Mahmoud Abumandour

Mahmoud Abumandour is a PhD candidate at Simon Fraser University, supervised by Professor Alaa Alameldeen. His interests span software systems, computer architecture, and the boundary where hardware and AI meet.


His current work focuses on protecting deep learning applications on edge devices against fault injection attacks, with an emphasis on real-world reliability and attack-aware design.

Portrait of Mahmoud Abumandour

PhD Candidate @ SFU

News

  • Joined the Max Planck Institute for Security and Privacy

    I joined the Max Planck Institute for Security and Privacy as a Research Intern, where I am working on an FPGA-based fault-injection framework to evaluate the security of hardened cores. I am working in Dr. Christof Paar's research group.

  • Paper accepted at CF 2026

    Our paper on protecting decentralized federated learning against model poisoning attacks has been accepted at the 23rd ACM International Conference on Computing Frontiers (CF'26)

  • Paper accepted at HASP 2025

    Our paper on Real-Time Bit-Flip Protection for Deep Neural Networks on GPUs has been accepted at Hardware and Architectural Support for Security and Privacy HASP 2025, co-located with MICRO 2025.

Experience

    • Working on an FPGA-based fault-injection framework to evaluate the security of hardened cores.
    • Assessed the impact of different attack vectors against quantized DNNs.
    • Devised a fatal bit-flip attack against quantized DNNs, including LLMs, and explored potential defenses.
    • Researched, modeled, and assessed CPU front-end features including instruction prefetching and caching.
    • Performed workload analysis to categorize instruction cache footprint and branch behavior.
    • Analyzed sources of miscorrelation between functional and cycle-accurate simulators.
    • Used profiler-guided optimizations to achieve an 8x performance improvement in the release generation process.
    • RTEMS: Release Generator
    • Designed and assessed the performance of two data storage schemes for design points data structures.

Publications

  • EnsembleHealer: Autonomous Recovery from Model Poisoning in Decentralized Federated Learning

    Srinija Ramichetty, Mahmoud Abumandour, Alaa Alameldeen, Guru Venkataramani

    The 23rd ACM International Conference on Computing Frontiers, 2026

    Paper DOI

  • WeightSentry: Real-Time Bit-Flip Protection for Deep Neural Networks on GPUs

    Mahmoud Abumandour, Srinija Ramichetty, Guru Venkataramani, Alaa Alameldeen

    International Workshop on Hardware and Architectural Support for Security and Privacy (HASP), 2025

    Paper DOI Slides

Teaching

  • CMPT 130 / Introduction to Computer Programming (Fall 2025)
  • CMPT 295 / Introduction to Computer Systems (Spring 2023, Summer 2024, Fall 2024, Spring 2025)
  • CMPT 379 / Principles of Compiler Design (Summer 2023, Summer 2025)
  • CMPT 431 / Distributed Systems (Fall 2025)
  • CMPT 750 / Computer Architecture (Fall 2023, Fall 2024, Spring 2026)

Education

Simon Fraser University BC, Canada

    • Advisor: Prof. Alaa Alameldeen
    • Thesis: Uncovering and Mitigating Bit-Flip Vulnerabilities in Full-Precision and Quantized DNNs

Mansoura University Mansoura, Egypt

    • Ranked first in class with a GPA of 3.96/4.0

Service