Wassila Ouerdane

My research investigates how AI-driven decisions can be understood, justified and assessed in terms of reliability. At the intersection of decision theory, formal methods, explainable AI and human-AI interaction, I study how the foundations of a decision can be represented, how its reasoning can be explained, and how explanations can support human understanding and the evaluation of AI systems.

My overarching research question is: How can we develop AI systems whose decisions are justifiable, whose explanations are meaningful to users, and whose reliability can be assessed as their models and interactions become increasingly complex?

This question connects five complementary research themes: preference modelling and learning; formal explanations of symbolic decisions; interactive explanation and human-machine dialogue; the explainability and hybridization of learned models; and the reliability and explainability of LLM-based multi-agent systems.

Across these themes, I investigate the relationships between the foundations of a decision, the justification of its outcome, the understanding developed by its users, and the reliability of the system that produces it.

Research Trajectory

My research follows a progression from the formal modelling of decisions to the explanation and assessment of increasingly complex AI systems. Rather than representing separate research topics, these directions address complementary aspects of a common scientific challenge: establishing why a system produces a decision, how that decision can be explained, and how its reliability can be assessed.

My work on preference modelling provides formal representations of decision makers' judgments. Research on symbolic models then investigates how the reasoning underlying decisions can be formally justified. Interactive explanation extends this perspective by considering how explanations can be adapted through dialogue and how understanding can be progressively constructed.

The development of learned AI models raises additional questions about the interpretability, grounding and reliability of their predictions. My research on explainability and hybrid AI investigates how explicit knowledge can complement learned models. My current work on LLM-based multi-agent systems extends these concerns to interacting agents, where errors, information and decisions may propagate across the system.

This trajectory connects formal decision modelling, explainable reasoning, human-AI interaction and system-level reliability, while allowing each research theme to retain its own methods and scientific objectives.

Research Themes

1. Preference Modelling, Elicitation and Learning

Research question. How can decision makers' preferences be represented and learned from limited, incomplete or imperfect information?

My research addresses preference modelling in multiple criteria decision aiding, including sorting models such as MR-Sort and constrained sorting, reference-point-based models, and preference learning from data that may only partially satisfy monotonicity assumptions. I also study the integration of preference models into optimization problems.

Objective. To develop preference models that faithfully represent decision makers' judgments while remaining interpretable, computationally tractable and suitable for decision support and optimization.

Related PhD theses. Manel Maamar, Jinyan Liu, Massinissa Mammeri, Pegdwendé Stéphane Minoungou and Ali Tlili.

2. Formal Explanations of Decisions from Symbolic Models

Research question. How can decisions be justified through explanations that are logically grounded, informative and concise?

My research investigates explanation schemes based on deductive reasoning, their formal properties—including soundness, completeness and conciseness—and the algorithms required to compute them. Applications include preference aggregation, scheduling and assignment, as well as rule-based systems, including possibilistic and fuzzy models capable of generating textual justifications.

Objective. To establish formal foundations for explanations that make the reasoning underlying recommendations, optimization solutions and rule-based decisions accessible to users.

Related PhD theses. Khaled Belahcène, Manuel Amoussou, Mathieu Lerouge, Francesco Sabatino, Karim El Mernissi and Ismaïl Baaj.

3. Interactive Explanation and Human-Machine Dialogue

Research question. How can explanations evolve through interaction to address uncertainty and support users' understanding?

I investigate explanatory dialogue protocols, the management of uncertainty in human-machine exchanges, and the interaction between explanation and preference elicitation. Applications include the interpretation of visual scenes and interactive decision support.

Objective. To develop dialogue-based explanation methods that adapt to users' needs, help resolve inconsistencies and support the progressive construction of shared understanding.

Related PhD theses. Dao Thauvin and Armand Gaudillier.

4. Explainability and Hybridization of Learned Models

Research question. How can learned models be made more interpretable and their predictions better grounded in explicit knowledge?

My research follows two complementary directions: explaining what models have learned, for example through concept-based explanations and the analysis of models operating on dynamic graphs; and combining learned models with explicit knowledge, including knowledge graphs, conceptual models and rules.

Objective. To investigate how learning and explicit knowledge representations can complement one another in the development of more interpretable and reliable AI systems. Applications include healthcare, disinformation detection, certification of critical systems and IT system monitoring.

Related PhD theses. Reda Arab, Hugo Miccinilli, Baptiste Sivy, William Kim, Charlotte Claye, Géraud Faye and Angélique Yameogo.

5. Reliability and Explainability of LLM-Based Multi-Agent Systems

Research question. How can the reliability and collective behaviour of interacting AI agents be assessed and explained?

This emerging research direction extends questions of explainability and reliability to systems composed of multiple interacting agents powered by large language models. It addresses the propagation of factual errors, interactions between agents, the assessment of collective reliability, and the explanation of distributed reasoning.

Objective. To develop methods for analysing error propagation, identifying the contributions of individual agents, establishing system-level reliability criteria and explaining collective outcomes.

This research direction is being developed notably within the TMAS research chair.

Common Scientific Thread

The common thread across my research is the relationship between decision foundations, explanation and reliability. In preference modelling, the focus is on representing the judgments that determine a decision. In formal explanation, it is on establishing why a decision follows from a model. In interactive explanation, it is on how users progressively develop an understanding of that decision. In learned and hybrid AI, it is on the relationship between predictions, explicit knowledge and explanatory claims. In multi-agent systems, it is on how these questions extend to interactions and collective behaviour.

My current research agenda builds on these complementary perspectives to investigate how explanations can support the assessment of reliability in increasingly complex AI systems. Formal justification, user understanding and empirical reliability are distinct objectives; my research explores how they can be connected without conflating them.