E cient Agent Negotiations in a Realistic Context

In practice, many problems can be modeled by resource allocation problems. However, even if their aim is often to identify a resource distribution maximizing a given objective, different approaches exist. Many solving techniques only focus on the best resource distribution. Such approaches consider resources from one side and agents on the other side. It is the case for all centralized techniques, like with combinatorial auctions and the so-called winner determination problems [5]. However, in many applications, resources are initially already distributed between agents of the population. In such cases, the aim is to identify a sequence of transactions leading from the initial resource allocation to an optimal solution. Then, centralized techniques are not suitable to solve allocation problems in such conditions. In this purpose, methods based on agent negotiations have been developed, where autonomous agents negotiate by pair to identify acceptable transactions. Studies in this field are most of the time theoretical. Some studies established the existence or not of transaction sequence to optimal solutions [4]. Classes of utility functions and payment functions are also studied to design convergent negotiation processes [1]. They also study different scenarios corresponding to different preference representation and to different acceptability criteria [2]. However, these studies did not focus on the mechanism to use in order to get an optimal solution, they characterize properties that may favor the