Marcus Hutter - Böcker

An Introduction to Universal Artificial Intelligence provides the formal underpinning of what it means for an agent to act intelligently in an unknown environment. First presented in Universal Algorithmic Intelligence (Hutter, 2000), UAI offers a framework in which virtually all AI problems can be formulated, and a theory of how to solve them. UAI unifies ideas from sequential decision theory, Bayesian inference, and algorithmic information theory to construct AIXI, an optimal reinforcement learning agent that learns to act optimally in unknown environments. AIXI is the theoretical gold standard for intelligent behavior.The book covers both the theoretical and practical aspects of UAI. Bayesian updating can be done efficiently with context tree weighting, and planning can be approximated by sampling with Monte Carlo tree search. It provides algorithms for the reader to implement, and experimental results to compare against. These algorithms are used to approximate AIXI. The book ends with a philosophical discussion of Artificial General Intelligence: Can super-intelligent agents even be constructed? Is it inevitable that they will be constructed, and what are the potential consequences?This text is suitable for late undergraduate students. It provides an extensive chapter to fill in the required mathematics, probability, information, and computability theory background. You can also visit the author website: http://www.hutter1.net/ai/uaibook2.htm.

An Introduction to Universal Artificial Intelligence provides the formal underpinning of what it means for an agent to act intelligently in an unknown environment. First presented in Universal Algorithmic Intelligence (Hutter, 2000), UAI offers a framework in which virtually all AI problems can be formulated, and a theory of how to solve them. UAI unifies ideas from sequential decision theory, Bayesian inference, and algorithmic information theory to construct AIXI, an optimal reinforcement learning agent that learns to act optimally in unknown environments. AIXI is the theoretical gold standard for intelligent behavior.The book covers both the theoretical and practical aspects of UAI. Bayesian updating can be done efficiently with context tree weighting, and planning can be approximated by sampling with Monte Carlo tree search. It provides algorithms for the reader to implement, and experimental results to compare against. These algorithms are used to approximate AIXI. The book ends with a philosophical discussion of Artificial General Intelligence: Can super-intelligent agents even be constructed? Is it inevitable that they will be constructed, and what are the potential consequences?This text is suitable for late undergraduate students. It provides an extensive chapter to fill in the required mathematics, probability, information, and computability theory background. You can also visit the author website: http://www.hutter1.net/ai/uaibook2.htm.

This book presents sequential decision theory from a novel algorithmic information theory perspective. While the former is suited for active agents in known environments, the latter is suited for passive prediction in unknown environments. The book introduces these two different ideas and removes the limitations by unifying them to one parameter-free theory of an optimal reinforcement learning agent embedded in an unknown environment. Most AI problems can easily be formulated within this theory, reducing the conceptual problems to pure computational ones. Considered problem classes include sequence prediction, strategic games, function minimization, reinforcement and supervised learning. The discussion includes formal definitions of intelligence order relations, the horizon problem and relations to other approaches. One intention of this book is to excite a broader AI audience about abstract algorithmic information theory concepts, and conversely to inform theorists about exciting applications to AI.

This volume contains the papers presented at the 18th International Conf- ence on Algorithmic Learning Theory (ALT 2007), which was held in Sendai (Japan) during October 1–4, 2007. The main objective of the conference was to provide an interdisciplinary forum for high-quality talks with a strong theore- cal background and scienti?c interchange in areas such as query models, on-line learning, inductive inference, algorithmic forecasting, boosting, support vector machines, kernel methods, complexity and learning, reinforcement learning, - supervised learning and grammatical inference. The conference was co-located with the Tenth International Conference on Discovery Science (DS 2007). This volume includes 25 technical contributions that were selected from 50 submissions by the ProgramCommittee. It also contains descriptions of the ?ve invited talks of ALT and DS; longer versions of the DS papers are available in the proceedings of DS 2007. These invited talks were presented to the audience of both conferences in joint sessions.

This book presents sequential decision theory from a novel algorithmic information theory perspective. While the former is suited for active agents in known environments, the latter is suited for passive prediction in unknown environments. The book introduces these two different ideas and removes the limitations by unifying them to one parameter-free theory of an optimal reinforcement learning agent embedded in an unknown environment. Most AI problems can easily be formulated within this theory, reducing the conceptual problems to pure computational ones. Considered problem classes include sequence prediction, strategic games, function minimization, reinforcement and supervised learning. The discussion includes formal definitions of intelligence order relations, the horizon problem and relations to other approaches. One intention of this book is to excite a broader AI audience about abstract algorithmic information theory concepts, and conversely to inform theorists about exciting applications to AI.

This volume contains the papers presented at the 21st International Conf- ence on Algorithmic Learning Theory (ALT 2010), which was held in Canberra, Australia, October 6–8, 2010. The conference was co-located with the 13th - ternational Conference on Discovery Science (DS 2010) and with the Machine Learning Summer School, which was held just before ALT 2010. The tech- cal program of ALT 2010, contained 26 papers selected from 44 submissions and ?ve invited talks. The invited talks were presented in joint sessions of both conferences. ALT 2010 was dedicated to the theoretical foundations of machine learning and took place on the campus of the Australian National University, Canberra, Australia. ALT provides a forum for high-quality talks with a strong theore- cal background and scienti?c interchange in areas such as inductive inference, universal prediction, teaching models, grammatical inference, formal languages, inductive logic programming, query learning, complexity of learning, on-line learning and relative loss bounds, semi-supervised and unsupervised learning, clustering,activelearning,statisticallearning,supportvectormachines,Vapnik- Chervonenkisdimension,probablyapproximatelycorrectlearning,Bayesianand causal networks, boosting and bagging, information-based methods, minimum descriptionlength,Kolmogorovcomplexity,kernels,graphlearning,decisiontree methods, Markov decision processes, reinforcement learning, and real-world - plications of algorithmic learning theory. DS 2010 was the 13th International Conference on Discovery Science and focused on the development and analysis of methods for intelligent data an- ysis, knowledge discovery and machine learning, as well as their application to scienti?c knowledge discovery. As is the tradition, it wasco-located and held in parallel with Algorithmic Learning Theory.