Optimizing Complex Automated Negotiation Using Sparse Pseudo-Input Gaussian Processes

Complex negotiations among rational autonomous agents is gaining a mass of attention due to the diversity of its possible applications. This paper deals with a prominent type of complex negotiations, namely, multi-issue negotiation that runs under realtime constraints and in which the negotiating agents have no prior knowledge about their opponents' preferences and strategies. We propose a novel negotiation strategy called Dragon which employs sparse pseudo-input Gaussian processes (SPGPs) to model efficiently the behavior of the negotiating opponents. Specifically, with SPGPs Dragon is capable of: (1) efficiently modeling unknown opponents by means of a non-parametric functional prior; (2) significantly reducing the computational complexity of this functional prior; and (3) effectively and adaptively making decisions during negotiation. The experimental results provided in this paper show, both from the standard mean-score perspective and the perspective of empirical game theory, that Dragon outperforms the state-of-the-art negotiation agents from the 2012 and 2011 Automated Negotiating Agents Competition (ANAC) in a variety of negotiation domains.