Umar Faruk Abdullahi

Umar Faruk Abdullahi

Machine Learning Engineer, Huawei R&D

I work on reliable intelligent systems. I am interested in safe planning, mechanistic diagnostics, and adversarial robustness of foundation models.

Recently completed a joint masters through Erasmus Mundus EDISS: M.Sc. in Computer Engineering (AI & ML) from Finland, and M.Sc. in Intelligent Systems (computer vision and robotics) from Spain. Previously a research assistant at Turku Bioscience and a software engineer at Mercurie. B.Sc. in Computer Science from Ahmadu Bello University.

Updates

Mar 2026

Our paper "Beyond Suffixes: Token Position in GCG Adversarial Attacks on Large Language Models" was accepted at the ICLR ICBINB 2026 Workshop.

Nov 2025

Graduated from the EDISS master's program. My thesis on Adaptive Bathymetric Planning can be found here.

Feb 2024

Presented ongoing work on Parkinson's detection & biomarker discovery at the EDISS Winter School in Västerås.

Oct 2023

Won the Nokia Challenge Hackathon for network bandwidth optimization with NetworkAsCode.

Aug 2023

Began the EDISS joint master's (Finland & Spain) on an Erasmus Mundus scholarship.

Research

  • Beyond Suffixes: Token Position in GCG Adversarial Attacks on Large Language Models

    Eddoubi, H., Abdullahi, U. F., Hassan, F.

    ICLR 2026 ICBINB Workshop, accepted · arXiv:2602.03265

  • Hybrid MCTS–CEM for Adaptive Bathymetric Planning

    Abdullahi, U. F.

    M.Sc. Thesis, 2025 · National Library of Finland · Doria Archives