View source for Exploring evolutionary, co-evolutionary and "island evolution" models for the multi-issue negotiation problem

This project takes a complex, well-known problem in agent systems, that of multi-issue negotiation and looks at it from a completely new angle. THe problem involves two agents, negotiating over a genericaally termed "set of items" (which can be seen as the composition of a bundle of goods in e-commerce, a set of resources/tasks in logistics planning etc). 
The utility functions of the agents over different combinations of items are non-linear, leading to a very roughly-shaped utility space, for each one of the agents.

Our approach aims to compare the Pareto-efficiency of the deals reached in two settings:
- One in which both agents reveal their preferences to a centralised mediator, who explores and proposes an optimal combination to both parties
- A decentralized one, in which agents only know (and can only explore) their own utility space, yet, in order to reach an agreement, optimisation has to take place in both utility spaces. For this, a co-evolution model could be used, in which each agent searches its wn utility space thorugh a co-evolving population. This model would involve parties exchanging sets of contracts (which form the best individuals of each population), and including these in the evaluation of the other agent.

In an initial stage, the model can be implemented for the case of bilateral negotiation and the outcomes obtained from the two approaches two approaches (co-evolutionary vs. centralised evolutionary) could be compared. If there are more than 2 agents, "island evolution" models could be used. In this case, the topology of the interaction adds an extra layer of complexity to the problem.

Interested parties: Valentin Robu, Jorge Tavares