH-mutation prediction. Co-author; preprint on medRxiv.

  • Radiomics features as the explanation backbone
  • LLM turns attributions into plain-language reports
  • Built for neuro-oncology decision support
XAIRadiomicsLLMsNeuro-oncology

Links coming soon

Publications

Journal articles, conference papers and preprints. My name is highlighted in each author list.

  1. 2026

    UPhAIR: A hybrid pipeline for generating understandable post-hoc AI reports in glioma IDH mutation status prediction

    Gorji, A., Shahverdi, H., Saberi, A., Gheiji, B., Farahani, S., Azemi, G., & Di Ieva, A.

    medRxiv

    Preprint
  2. 2025

    Lightweight Transformer for robust human activity recognition using smartphone IMU data

    Shahverdi, H. & Ghorashi, S. A.

    IHIET-FS 2025, University of East London

    Conference
  3. 2024

    Assessing the efficacy of 3D Dual-CycleGAN for multi-contrast MRI synthesis

    Mahboubisarighieh, A., Shahverdi, H., Jafarpoor Nesheli, S., Alipoor Kermani, M., Niknam, M., Torkashvand, M., & Rezaeijo, S. M.

    Egyptian Journal of Radiology and Nuclear Medicine, 55(1)

    Journal
  4. 2024

    A CSI-based human activity recognition using Canny edge detector

    Shahverdi, H., Fard Moshiri, P., Nabati, M., Asvadi, R., & Ghorashi, S. A.

    ABC 2022, CRC Press / Taylor & Francis

    Conference
  5. 2023

    Enhancing CSI-based human activity recognition by edge-detection techniques

    Shahverdi, H., Nabati, M., Fard Moshiri, P., Asvadi, R., & Ghorashi, S. A.

    Information, 14(7)

    Journal
  6. 2023

    Convolutional neural networks for CSI-based human activity recognition

    Shahverdi, H., Shahbazian, R., Fard Moshiri, P., Asvadi, R., & Ghorashi, S. A.

    Int. Journal of Information & Communication Technology Research

    Journal
  7. 2023

    Use of deep image-to-image translations to assess complementary value of imaging modalities: PET and CT in head and neck cancer

    Rezaeijo, S. M., Mahboubisarighieh, A., Jafarpoor Nesheli, S., Shahverdi, H., Hosseinzadeh, M., Hacihaliloglu, I., Rahmim, A., & Salmanpour, M. R.

    Journal of Nuclear Medicine, 64(S1)

    Abstract
  8. 2023

    Enhancing multi-contrast MRI synthesis: a novel 3D Dual-CycleGAN approach

    Mahboubisarighieh, A., Shahverdi, H., Jafarpoor Nesheli, S., Niknam, M., Torkashvand, M., & Rezaeijo, S. M.

    Research Square

    Preprint

Open questions

Experiments I'm running or about to run, each one a bridge between my research and security. Flip a card to see the approach.

Planned

PacketLens

Can a time-series Transformer spot intrusions in raw network flows?

Planned

Mirage

How easily can sensor-based activity models be fooled, and defended?

Building

Featherweight

How small can a HAR Transformer get before accuracy breaks?

Concept

WaveSense

Room-level presence sensing with a five-dollar Wi-Fi chip?

Methods and tools

Three dots means I use it every week; one means I'm actively learning it.

Machine learning

  • PyTorch
  • TensorFlow
  • Transformers
  • Time-series modelling
  • GANs
  • Explainable AI
  • scikit-learn
  • OpenCV

Signals

  • IMU and sensor streams
  • Wi-Fi CSI
  • EEG
  • MATLAB
  • Particle filters

Security and systems

  • Network protocols
  • Linux
  • Nmap and Wireshark
  • Burp Suite
  • Metasploit
  • Docker and Kubernetes

The journey so far

  1. 2026

    UPhAIR preprint, and PhD applications

    medRxiv

    Explainable AI for glioma IDH prediction. Now applying to PhD programmes in efficient and trustworthy AI.

  2. 2025

    Thesis presented in London

    IHIET-FS 2025, University of East London

    The lightweight HAR Transformer, presented at an international conference.

  3. 2024

    3D Dual-CycleGAN published

    Egyptian J. Radiology and Nuclear Medicine

    Cross-contrast MRI synthesis with volumetric cycle-consistent GANs.

  4. 2023

    M.Sc. completed, thesis graded 20/20

    Shahid Beheshti University

    Two journal papers on Wi-Fi CSI sensing, a JNM abstract on PET/CT translation, and a research internship working with EEG.

  5. 2020

    Ranked 312 of 50,000+ nationwide

    Iranian M.Sc. entrance exam

    Joined the M.Sc. in Telecommunications Engineering and the NAIRG lab; later ranked 3rd in GPA in the cohort.

  6. 2014 – 2019

    B.Sc. in Electrical Engineering

    Sahand University of Technology

    Top 4% of 500,000+ in the entrance exam. Capstone on object tracking with particle filters; TA for DSP and networks.

Looking for a PhD student who ships?

I'm open to PhD positions and research collaborations in efficient, reliable and explainable deep learning.

Email me
Google Scholar

References

  • Prof. Seyed Ali Ghorashi

    M.Sc. thesis advisor, University of East London

  • Dr. Reza Shahbazian

    Research collaborator, University of Calabria

  • Dr. Arman Gorji

    Supervisor and Collaborator, Hamadan University of Medical Sciences and Health Services

Contact details available on request.