Catálogo de publicaciones - libros

This paper proposes an adaptive subcluster merging (ASM) based clustering algorithm. The ASM algorithm has two stages: subcluster partition and subcluster merging. Specifically, it first applies local expanding with variance constraint to partition subclusters with uniform granularity, and then it adaptively merges the subclusters into clusters with the notion of density. Through these two stages, ASM algorithm can identify clusters of heterogeneous structures. The feasibility of the algorithm has been successfully tested on both synthetic and real-world data sets. Comparative experimental studies of various clustering algorithms are also performed. The results demonstrate that the proposed algorithm performs better than K-means, complete-link hierarchial, density-based and maximum variance algorithms.

Video Association Mining(VAM) is the process of discovering associations in a given video. Two key phases of VAM are (i) Transformation and (ii) Frequent Temporal Pattern Mining. The transformation phase converts the original input video to an alternate transactional format, namely a cluster sequence. Frequent temporal pattern mining phase concerns the generation of patterns subject to the temporal distance and support thresholds. The paper addresses the issue of frequent temporal pattern mining and studies algorithms for the same. The existing Apriori based algorithm is compared with three other approaches highlighting the case specific situations suited by each.

Ubiquitous Healthcare (u-healthcare) which focuses on automated applications that can provide healthcare to citizens anywhere/anytime using wired and wireless mobile technologies is becoming increasingly important. Ubiquitous healthcare data provides a mine of hidden knowledge which can be exploited in preventive care and “wellness” recommendations. Data mining is therefore a significant aspect of such systems. Distributed Data mining (DDM) techniques for knowledge discovery from databases help in the thorough analysis of data collected from healthcare facilities enabling efficient decision-making and strategic planning. This paper presents and discusses the development of a prototype ubiquitous healthcare system. The prospects for integrating data mining into this framework are studied using a distributed data mining system. The DDM system employs a mixture modelling mechanism for data partitioning. Initial results with some standard medical databases offer a plausible outlook for future integration.

The judgment that whether an email is spam or non-spam may vary from person to person. Different individuals can have totally different responses to the same email based on their preferences. This paper presents an innovative approach that incorporates user preferences to construct an anti-spam mail system, which is different from the conventional content-based approaches. We build a user preference ontology to formally represent the important concepts and rules derived from a data mining process. Then we use an inference engine that utilizes the knowledge to predict the user’s action on new incoming emails. We also suggest a new rule optimization procedure inspired from logic synthesis to improve comprehensibility and exclude redundant rules. Experimental results showed that our user preference based architecture achieved good performance and the rules derived from the architecture and the optimization method have better quality in terms of comprehensibility.

In this paper we introduce a system that automatically summarizes multiple biomedical documents relevant to a question. The system extracts biomedical and general concepts by utilizing concept-level knowledge from domain-specific and domain-independent sources. Semantic role labeling, semantic subgraph-based sentence selection and automatic post-editing are involved in the process of finding the information need. Due to the absence of expert-written summaries of biomedical documents, we propose an approximate evaluation by taking MEDLINE abstracts as expert-written summaries. Evaluation results indicate that our system does help in answering questions and the automatically generated summaries are comparable to abstracts of biomedical articles, as evaluated using the ROUGE measure.

Different types of rules are mined from transaction databases often with the goal of improving sales and services. In this paper, we link the interestingness of rules with the context of business marketing. We consider the profits generated from some specific marketing strategies that are developed based on particular discovered rules. This leads to a profit-based business model for evaluating rule interestingness. With this additional utility, we investigate some relationships between different marketing strategies and fundamental properties of rules for profit increasing.

The present paper is about a framework of formal topological reasoning originating from the fundamental work of Moss and Parikh relating to this. We add a set of unary operators involving sets to that system. This new means of expression gives us considerably more expressive power with regard to spatial operations, but the accompanying logic remains sound and semantically complete with respect to the class of all subset spaces that are enriched accordingly. Moreover, the new logic turns out to be decidable. We prove these results by relying heavily on a particular extension of the common modal formalism, viz hybrid logic.

Various families of Neural Networks (NN) have been used in the study and development of the field of Artificial Intelligence (AI). More recently, the Hodgkin-Huxley (HH) has attracted a fair bit of attention. Apart from the HH neuron possessing desirable “computing” properties, it also can be used to reasonably model biological phenomena, and in particular, in modeling neurons which are “synchronized/desynchronized”. This paper, which we believe is a of pioneering sort, derives some of the analytic/learning properties of the HH neuron, and neural network.

The HH Neuron is a nonlinear system with two equilibrium states: A fixed point and a limit cycle. Both of them co-exist and are stable. The behavior of this neuron can be switched between these two equilibria, namely and respectively, by using a perturbation method. The change from spiking to resting is referred to as . In this paper, we analytically prove the existence of a brief excitation (input) which, when delivered to the HH neuron during its repetitively firing state, annihilates its spikes. We also formally derive the characteristics of this brief excitation. We thus believe that our analysis of the HH neuron has practical implications in clinical applications, especially in the case of the of neuronal populations.

Importance sampling-based algorithms are a popular alternative when Bayesian network models are too large or too complex for exact algorithms. However, importance sampling is sensitive to the quality of the importance function. A bad importance function often leads to much oscillation in the sample weights, and, hence, poor estimation of the posterior probability distribution. To address this problem, we propose the technique to adjust the samples with extremely large or extremely small weights, which contribute most to the variance of an importance sampling estimator. Our results show that when we adopt this technique in the EPIS-BN algorithm[14], adaptive split-rejection control helps to achieve significantly better results.

When using Bayesian networks, practitioners often express constraints among variables by conditioning a common child node to induce the desired distribution. For example, an ‘or’ constraint can be easily expressed by a node modelling a logical ‘or’ of its parents’ values being conditioned to true. This has the desired effect that at least one parent must be true. However, conditioning also alters the distributions of further ancestors in the network. In this paper we argue that these are undesirable when constraints are added during model design. We describe a method called shielding to remove these side effects while remaining within the directed language of Bayesian networks. This method is then compared to chain graphs which allow undirected and directed edges and which model equivalent distributions. Thus, in addition to solving this common modelling problem, shielded Bayesian networks provide a novel method for implementing chain graphs with existing Bayesian network tools.