Hamza Ba-mohammed

PhD Candidate at UM6P College of Computing. Adversarial Robustness in Cooperative Multi-Agent Learning.

I am a PhD Candidate in Computer Science / Artificial Intelligence at the College of Computing (CC), Mohammed VI Polytechnic University (UM6P), Morocco.

My research investigates adversarial robustness in cooperative Multi-Agent Reinforcement Learning (MARL): when autonomous agents coordinate in shared environments, how do localized observation perturbations propagate across interacting teammates, and how can we design adaptive, parameter-preserving execution-time defenses to guarantee collective resilience?

I am co-supervised by:

Previously, I received a State-Engineer Diploma in Artificial Intelligence Engineering (accredited as equivalent to an M.Sc. in Engineering) from ENSIAS, Rabat, following two years of intensive Classes Préparatoires (MPSI/MP) at Lycée Moulay Youssef.

Research & Publications Curriculum Vitae Google Scholar ORCID GitHub Email

Research Focus: Adversarial Robustness in Cooperative Multi-Agent Learning

In cooperative multi-agent reinforcement learning (MARL), autonomous agents learn to coordinate in shared environments where individual decisions affect the entire group. When observations are deliberately manipulated, consequences need not remain confined to the targeted agent: a robot receiving corrupted sensor information may take inappropriate actions that force peers to react, either containing the disturbance or compounding its damage.

My research investigates the relationship between individual vulnerability and collective performance in cooperative teams:

  • Individual vs. Collective Consequences: Analyzing how localized observation perturbations propagate through cooperative teams, and characterizing whether peer responses represent harmful disruption or useful compensation.
  • Team Organization (Heterogeneity & Communication): Examining how differences in agents’ learned behaviors (behavioral heterogeneity) and inter-agent communication influence team robustness, without assuming diversity alone guarantees resilience.
  • Adaptive Execution-Time Protection: Developing defense mechanisms that operate during execution while preserving pretrained policy parameters. This line of inquiry explores principles of differentiated collective protection—such as artificial immune systems—with multi-robot teams as a prospective empirical direction.

Recent News

  • [Oct 2025] Presented research poster on Adversarial Attacks and Defenses in Multi-Agent Deep Reinforcement Learning at the UM6P Doctoral Days 2025 (Benguerir). [ Poster PDF]
  • [Dec 2025] First author of comprehensive survey paper “A Survey on Adversarial Multi-Agent Deep Reinforcement Learning” (with L. El Bouga, A. Andam, M. Essoullami, D. Benhlima, J. Bentahar, and M. Hedabou), currently under major revision at ACM Computing Surveys (ACM CSUR).
  • [Nov 2024] Awarded 1st Rank & Global Nominee at the NASA Space Apps Challenge 2024 (Ifrane Section) and qualified for international global judging.

Research Spotlight

A Survey on Adversarial Multi-Agent Deep Reinforcement Learning

Hamza Ba-mohammed, Latifa El Bouga, Amine Andam, Mostapha Essoullami, Douae Benhlima, Jamal Bentahar, Mustapha Hedabou
Under Major Revision — ACM Computing Surveys (ACM CSUR)
A comprehensive synthesis connecting multi-agent decision models (Markov games, POSGs, Dec-POMDPs) to game-theoretic solution concepts. Develops a multi-dimensional attack taxonomy, examines threat-model access assumptions, provides a functional defense categorization (robustness evaluation vs. enhancement), and critically reviews published empirical benchmarks.

Adversarial Robustness in Cooperative Multi-Agent Learning

Hamza Ba-mohammed (Doctoral Research, CC UM6P)
Doctoral Investigation (2024 – Present)
Investigating the propagation of localized observation perturbations in cooperative teams. Developing an analytical formulation distinguishing effects on targeted agents, peer reactions, and collective team performance under policy heterogeneity and communication constraints, paired with simulation-based experiments in multi-agent environments.