International Journal on Artificial Intelligence Tools

AI planning systems tend to be disembodied and are not situated within the environment for which plans are generated, thus losing information concerning the interaction between the system and its environment. This paper argues that such information may potentially be valuable in constraining plan formulation, and presents both an agent- and domain-independent architecture that extends the classical AI planning framework to take into account context, or the interaction between an autonomous situated planning agent and its environment. The paper describes how context constrains the goals an agent might generate, enables those goals to be prioritised, and constrains plan selection.

We present a new approach in web search engines. The web creates new challenges for information retrieval. The vast improvement in information access is not the only advantage resulting from the keyword search. Additionally, much potential exists for analyzing interests and relationships within the structure of the web. The creation of a hyperlink by the author of a web page explicitly represents a relationship between the source and destination pages which demonstrates the hyperlink structure between web pages. Our web search engine searches not only for the keywords in the web pages, but also for the hyperlink structure between them. Comparing the results of structural web search versus keyword-based search indicates an improved ability to access desired information. We also discuss steps toward mining the queries input to the structural web search engine.

In this study, we compare the use of genetic algorithms (GAs) and other forms of heuristic search in the cryptanalysis of short cryptograms. This paper expands on the work presented at FLAIRS-2003, which established the feasibility of a word-based genetic algorithm (GA) for analyzing short cryptograms.¹ In this study the following search heuristics are compared both theoretically and experimentally: hill-climbing, simulated annealing, word-based and frequency-based genetic algorithms. Although the results reported apply to substitution ciphers in general, we focus in particular on short substitution cryptograms, such as the kind found in newspapers and puzzle books. Short cryptograms present a more challenging form of the problem. The word-based approach uses a relatively small dictionary of frequent words. The frequency-based approaches use frequency data for 2-, 3- and 4-letter sequences. The study shows that all of the optimization algorithms are successful at breaking short cryptograms, but perhaps more significantly, the most important factor in their success appears to be the choice of fitness measure employed.

We present an algorithm for the inference of context-free graph grammars from examples. The algorithm builds on an earlier system for frequent substructure discovery, and is biased toward grammars that minimize description length. Grammar features include recursion, variables and relationships. We present an illustrative example, demonstrate the algorithm's ability to learn in the presence of noise, and show real-world examples.

An intelligent home is likely in the near future. An important ingredient in an intelligent environment such as a home is prediction – of the next low-level action, the next location, and the next high-level task that an inhabitant is likely to perform. In this paper we model inhabitant actions as states in a simple Markov model. We introduce an enhancement to this basic approach, the Task-based Markov model (TMM) method. TMM discovers high-level inhabitant tasks using the supplied unlabeled data. We investigate clustering of actions to identify tasks, and integrate clusters into a hidden Markov model that predicts the next inhabitant action. We validate our approach and observe that for simulated data we achieve good accuracy using both the simple Markov model and the TMM, whereas on real data we see that simple Markov models outperform the TMM. We also perform an analysis of the performance of the HMM in the framework of the TMM when diverse patterns are introduced into the data.

We present an experimental methodology and results for a machine learning approach to learning opening strategy in the game of Go, a game for which the best computer programs play only at the level of an advanced beginning human player. While the evaluation function in most computer Go programs consists of a carefully crafted combination of pattern matchers, expert rules, and selective search, we employ a neural network trained by self-play using temporal difference learning. Our focus is on the sequence of moves made at the beginning of the game. Experimental results indicate that our approach is effective for learning opening strategy, that including higher-level features of the game can improve the quality of the learned evaluation function, and that different input representations of higher-level information can substantially affect performance.

Neural-Symbolic Systems concern the integration of the symbolic and connectionist paradigms of Artificial Intelligence. Distributed knowledge representation is traditionally seen under a symbolic perspective. In this paper, we show how neural networks can represent distributed symbolic knowledge, acting as multi-agent systems with learning capability (a key feature of neural networks). We apply the framework of Connectionist Modal Logics to well-known testbeds for distributed knowledge representation formalisms, namely the muddy children and the wise men puzzles. Finally, we sketch a full solution to these problems by extending our approach to deal with knowledge evolution over time.

One challenge in text processing is the treatment of case insensitive documents such as speech recognition results. The traditional approach is to re-train a language model excluding case-related features. This paper presents an alternative two-step approach whereby a preprocessing module (Step 1) is designed to restore case-sensitive form which is subsequently processed by the original system (Step 2). Step 1 is mainly implemented as a Hidden Markov Model trained on a large raw corpus of case sensitive documents. It is demonstrated that this approach (i) outperforms the feature exclusion approach for named entity tagging, (ii) leads to limited degradation for parsing, relationship extraction and case insensitive question answering, (iii) reduces system complexity, and (iv) has wide applicability: the restored text can be used in both statistical model and rule-based systems.

We present a new keyword extraction algorithm that applies to a single document without using a corpus. Frequent terms are extracted first, then a set of co-occurrences between each term and the frequent terms, i.e., occurrences in the same sentences, is generated. Co-occurrence distribution shows importance of a term in the document as follows. If the probability distribution of co-occurrence between term a and the frequent terms is biased to a particular subset of frequent terms, then term a is likely to be a keyword. The degree of bias of a distribution is measured by the χ²-measure. Our algorithm shows comparable performance to tfidf without using a corpus.

In this paper a mereological theory of frames of reference is presented. It shows that mereology extended by the notion of granularity and approximation is sufficient to provide a theory for location based features of frames of reference. More complex theories, taking also into account orientation and metric properties can be built as extensions of the presented theory. In order to take the hierarchical organization of frames of reference into account we introduce the notion of stratified approximation to facilitate transformations between different levels of granularity. This paper shows that the ontological grounding of the theory of frames of reference into mereology allows us to give a clear semantics to the notion 'degree of parthood' which is central to the notion of approximation. It also shows how epistemic aspects which affect the use of frames of reference help us to understand the feature of epistemic vagueness and the way it affects the notions of approximation and of degree of parthood.

Distributed multimedia e-health applications have a set of specific requirements which must be taken into account if effective use is to be made of the limited resources provided by public telecommunication networks. Moreover, there is an architectural gap between the provision of network-level Quality of Service (QoS) and user requirements of e-health applications. In this paper, we address the problem of bridging this gap from a multi-criteria decision-making perspective in the context of a remote collaborative environment for back pain treatment. We propose adaptable data transport mechanisms, capable of responding to evolving networking, application and user requirements. To this end we present a framework which allows for not only construction of tailored multimedia communication protocols, but also for the inclusion of user requirements in such protocols. This is achieved by equipping our protocol adaptation mechanism with multi-criteria decision making based techniques that can be used to allocate resources according to their ratio-scale priorities in response to detection of system degradation or evolving user preferences. The applicability of our approach is illustrated in three usage scenarios.

Recommender systems combine research from user profiling, information filtering and artificial intelligence to provide users with more intelligent information access. They have proven to be useful in a range of Internet and e-commerce applications. Recent research has shown that a content-based (or case-based) perspective on collaborative filtering for recommendation can provide significant benefits in decision support accuracy over traditional collaborative techniques, particularly as dataset sparsity increases. These benefits derive both from the use of more sophisticated case-based similarity metrics and from the proactive maintenance of item similarity knowledge using data mining. This article presents a natural next step in this ongoing research to improve the quality of recommender systems by validating these findings in the context of more complex models of collaborative filtering, as well as by demonstrating that such techniques also preserve recommendation diversity, one of the key issues affecting traditional recommender systems.

Multi-case-base reasoning (MCBR) extends case-based reasoning to draw on multiple case bases that may address somewhat different tasks. In MCBR, an agent selectively supplements its own case-base as needed, by dispatching problems to external case-bases and using cross-case-base adaptation to adjust their solutions for inter-case-base differences. MCBR has been advocated as a means to facilitate handling large case-bases when storage is limited, or to enable use of distributed case sources. However, this raises an important question: When storage is not an issue, and the cases from all external case sources could be merged into a single case-base, is there any reason for MCBR? This article answers that question with an experimental assessment of how MCBR affects the quality of solutions generated. It demonstrates that for a given local case-base and an external case-base for a task environment that is similar to, but different from, the local task environment, MCBR can improve accuracy compared to merging the case-bases into a single case-base. This improvement holds even if the cross-case-base adaptation method used by MCBR is also applied to the external cases before merging. The article hypothesizes an explanation of this behavior in terms of the ability of MCBR to exploit the tradeoffs between similarity of problems and similarity of solution contexts. It provides experimental evidence to support this hypothesis, and also demonstrates that MCBR is a useful framework for guiding case-base maintenance by selecting cases to add to a case-base.

Space systems are an important part of everyday life. They provide global positioning data, communications, and Earth science data such as weather information. All space systems require satellite operators to ensure high performance and continuous operations in the presence of off-nominal conditions due to space weather and onboard anomalies. Similar to other high-stress, time critical operations (e.g., piloting an aircraft or operating a nuclear power plant), situation awareness is a crucial factor in operator performance during these conditions. Because situation awareness is largely acquired by monitoring large numbers of parameters, it is difficult to rapidly and accurately fuse the data to develop an accurate assessment. To aid operators in this task, we have developed a prototype Multi-Agent Satellite System for Information Fusion (MASSIF) for automated data fusion and situation awareness. This system is based on human cognitive decision-making models and integrates a fuzzy logic system for semantic data processing, Bayesian belief networks for multi-source data fusion and situation assessment, and rule-bases for automatic network construction. This paper describes initial simulation-based results to establish feasibility and baseline performance. We describe knowledge engineering efforts, belief network construction, and operator-interfaces for automated data fusion and situation awareness for a hypothetical geosynchronous satellite.