Ban lãnh đạo nghiên cứu : Zahia GUESSOUM
Đồng hướng dẫn : DOCTOR Sylvain, ROUSSEL Olivier
For most companies, the time gap between the moment when they begin to spend money and the moment when they receive payment from their customer generates a working capital requirement. The latter may create issues such as contract lost or bankruptcy. In the current context, getting credits from banks may be either costly or even impossible for Small and Medium Enterprises. Indeed, many banks refuse to lend money for small companies or when they estimate that there is a risk. It is then possible to rely on other solutions as factoring and get immediate credit from the invoices.
In this thesis, we propose a factoring marketplace based on related to curious agents that try to infer private information from their partners. This kind of behavior is harmful both for the agents and for the platform more generally. We thus propose to design a negotiation protocol resistant to such behavior by endowing the agents with an incentive to negotiate only when they really try to get a good.
We then propose an automated negotiation agent which relies on Monte Carlo Tree Search techniques. Those techniques have proved to be quite efficient in AI for games. Our agent is able to learn information from its partner, though it is not its main objective in order to get to a beneficial agreement. For this purpose, it relies on opponent modeling techniques and machine learning such as Bayesian Learning and Gaussian Process Regression.
Bảo vệ luận án : 05/17/2018 - 10h - Site Jussieu 25-26/105
Mme Maroua Bouzid-Bouazzid, Univ. de Caen-Basse Normandie [Rapporteur]
M. René Mandiau, rapporteur, Univ. de Valenciennes [Rapporteur]
Mme. Amal El Fallah Seghrouchni, Univ. Pierre et Marie Curie
M. Vincent Chevrier, Univ. de Lorraine
Mme Zahia Guessoum, Univ. Pierre et Marie Curie
M. Sylvain Ductor, Univ. Fédérale du Ceará
M. Olivier Roussel, Kyriba