DENOS Lab, University of Calgary ← Back to team
Hossein KhademSohi

Hossein KhademSohi

PhD Candidate in Electrical and Software Engineering, University of Calgary

Federated Learning

Google Scholar

Biography

Hossein KhademSohi is a PhD candidate at DENOS Lab working on data science and software engineering for efficient deep learning. His research covers early exit mechanisms that cut inference cost without labels, contribution estimation that makes federated learning fair and efficient, and model-agnostic federated video super-resolution.

Projects

  • Contribution estimation in federated learning

    Owen sampling and SNIP-Owen values that accelerate how client contributions are estimated in heterogeneous federated training.

  • Efficient inference

    Selfxit, an unsupervised early exit mechanism that lets deep networks stop computing once a prediction is settled.

Publications

  • TMLR2024

    Selfxit: An Unsupervised Early Exit Mechanism for Deep Neural Networks

    Hossein KhademSohi, M. Abedi, Y. Ioannou, Steve Drew, P. Jamshidi, and Hadi Hemmati

  • ACM MMSys2026

    FedVSR: Towards Model-Agnostic Federated Learning in Video Super-Resolution

    A. M. Dehaghi, Hossein KhademSohi, R. Razavi, Steve Drew, and M. Moshirpour

  • arXiv2025

    Owen Sampling Accelerates Contribution Estimation in Federated Learning

    Hossein KhademSohi, Hadi Hemmati, Jiayu Zhou, and Steve Drew

    paper

  • SNOWFL: Efficient and Heterogeneous Federated Learning with SNIP-Owen-values

    Hossein KhademSohi, A. M. Dehaghi, Hadi Hemmati, Jiayu Zhou, M. Moshirpour, and Steve Drew

  • IEEE ICSPIS2016

    Accuracy-energy Optimized Location Estimation Method for Mobile Smartphones by GPS/INS Data Fusion

    Hossein KhademSohi, S. Sharifian, and K. Faez

← All team members