Artificial Intelligence: Methodology, Systems, and Applications

This paper proposes a method of learning probabilistic semantic networks which represent visual features and semantic features of object-oriented actions and their contextual activities. In this method, visual motion feature classes of actions and activities are learned by an unsupervised Incremental Probabilistic Latent Component Analysis (IPLCA) and these classes and their semantic tags in the form of case triplets are integrated into probabilistic semantic networks to visually recognize and verbally infer actions in the context of activities. Through experiments using video clips captured with the Kinect sensor, it is shown that the method can learn, recognize and infer object-oriented actions in the context of activities.

In this paper, we present a novel semantic-aware clustering approach for partitioning of experts represented by lists of keywords. A common set of all different keywords is initially formed by pooling all the keywords of all the expert profiles. The semantic distance between each pair of keywords is then calculated and the keywords are partitioned by using a clustering algorithm. Each expert is further represented by a vector of membership degrees of the expert to the different clusters of keywords. The Euclidean distance between each pair of vectors is finally calculated and the experts are clustered by applying a suitable partitioning algorithm.

Online discussions such as forums are very popular and enable participants to read other users’ previous interventions and also to express their own opinions on various subjects of interest. In online discussion forums, there is often a mixture of positive and negative opinions because users may have similar or conflicting opinions on the same subject. Therefore, it is challenging to track the flow of opinions over time in online discussion forums. Past research in the field of opinion propagation has dealt mainly with online social networks. In this paper, by contrast, we address the opinion propagation in discussion forum threads. We proposed a user-level opinion propagation analysis method in the discussion forum threads. This method establishes for a given time step whether the discussion will result in complete agreement between participants or in disparate and even contrary opinions.

This paper presents research on social news recommendation at the biggest social network Facebook. The recommendation strategies which are used are based on content and social trust as the trust is selected as more reliable for recommendation. In order the news to get old in time a decay factor for the score is proposed. Both offline and online evaluation are made as the feedbacks shows that users find the application interesting and useful.

We propose a new approach for Collaborative filtering which is based on Boolean Matrix Factorisation (BMF) and Formal Concept Analysis. In a series of experiments on real data (MovieLens dataset) we compare the approach with an SVD-based one in terms of Mean Average Error (MAE). One of the experimental consequences is that it is enough to have a binary-scaled rating data to obtain almost the same quality in terms of MAE by BMF as for the SVD-based algorithm in case of non-scaled data.

In this paper, a semi-supervised multi-view teaching algorithm based on Bayesian learning is proposed for image segmentation. Beforehand, only a small amount of pixels should be classified by a teacher. The rest of the pixels are used as unlabeled examples. The algorithm uses two views and learns a separate classifier on each view. The first view contains the coordinates of the pixels and the second – the RGB values of the points in the image. Only the weaker classifier is improved by an addition of more examples to the pool of labelled examples. The performance of the algorithm for image segmentation is compared to a supervised classifier and shows very good results.

Rumor spreading as a basic mechanism for information on online social network has a significant impact on people’s life. In Web 2.0 media age, microblog has become a popular means for people to gain new information. Rumor as false information inevitably become a part of this new media. In this study, a modified rumor spreading model called SIRe is introduced, which compared to traditional rumor spreading model, have included the stifler’s broadcasting effect and social intimacy degree between people. In order to verify the reasonableness of SIRe model, real rumor spreading data set and microblog network structure data set are obtained using Sina API. Then rumor predicting results using different models are compared. Finally, for the purpose of finding the characteristics of rumor spreading in community scale, a clustering method is used to discover the user communities. Analysis results have revealed that communities with higher closeness centrality tend to have higher max ratio of spreaders, and scattered immunization is better than centralized immunization, resulting lower max ratio of spreaders. Both the results and explanations are shown in this paper.

In alpine regions, traffic infrastructure may be endangered by snow avalanches. If not protected by physical structures, roads need to be temporarily closed in order to prevent fatal accidents. For assessing the danger of avalanches, local avalanche services use, amongst others, meteorological data measured on a daily basis as well as expert knowledge about avalanche activity. Based on this data, a system for decision support in avalanche warning has been developed. Feasible models were trained using Balanced Random Forests and Weighted Random Forests, yielding a performance useful for human experts. The results are discussed and options for further improvements are pointed out.

This paper presents a research project integrating language technologies and a business intelligence tool that help to discover new knowledge in a very large repository of patient records in Bulgarian language. The ultimate project objective is to accelerate the construction of the Register of diabetic patients in Bulgaria. All the information needed for the Register is available in the outpatient records, collected by the Bulgarian National Health Insurance Fund. We extract automatically from the records’ free text essential entities related to the drug treatment such as drug names, dosages, modes of admission, frequency and treatment duration with precision 95.2%; we classify the records according to the hypothesis “having diabetes” with precision 91.5% and deliver these findings to decision makers in order to improve the public health policy and the management of Bulgarian healthcare system. The experiments are run on the records of about 436,000 diabetic patients.

The Interval Algebra (

IA

) and a fragment of the Region Connection Calculus (

RCC

), namely,

RCC

-8, are the dominant Artificial Intelligence approaches for representing and reasoning about qualitative temporal and topological relations respectively. In this framework, one of the main tasks is to compute the path consistency of a given Qualitative Constraint Network (

QCN

). We concentrate on the partial path consistency checking problem problem of a

QCN

, i.e., the path consistency enforced on an underlying chordal constraint graph of the

QCN

, and propose an algorithm for maintaining or enforcing partial path consistency for growing constraint networks, i.e., networks that grow with new temporal or spatial entities over time. We evaluate our algorithm experimentally with

QCNs

of

IA

and

RCC

-8 and obtain impressive results.

Knowledge Representation and Reasoning has been quite successfull in dealing with the concepts of time and space separately. However, not much has been done in designing qualitative spatiotemporal representation formalisms, let alone reasoning systems for that formalisms. We introduce a qualitative constraint-based spatiotemporal framework using Point Algebra (

PA

), that allows for defining formalisms based on several qualitative spatial constraint languages, such as

RCC

-8, Cardinal Direction Algebra (

CDA

), and Rectangle Algebra (

RA

). We define the notion of a qualitative spatiotemporal constraint network (

QSTCN

) to capture such formalisms, where pairs of spatial networks are associated to every base relation of the underlying network of

PA

. Finally, we analyse the computational properties of our framework and provide algorithms for reasoning with the derived formalisms.

This paper focuses on the main aspects of development of a qualitative system for dynamic content filtering. These aspects include collection of meaningful training data and the feature selection techniques. The Web changes rapidly so the classifier needs to be regularly re-trained. The problem of training data collection is treated as a special case of the focused crawling. A simple and easy-to-tune technique was proposed, implemented and tested. The proposed feature selection technique tends to minimize the feature set size without loss of accuracy and to consider interlinked nature of the Web. This is essential to make a content filtering solution fast and non-burdensome for end users, especially when content filtering is performed using a restricted hardware. Evaluation and comparison of various classifiers and techniques are provided.

We consider the problem of selection the set of features that are the most significant for partitioning two given data sets. The criterion for selection which is to be maximized is the symmetric information distance between distributions of the features subset in the two classes. These distributions are estimated using Bayesian approach for uniform priors, the symmetric information distance is given by the lower estimate for corresponding average risk functional using Rademacher penalty and inequalities from the empirical processes theory. The approach was applied to a real example for selection a set of manufacture process parameters to predict one of two states of the process. It was found that only 2 parameters from 10 were enough to recognize the true state of the process with error level 8%. The set of parameters was found on the base of 550 independent observations in training sample. Performance of the approach was evaluated using 270 independent observations in test sample.

We introduce a novel method for analysis of logical proofs constructed by undergraduate students that employs sequence mining for manipulation with temporal information about all actions that a student performed, and also graph mining for finding frequent subgraphs on different levels of generalisation. We show that this representation allows one to find interesting subgroups of similar solutions and also to detect outlying solutions. Specifically, distribution of errors is not independent of behavioural patterns and we are able to find clusters of erroneous solutions. We also observed significant dependence between time duration and an appearance of the most serious error.

The paper proposed an algorithm for script discrimination using adjacent local binary patterns (ALBP). In the first stage, each letter is modeled according to its height. The real data are extracted from the probability distribution of the letter heights. Then, the gray scale co-occurrence matrix is computed. It is used as a starting point for the feature extraction. The extracted features are classified according to ALBP. Because of the variety in script characteristics, the statistical analysis shows the differences between scripts. Accordingly, the linear discrimination function is proposed to distinct the scripts. The proposed method is tested on the samples of the printed documents, which include Cyrillic and Glagolitic script. The results of experiments are encouraging.

A hybrid approach to automated identification and monitoring of technology trends is presented. The hybrid approach combines methods of ontology based information extraction and statistical methods for processing OBIE results. The key point of the approach is the so called ‘black box’ principle. It is related to identification of trends on the basis of heuristics stemming from an elaborate ontology of a technology trend.

A modified version of particle swarm optimization algorithm is proposed for minimization of maximal real part of a polytopic system eigenvalues. New initialization procedure and special projection operation are introduced to keep all particles working effectively inside a simplex of feasible positions. The algorithm is tested on several benchmarks and statistical evidences for its’ high efficiency are provided.

The present paper describes the back-propagation learning of a partial functional differential equation with reaction-diffusion term. The time-dependent recurrent learning algorithm is developed for a delayed recurrent neural network with the reaction-diffusion term. The proposed simulation methods are illustrated by the back-propagation learning of continuous multilayer Hopfield neural network with a discrete time delay and reaction-diffusion term using the prey-predator system as a teacher signal. The results show that the continuous Hopfield neural networks are able to approximate the signals generated from the predator-prey system with Hopf bifurcation.

In the presented investigation a recently proposed approach for multidimensional data clustering was applied to create a 3D “sound picture” of the data collected by a microphone array antenna. For this purpose records of acoustic pressure at each point (a microphone in the array) collected for a given period of time were used. Features for classification are extracted using overlapping receptive fields based on the model of direction selective cells in the middle temporal (MT) cortex. Next the clustering procedure using Echo state network and subtractive clustering algorithm is applied to separate these receptive fields into proper number of classes. Obtained for each time step two dimensional “sound pictures” were combined to create a 3D representation of dynamic changes in the sound pressure. We compare our results with the sonograms created by the original software of the producer of microphone array. Although our approach did not account for the distance to the noise source, it allows consideration of dynamically changing sounds.

The subject of the presented research is to determine the complete neural procedure for classifying inaccurate information, as given in the form of an interval vector. For such a formulated task, a basic functionality Probabilistic Neural Network was extended upon the interval type of information. As a consequence, a new type of neural network has been proposed. The presented methodology was positively verified using random and benchmark data sets. In addition, a comparative analysis of existing algorithms with similar conditions was made.

Formal Concept Analysis Research Toolbox (FCART) is an integrated environment for knowledge and data engineers with a set of research tools based on Formal Concept Analysis. FCART allows a user to load structured and unstructured data (including texts with various metadata) from heterogeneous data sources into local data storage, compose scaling queries for data snapshots, and then research classical and some innovative FCA artifacts in analytic sessions.

Enabling applications with natural language processing capabilities facilitates user interaction, especially in the case of complex applications such as a mobile banking. In this paper we introduce the steps required for building such a system, starting from the presentation of different alternatives alongside their problems and benefits, and ending up with integrating them within our implemented system. However, one of the main problems with voice recognition models is that they tend to use different approximations and thresholds that aren’t completely reliable; therefore, the best solution consists of combining multiple approaches. Consequently, we opted to implement two different and complementary recognition models, and to detail in the end how their integration within the framework’s architecture leads to encouraging results.

The implementation of effective Semantic Web Services (SWS) platforms allowing the composition and, in general, the orchestration of services presents several problems. Some of them are intrinsic within the formalisms adopted to describe SWS, especially when trying to combine the dynamic aspect of SWS effects and the static nature of their ontological representation in Description Logic (DL). This paper proposes a mapping of OWL-S with a DL action formalism in order to evaluate executability and projection by means of the notion of Contexts.

Recently some novel strategies have been proposed for neural network training that set randomly the weights from input to hidden layer, while weights from hidden to output layer are analytically determined by Moore-Penrose generalised inverse; such non-iterative strategies are appealing since they allow fast learning. Aim of this study is to investigate the performance variability when random projections are used for convenient setting of the input weights: we compare them with state of the art setting i.e. weights randomly chosen according to a continuous uniform distribution. We compare the solutions obtained by different methods testing this approach on some UCI datasets for both regression and classification tasks; this results in a significant performance improvement with respect to conventional method.

Test Case prioritization having a key role to play in prioritizing test scenarios from a pile of scenarios, to best of our knowledge, has not been employed in Agile environment for prioritizing test cases in Automated Test Plans. Considering automated testing in agile environment esp scrum, a prioritized test plan containing high priority test cases is emanated using Genetic Algorithms. This prioritization is courtesy to base factors such as operational profile, test scenario criticality, and faults uncovered by each test case; used to weight test scenarios. Proposed technique exhibits great performance by ameliorating the rate of fault detection by dynamically prioritizing NUnit based test scenarios.

We consider a graph representation for a paragraph of text. It widely uses linguistic theories of discourse to extend the set of edges between vertices corresponding to words. Parse thickets is a set of syntactic parse trees augmented by a number of inter-sentence coreference links and links based on Speech Act and Rhetoric Structures Theories. Similarity of parse thickets is defined by means of intersection operation taking common parts of the thickets. Several approaches to computing intersection of parse thickets are proposed and compared. Projections as approximation means are considered.

This paper presents a modified natural selection based quantum behaved particle swarm optimization (SelQPSO) algorithm for the path planning of mobile robot vehicles. To ensure the global searching and the high efficiency of the QPSO’s searching process, the particle swarms are sorted by fitness and the group of the particles with worst fitness are replaced by the group with best fitness in each iteration of the whole procedure. The effectiveness and feasibility of this algorithm are demonstrated by the results from numerical experiments on well-known benchmark functions. Then, this algorithm is employed to estimate the basic parameters of the mobile robot path planning in the barrier free environment. The convergency of the estimation method versus particle numbers and iteration times is studied with variation of particle dimension. A unary linear regression equation taking the particle number, maximum generation and particle dimension as variables is formulated. The results from experiments for optimal path planning of a mobile robot in complex environment justifies the estimation method.

Knowledge engineering for automated planning is still in its childhood and there has been little work done on how to model planning problems. The prevailing approach in the academic community is using the PDDL language that originated in planning competitions. In contrast, real applications require more modeling flexibility and different modeling languages were designed in order to allow efficient planning. This paper focuses on the role of a domain modeling formalism as an interface between a domain modeler and a planner.

Path planning is typically considered in Artificial Intelligence as a graph searching problem and R* is state-of-the-art algorithm tailored to solve it. The algorithm decomposes given path finding task into the series of subtasks each of which can be easily (in computational sense) solved by well-known methods (such as A*). Parameterized random choice is used to perform the decomposition and as a result R* performance largely depends on the choice of its input parameters. In our work we formulate a range of assumptions concerning possible upper and lower bounds of R* parameters, their interdependency and their influence on R* performance. Then we evaluate these assumptions by running a large number of experiments. As a result we formulate a set of heuristic rules which can be used to initialize the values of R* parameters in a way that leads to algorithm’s best performance.

In this paper, the analysis of American football strategies is by applying Nash equilibrium. Up to the offensive or defensive team-role, each player usually practices the relevant plays for his role; each play is qualified regarding the benefit that could add to the team success. The team’s strategies, that join the individual’s plays, are identified by means of the strategy profiles of a normal game formal setting of American football, and valued by the each player’s payoff function. Hence, the Nash equilibrium strategy profiles can be identified and used for the actions decision making in a match gaming.

Recently, in the field of activity recognition a number of approaches that utilise probabilistic symbolic models have been proposed. Such approaches rely on the combination of symbolic state-space models and probabilistic inference techniques in order to recognise the user activities in situations with uncertainty. One problem with such approaches is the huge state space that can be generated just by a few rules. In this work we investigate the effects of a mechanism for reducing the model complexity on symbolic level. To illustrate the approach, we present one possible strategy and discuss its effects on the model size and the probability of selecting the correct action in an office scenario.