The overall objective in this project is to show how socially self-interested autonomous agents (‘economic software agents') and Multiple Party Decision-Making methods could be used for processing Automated Negotiation which will be a key component of electronic markets. In automated negotiation, computational agents find contracts on behalf of the real world parties that they represent. In the original contract net framework, agents solve problems by iteratively making decisions about issues such as the allocation of tasks among themselves. We will extend the contract net for use among socially self-interested, computationally bound agents in several ways. Self-interested autonomous agents, by definition, simply choose a course of action which maximizes their own utility.
Furthermore, we will use learning to improve the negotiation capabilities of the self-interested autonomous agents. In particular we will base our approach on Reinforcement Learning (RL). RL Q-Learning has already proven to be successful in solving a number of real world problems in several domains.
Automated Negotiation is a form of decision-making where two or more parties jointly search for possible solutions with the goal of reaching a consensus. Economics and Multiple Criteria Decision-Making describe such interactions in terms of protocols, strategies, and tactics. It is important to show that these existing methods are applicable in order to get efficiently an automated negotiation. Therefore this project will be carried out at different sites and the following main objectives have to be achieved:
To build a Trial server which implements a decentrally mediated market: The socially self-interested autonomous agents are allowed to send bids and ask for combinations of items to the server. These combinatorial auctions allow users to express complementary interest among items. This capability is particularly important in illiquid, highly volatile, or non-commoditized markets where it is unsure whether one can acquire the items of a desired bundle one at a time. The server determines the winners of the combinatorial auction which means identifying profitable contracts for the agents. Optimal winner determination is NP-complete.
To use several new techniques and tools to speed up the search without
By doing so, the NETMARKET consortium will attempt to construct a paradigm value-negotiations and the benefits will be to reduce drastically certain type of frictional cost and time incurred in trading. This comprehensible conception of value creation is shared by both buyers and sellers.
To show how learning approaches based on Bayesian networks and Reinforcement Learning can be used: - to 'mark' WEB pages via ‘ant-based’ or ‘social insect collective intelligence’ paradigms, in order to find more easily the best path to go to a related data; or – to 'interior' the memorizing of the paths thanks to a 'cognitive map' of the visited WEB pages that drastically reduces the high bandwidth communications time as soon as every mobile agent uses it.
To automatize search, selection and automated negotiation for e-Markets in B2B and B2G environments and help SME Cybernaut purchasers decide ‘What to Buy’.
NETMARKET General Architecture
The Agents Coordinators are used in order to manage the events about the potential Agents Sellers, the Negotiation of the purchase of the product wanted, and the securization of the transaction during the payment phase.
| Français | News
| Site Map
@ 2003 by Tiga Technologies
Information or comments: contact firstname.lastname@example.org