International Journal on Artificial Intelligence Tools

Knowing the norms of a domain is crucial, but there exists no repository of norms. We propose a method to extract them from texts: narrative texts generally do not describe a norm, but rather the discrepancies between what actually happened, and the corresponding normal sequence of events. Answers about the causes of an event often reveal the implicit norms. We apply this idea to the domain of driving, and validate it by designing algorithms that identify, in a text, the "basic" norms to which it refers implicitly.

This paper describes work on methods that evaluate algebraic solutions to word problems in physics. Many current tutoring systems rely on substantial scaffolding and consequently require students to completely describe every variable used in the solution. A heuristic, based on constraint propagation, capable of inferring the description of variables (i.e., the possible dimensions and physics concepts) is shown to be highly reliable on three real world data sets, one covering a few problems with a small number of student answers and two others covering a large class of problems (~100) with a large number of student answers (~11,000). The heuristic uniquely determines the dimensions of all the variables in 91–92% of the equation sets. By asking the student for dimension information about one variable, an additional 3% of the sets can be determined. An ITS tutoring system can use this heuristic to reason about a student's answers even when the scaffolding and context are removed.

This paper presents a hybrid algorithm that combines local search and constraint programming techniques to solve a network routing problem. The problem considered is that of routing traffic demands from a set of requests over a network with limited capacity so as to minimise the cost of any unrouted demands. The hybridisation is twofold: pure local search is used to find a good cost bound for a subsequent branch-and-bound optimisation phase, with local search again applied at the nodes of the branch-and-bound search tree. Constraint propagation occurs in the search tree to reduce the domains of the decision variable, using a set of constraints that are independent of the action of local search at the nodes. In contrast to previous constraint programming/local search hybridisations, here local search is used to satisfy the hard problem constraints, while optimisation is handled in the framework of constraint programming. The resulting algorithm is incomplete, but is shown to compare favourably with a complete approach to this problem.

Traditional approaches to semantics give a set-theoretic account of category conjunction, in which an item will be a member of a conjunction A&B only if it is a member of the single category A and a member of the single category B. However, a number of studies have found that people do not follow this set-theoretic account when classifying items in certain natural-language conjunctions. For example, Hampton¹ found that while people tend to classify 'toucans' as non-members of the single category 'pet', they classify 'toucans' as members of the conjunction 'pet bird'. This paper describes a general approach to category conjunction that gives a logically consistent explanation for the occurrence of overextension in these conjunctions. The paper also describes a computational model implementing this approach, and an experiment using conjunctions of controlled, laboratory-learned categories to test this model. This computational model gave a good fit to both classification and overextension results in the experiment. The approach to conjunction implemented in this model may provide a useful tool for AI models of classification, allowing them to reason about category conjunction in a way that reflects human reasoning.

ABSURDIST II, an extension to ABSURDIST, is an algorithm using attributed graph matching to find translations between conceptual systems. It uses information about the internal structure of systems by itself, or in combination with external information about concept similarities across systems. It supports systems with multiple types of weighted or unweighted, directed or undirected relations between concepts. The algorithm exploits graph sparsity to improve computational efficiency. We present the results of experiments with a number of conceptual systems, including artificially constructed random graphs with introduced distortions.

Much of current data mining research is focused on discovering sets of attributes that discriminate data entities into classes, such as shopping trends for a particular demographic group. In contrast, we are working to develop data mining techniques to discover patterns consisting of complex relationships between entities. Our research is particularly applicable to domains in which the data is event-driven or relationally structured. In this paper we present approaches to address two related challenges; the need to assimilate incremental data updates and the need to mine monolithic datasets. Many realistic problems are continuous in nature and therefore require a data mining approach that can evolve discovered knowledge over time. Similarly, many problems present data sets that are too large to fit into dynamic memory on conventional computer systems. We address incremental data mining by introducing a mechanism for summarizing discoveries from previous data increments so that the globally-best patterns can be computed by mining only the new data increment. To address monolithic datasets we introduce a technique by which these datasets can be partitioned and mined serially with minimal impact on the result quality. We present applications of our work in both the counter-terrorism and bioinformatics domains.

In this article we will present a classification and an analysis, by means of Applicative and Combinatory Categorial Grammar (ACCG), of relative, completive and indirect interrogative propositions in French introduced by "que" and "qui". Applicative and Combinatory Categorial Grammar is a generalization of standard Categorial Grammar. It is represented by a canonical association between Steedman's Combinatory Categorial rules and Curry's combinators. This model is included in the general framework of Applicative and Cognitive Grammar with three levels of representation: (i) phenotype (concatened expressions); (ii) genotype (applicative expressions); (iii) the cognitive representations (meaning of linguistic predicates). We are interested only in phenotype and genotype levels.

The Intensive Care Unit (ICU) is a challenging environment to both patient and caregiver. Continued shortages in staffing increase risk to patients. To evaluate the use of intelligent systems in the improvement of patient care, an intelligent agent was developed to regulate ICU patient sedation. A temporal differencing form of reinforcement learning was used to train the agent in the administration of intravenous propofol in simulated ICU patients. The agent utilized a well-studied pharmacokinetic model to calculate the distribution of drug within the patient. Pharmacodynamics were then estimated for the drug effect. A derivative of the electroencephalograms, the bispectral index, served as the system control variable. The agent demonstrated satisfactory control of the simulated patient's consciousness level in static and dynamic setpoint conditions. The agent demonstrated superior stability and responsiveness when compared to a well-tuned PID controller, the control method of choice in closed-loop sedation control literature.

We develop the default logic for pragmatic analysis of natural language queries. Ambiguous sentences are considered so that each set of meanings is assigned an extension of default system. The approach is applied to natural language question answering, where even a correct semantic representation needs to be modified in accordance to the set of default rules to better match a knowledge domain.

Dealing with constraints due to obstacles is an important topic in constraint-based spatial clustering. In this paper, we proposed the DBRS_O method to identify clusters in the presence of intersected obstacles. Without doing any preprocessing, DBRS_O processes the constraints during clustering. DBRS_O can also avoid unnecessary computations when obstacles do not affect the clustering result. As well, DBRS_O can find clusters with arbitrary shapes, varying densities, deal with significant non-spatial attributes and handle large datasets.

In this article, we study the subspace function granularity and present a method to estimate the sharing distance and the optimal population size. To achieve multimodal function optimization, niching techniques diversify the population of Evolutionary Algorithms (EA) and encourage heterogeneous convergence to multiple optima. The key to a successful diversification is effective resource sharing. Without knowing the fitness landscape, resource sharing is usually determined by uninformative assumptions on the number of peaks. Using the Probably Approximately Correct (PAC) learning theory and the ∊-cover concept, a PAC neighborhood for a set of samples is derived. Within this neighborhood, we sample the fitness landscape and compute the subspace Fitness Distance Correlation (FDC) coefficients. Using the estimated granularity feature of the fitness landscape, the sharing distance and the population size are determined. Experiments demonstrate that by using the estimated population size and sharing distance an Evolutionary Algorithm successfully identifies multiple optima.

Recommender systems bring together ideas from information retrieval and filtering, user profiling, and machine learning in an attempt to provide users with more proactive and personalized information systems. Forwarded as a response to the information overload problem, recommender systems have enjoyed considerable theoretical and practical successes, with a range of core techniques and a compelling array of evaluation studies to demonstrate success in many real-world domains. That said, there is much yet to understand about the strengths and weaknesses of recommender systems technologies and in this article, we make a fine-grained analysis of a successful case-based recommendation approach. We describe a detailed, fine-grained ablation study of similarity knowledge and similarity metric contributions to improved system performance. In particular, we extend our earlier analyses to examine how measures of interestingness can be used to identify and analyse relative contributions of segments of similarity knowledge. We gauge the strengths and weaknesses of knowledge components and discuss future work as well as implications for research in the area.

Computational modeling of the human visual system is of current interest to developers of artificial vision systems, primarily because a biologically-inspired model can offer solutions to otherwise intractable image understanding problems. The purpose of this study is to present a biologically-inspired model of selective perception that augments a stimulus-driven approach with a high-level algorithm that takes into account particularly informative regions in the scene. The representation is compact and given in the form of a topographic map of relative perceptual conspicuity values. Other recent attempts at compact scene representation consider only low-level information that codes salient features such as color, edge, and luminance values. The previous attempts do not correlate well with subjects' fixation locations during viewing of complex images or natural scenes. This study uses high-level information in the form of figure/ground segmentation, potential object detection, and task-specific location bias. The results correlate well with the fixation densities of human viewers of natural scenes, and can be used as a preprocessing module for image understanding or intelligent surveillance applications.

Algorithms reducing the storage requirement of the nearest neighbor classifier (NNC) can be divided into three main categories: Fast searching algorithms, Instance-based learning algorithms and Prototype based algorithms. We propose an algorithm, LVQPRU, for pruning NNC prototype vectors and a compact classifier with good performance is obtained. The basic condensing algorithm is applied to the initial prototypes to speed up the learning process. The learning vector quantization (LVQ) algorithm is utilized to fine tune the remaining prototypes during each pruning iteration. We evaluate LVQPRU on several data sets along with 12 other algorithms using ten-fold cross-validation. Simulation results show that the proposed algorithm has high generalization accuracy and good storage reduction ratios.

A method which modifies the objective function used far designing neural network classifiers is presented. The classical mean-square error criteria is relaxed by introducing two types of local error bias which are treated like free parameters. Open and closed form solutions are given for finding these bias parameters. The new objective function is seamlessly integrated into existing training algorithms such as back propagation (BP), output weight optimization (OWO), and hidden weight optimization (HWO). The resulting algorithms are successfully applied in training neural net classifiers having a linear final layer. Classifiers are trained and tested on several data sets from pattern recognition applications. Improvement over classical iterative regression methods is clearly demonstrated.

AI agents and robots that can adapt and handle multiple tasks in real time promise to be a powerful tool. To address the control challenges involved in such systems, the underlying control approach has to take into account the important sensory information. Modern sensors, however, can generate huge amounts of data, rendering the processing and representation of all sensor data in real time computationally intractable. This issue can be addressed by developing task-specific focus of attention strategies that limit the sensory data that is processed at any point in time to the data relevant for the given task. Alone, however, this mechanism is not adequate for solving complex tasks since the robot also has to maintain selected pieces of past information. This necessitates AI agents and robots to have the capability to remember significant past events that are required for task completion. This paper presents an approach that considers focus of attention as a problem of selecting controller and feature pairs to be processed at any given point in time to optimize system performance. This approach is further extended by incorporating short term memory and a learned memory management policy. The result is a system that learns control, sensing, and memory policies that are task-specific and adaptable to real world situations using feedback from the world in a reinforcement learning framework. The approach is illustrated using table cleaning, sorting, stacking, and copying tasks in the blocks world domain.

We present results from experiments in using several pitch representations for jazz-oriented musical tasks performed by a recurrent neural network. We have run experiments with several kinds of recurrent networks for this purpose, and have found that Long Short-term Memory networks provide the best results. We show that a new pitch representation called Circles of Thirds works as well as two other published representations for these tasks, yet it is more succinct and enables faster learning. We then discuss limited results using other types of networks on the same tasks.

Nobody would deny that music may evoke deep and profound emotions. In this paper, we present a perceptual music composition system that aims at the controlled manipulation of a user's emotional state. In contrast to traditional composing techniques, the single components of a composition, such as melody, harmony, rhythm and instrumentation, are selected and combined in a user-specific manner without requiring the user to continuously provide comments on the music employing input devices, such as keyboard or mouse.

Corrosion is one of the degrading mechanisms that greatly affect the structural integrity of aerospace components. Corrosion begins with small imperfections in the given material referred to as "pits". Over time, and with exposure to the environment and loading stresses, these pits grow into structural cracks. These cracks, under further exposure to elements and stress, will eventually lead to material failure. To better understand how corrosion growth process takes place in materials, research is being conducted to simulate corrosion pit growth process using computational intelligence methods. The objective of this study is to develop a discrete dynamic model using cellular automata to simulate the pitting corrosion growth process in aircraft materials, and see how it affects the structural integrity.