Jointly conducted by UGB and IRD, this research proposes an approach to improve coherence in decentralized multi-agent systems.
Quang-Anh Nguyen Vu, researcher at MSI; UMMISCO; LIRIS (IFI).
Thesis defended in 2012 at the doctoral school École doctorale InfoMaths (Lyon ; 2009-....).
This research is the result of an international co-supervision between several partner institutions.
Referenced in the ABES/STAR network, this thesis meets the rigorous criteria of French higher education.
The thesis of Quang-Anh Nguyen Vu addresses a crucial topic in the field of multi-agent systems (MAS), focusing on coherence and robustness in the face of defective information sources. In a world where information is often incomplete or erroneous, the ability of a system to maintain coherence in its data is paramount. Indeed, the reliability of information is a major issue in many sectors, ranging from finance to health, including public safety. The research highlights a trust model, TrustSet, which allows agents to self-organize and assess the reliability of the information they collect. This model relies on a dynamic interaction between agents, where each agent can both share and receive information while evaluating its relevance.
One of the strengths of this thesis is the emphasis on a systemic approach. By considering the coupling between the collection of direct information and communication among agents, the author proposes a method to limit the spread of erroneous information. This is particularly relevant in contexts where agents may fail, whether due to human errors or technical failures. For example, in a natural disaster management system, agents may collect data on weather conditions. If an agent transmits erroneous information about an impending storm, it can lead to inappropriate decisions, endangering human lives. Simulations conducted within collaborative mapping applications show that this approach can significantly improve the quality of the information collected.
However, several questions arise. How can this approach be implemented on a large scale? What are the technical and human challenges to overcome to ensure successful adoption? The research paves the way for reflections on the integration of these systems in various fields, such as natural resource management, security, or public health. For instance, in the health sector, a multi-agent system could collect data on the spread of epidemics, ensuring that the information disseminated is reliable and relevant. This could not only improve the response of health authorities but also strengthen public trust in the information provided.
It is also essential to consider the ethical implications of using such systems. The collection of information, especially in sensitive contexts, raises questions about privacy and data protection. Agents must be designed to adhere to strict ethical standards, ensuring that the information collected does not infringe on individuals' rights. Furthermore, the issue of accountability in the event of errors in the transmitted data must be clearly defined.
In conclusion, this thesis is not limited to a simple theoretical advancement. It proposes concrete solutions to current issues while highlighting the importance of trust and communication in decentralized systems. Decision-makers and stakeholders in the technology sector have a vested interest in leveraging these results to improve the reliability of information systems in various contexts. By integrating models like TrustSet, it is possible to create more resilient systems adapted to contemporary challenges, whether in crisis management, environmental protection, or public health.
Thus, this research opens the door to a multitude of potential applications, reinforcing the idea that technology, when used thoughtfully and ethically, can be a powerful lever for improving our society.
Access the full study
Sources and Access
Quang-Anh Nguyen Vu. Cohérence et robustesse dans un système multiagent perturbé : application à un système décentralisé de collecte d’information distribué. Autre [cs.OH]. Université Claude Bernard - Lyon I, 2012. Français. ⟨NNT : 2012LYO10232⟩. ⟨tel-00987118⟩