We propose a memory-based approach to the problem of goal-schema recognition. We use a generic episodic memory module to perform incremental goal schema recognition and to build the plan library. Unlike other case-based plan recognizers it does not require complete knowledge of the planning domain or the ability to record intermediate planning states. Similarity of plans is computed incrementally using a semantic matcher that considers the type and parameters of the observed actions.  We evaluate this approach on two datasets and show that it is able to achieve similar or better performance compared to a statistical approach, but offers important advantages: plan library is acquired incrementally and the memory structure it builds is multi-functional and can be used for other tasks such as plan generation or classification.

Coinduction has recently been introduced into logic programming by Simon et al. The resulting paradigm, termed coinductive logic programming (co-LP), allows one to model and reason about infinite processes and objects. Co-LP extended with negation has many interesting applications: for instance in developing top-down, goaldirected evaluation strategies for Answer Set Programming. In this paper we show yet another application of co-LP, namely, elegantly realizing Boolean SAT solvers

Can't Stop is a jeopardy stochastic game played on an octagonal game board with four six-sided dice. Optimal strategies have been computed for some simplified versions of Can't Stop by employing retrograde analysis and value iteration combined with Newton's method. These computations result in databases that map game positions to optimal moves. Solving the original game, however, is infeasible with current techniques and technology. This paper describes the creation of heuristic strategies for solitaire Can't Stop by generalizing an existing heuristic and using genetic algorithms to optimize the generalized parameters. The resulting heuristics are easy to use and outperform the original heuristic by 19%. Results of the genetic algorithm are compared to the known optimal results for smaller versions of Can't Stop, and data is presented showing the relative insensitivity of the particular genetic algorithm used to the balance between reduced noise and increased population diversity.

A reinforcement learning agent involved in life-long learning in a complex and dynamic environment has to have the ability to utilize control knowledge acquired in one situation in novel contexts. As part of this, it is important for the learning agent not only to be able to learn a new skill for a specific instance of a task but also to identify similar tasks, form a reusable skill and representational abstractions for the corresponding ''task type'', and to apply these abstractions in new, previously unseen contexts. This paper presents a new approach to policy generalization that derives an abstract policy for a set of similar tasks (a ''task type'') by constructing a partial policy homomorphism from a set of basic policies learned for previously seen task instances. The resulting generalized policy can then be applied in new contexts to address new instances of similar tasks. As opposed to many recent approaches in lifelong learning systems, this approach allows to identify similar tasks based on the functional characteristics of the corresponding skills and provides a means of transferring the learned knowledge to new situations without the need for complete knowledge of the state space and the system dynamics in the new environment. To illustrate the new policy generalization method and to demonstrate its ability to reuse the gained knowledge in new contexts, it is applied to a set of grid world examples.

Instance classification using machine learning techniques has numerous applications, from automation to medical diagnosis. In many problem domains, such as spam filtering, classification must be performed quickly across large datasets. In this paper we begin with machine learning techniques based on the naive Bayes classification and attempt to improve classification performance by taking into account attribute confidence intervals.  Our prediction functions operate over nominal datasets and retain the asymptotic complexity of one-pass learning and prediction functions. We present preliminary results indicating a modest, albeit inconsistent improvement over the naive Bayes classifier alone.

Ontology mapping plays an important role in interoperability over ontologies. Many researchers have proposed algorithms and tools for (semi-)automatically mapping one concept to another concept. Among them, WordNet is widely used as the domain knowledge support in the mapping process. To our knowledge, however, most of them only use synonym, hypernym and hyponym relations in WordNet and the actual meanings provided in natural English(as gloss) are often ignored. In this paper, we treat the concepts(c) as English words (w) and propose an ontology mapping technique where we use the meanings of the words as given in Wordnet (in English) for semantic mapping by constructing their parse trees first and simplifying them for computing similarity measures. Our experimental results show that our method performs better in Recall and F1-Measure than many techniques reported in the literature.

What is the biggest difference between playing a game against a human or against a computer generated player? Why do many people believe it is more challenging to play with humans than playing with an artificial player? The big success of massive multiplayer games and the huge number of so-called "LAN parties", where players meet and play with each other, seems to be related to the human demeanor of the players. All this indicates, that the current state of game AI is unsatisfactory compared to the performance of human players. This paper introduces a tool for analyzing basic computer games with incorporated AI modules which store strategies for performing the behavior of artificial players. This sets the stage for a systematic evaluation and refinement of rule based game AI.

Augmenting probabilities to conditional logic yields an expressive mechanism for representing uncertainty. The principle of optimum entropy allows one to reason in probabilistic logic in an information-theoretic optimal way by completing the given information as unbiasedly as possible. In this paper, we introduce the MEcore system that realises the core functionalities for an intelligent agent reasoning at optimum entropy and that provides powerful mechanisms for belief management operations like revision, update, diagnosis, or hypothetical what-if-analysis.

We present a modified navigation mesh generation algorithm that allows the mesh to be dynamically altered at runtime. We accomplish this using an extension to the existing spatial decomposition algorithm ASFV (Adaptive Space Filling Volumes) that will allow the algorithm to dynamically adapt to changes to the underlying world geometry without having to rebuild the entire spatial decomposition. This is accomplished by providing two algorithms to deal with alterations to the world. The ability is provided to add arbitrary obstructions into what was negative space and then to build a new correct spatial decomposition around the new obstruction. Functionality is also provided to remove existing obstructions and then to build up new decompositions to fill in the newly created negative space. Finally, we show via an experiment that our dynamic extensions to ASFV reduces the cost of correcting an invalidated decomposition by 90% or more.

In this paper, we propose a universal solution to web search and web browsing on handheld devices for visually impaired people. For this purpose, we propose (1) to automatically cluster web page results and (2) to summarize all the information in web pages so that speech-to-speech interaction is used efficiently to access information.