Research Activities
We studied preference profiles for which majority rule voting does not lead to a cycle.
Result: Transitivity of the Majority Rule. We demonstrated that Condorcet’s paradox occurs if and only if the preferences of at least three voters form a cycle on three candidates, highlighting the importance of focusing on acyclicity in collective preferences and conditions guaranteeing the absence of cycles.
PhD research, Universities of Avignon and Dschang, France and Cameroon, November 2019 – December 2022
Our objective was to design and develop innovative mathematical and algorithmic models to enhance the cyber defense of networks, particularly against stealthy attacks using cyber deception techniques. This included applying stochastic and Bayesian models to optimize defense strategies.
Result: Cyber defense and network epidemics. Development of a Partially Observable Stochastic Game model to counter the spread of botnet attacks in networks via cyber deception techniques. The main innovation lies in the application of the Value Iteration algorithm in the context of epidemic control to maximize the defender’s efficiency against stealthy attacks.
Result: Bayesian game model of centrality. First approach combining game theory, graph theory, and epidemiological modeling to analyze defense strategies against cyberattacks.
Two axes:
Result: Dynamic optimization. Optimization of dynamic allocation of requests to AI models, considering capacity and precision constraints. The objective is to minimize unprocessed requests, respect precision thresholds, and reduce model activation to save resources. The proposed solution is a neural network trained on optimized situations by piecewise linearizing a rational function. This is a contribution to the advancement of the PARFAIT project.
Result: MDP and complexity reduction. Exploration of Markov Decision Process complexity reduction to solve problems with large state spaces. This work focused primarily on load balancing and dynamic resource allocation in 5G networks.
Result: Evolution of the centrality game. Development of new defense strategies based on centrality measures in networks, using the Shapley value from cooperative game theory to rank vulnerabilities in attack graphs. This approach improves the allocation of cybersecurity resources and cyber deception strategies, demonstrating the effectiveness of defense techniques against lateral movement attacks. These results were presented at GameSec 2023.
Result: Collaboration for cybersecurity. Development of a continuous-time differential game framework to model security investments and information sharing between companies, while minimizing privacy risks and preserving information quality.
