http://dl.gi.de/handle/20.500.12116/11091 2026-07-21T13:33:08Z 2026-07-21T13:33:08Z Finnsson, Hilmar Björnsson, Yngvi http://dl.gi.de/handle/20.500.12116/11191 2018-03-20T10:33:08Z 2011-01-01T00:00:00Z

CadiaPlayer: Search-Control Techniques Finnsson, Hilmar; Björnsson, Yngvi Effective search control is one of the key components of any successful simulation-based game-playing program. In General Game Playing (GGP), learning of useful search-control knowledge is a particularly challenging task because it must be done in real-time during online play. In here we describe the search-control techniques used in the 2010 version of the GGP agent CadiaPlayer, and show how they have evolved over the years to become increasingly effective and robust across a wide range of games. In particular, we present a new combined search-control scheme (RAVE/MAST/FAST) for biasing action selection. The scheme proves quite effective on a wide range of games including chess-like games, which have up until now proved quite challenging for simulation-based GGP agents.

2011-01-01T00:00:00Z Haufe, Sebastian Michulke, Daniel Schiffel, Stephan Thielscher, Michael http://dl.gi.de/handle/20.500.12116/11189 2018-03-20T10:33:08Z 2011-01-01T00:00:00Z

Knowledge-Based General Game Playing Haufe, Sebastian; Michulke, Daniel; Schiffel, Stephan; Thielscher, Michael Although we humans cannot compete with computers at simple brute-force search, this is often more than compensated for by our ability to discover structures in new games and to quickly learn how to perform highly selective, informed search. To attain the same level of intelligence, general game playing systems must be able to figure out, without human assistance, what a new game is really about. This makes General Game Playing in ideal testbed for human-level AI, because ultimate success can only be achieved if computers match our ability to master new games by acquiring and exploiting new knowledge. This article introduces five knowledge-based methods for General Game Playing. Each of these techniques contributes to the ongoing success of our FLUXPLAYER (Schiffel and Thielscher in Proceedings of the National Conference on Artificial Intelligence, pp. 1191–1196, 2007), which was among the top four players at each of the past AAAI competitions and in particular was crowned World Champion in 2006.

2011-01-01T00:00:00Z Kirci, Mesut Sturtevant, Nathan Schaeffer, Jonathan http://dl.gi.de/handle/20.500.12116/11193 2018-03-20T10:33:08Z 2011-01-01T00:00:00Z

A GGP Feature Learning Algorithm Kirci, Mesut; Sturtevant, Nathan; Schaeffer, Jonathan This paper presents a learning algorithm for two-player, alternating move GGP games. The Game Independent Feature Learning algorithm, GIFL, uses the differences in temporally-related states to learn patterns that are correlated with winning or losing a GGP game. These patterns are then used to inform the search. GIFL is simple, robust and improves the quality of play in the majority of games tested. GIFL has been successfully used in the GGP program Maligne.

2011-01-01T00:00:00Z Ruan, Ji http://dl.gi.de/handle/20.500.12116/11190 2018-03-20T10:33:08Z 2011-01-01T00:00:00Z

Reasoning about Time, Action and Knowledge in Multi-Agent Systems Ruan, Ji This thesis is in the area of Multi-Agent Systems (MASs). In a MAS, multiple agents act on their own behalf or of other stakeholders, and key issues here are that they are situated, intelligent, rational and social. They are situated in the sense that they need to be able to sense their environment, intelligent in the sense that they need to model the world around them and make decisions in time and with incomplete information, rational in the sense that they make strategic deliberations when pursuing their own interest, and social in the sense that they are aware of other agents, and their level of intelligence, rationality and social skills. The General Game Playing competition tests the ability of multiple autonomous game playing agents on achieving pre-defined goals. In order to build such agents, one needs to study how agents can represent knowledge (or information) about the world, how their actions may change the world and how a MAS evolves over time due to actions performed by agents. We provide a logic-based account for the specification and verification of MASs, in terms of time, action and knowledge. The contributions are divided into two research themes.The full dissertation can be downloaded at http://ac.jiruan.net/thesis.

2011-01-01T00:00:00Z