Collective classification has been intensively studied due to its impact in many important applications, such as web mining, bioinformatics and citation analysis. Collective classification approaches exploit the dependencies of a group of linked objects whose class labels are correlated and need to be predicted simultaneously. In this paper, we focus on studying the collective classification problem in heterogeneous networks, which involves multiple types of data objects interconnected by multiple types of links. Intuitively, two objects are correlated if they are linked by many paths in the network. However, most existing approaches measure the dependencies among objects through directly links or indirect links without considering the different semantic meanings behind different paths. In this paper, we study the collective classification problem taht is defined among the same type of objects in heterogenous networks. Moreover, by considering different linkage paths in the network, one can capture the subtlety of different types of dependencies among objects. We introduce the concept of meta-path based dependencies among objects, where a meta path is a path consisting a certain sequence of linke types. We show that the quality of collective classification results strongly depends upon the meta paths used. To accommodate the large network size, a novel solution, called HCC (meta-path based Heterogenous Collective Classification), is developed to effectively assign labels to a group of instances that are interconnected through different meta-paths. The proposed HCC model can capture different types of dependencies among objects with respect to different meta paths. Empirical studies on real-world networks demonstrate that effectiveness of the proposed meta path-based collective classification approach.

Multi-task learning leverages shared information among data sets to improve the learning performance of individual tasks. The paper applies this framework for data where each task is a phase-shifted periodic time series. In particular, we develop a novel Bayesian nonparametric model capturing a mixture of Gaussian processes where each task is a sum of a group-specific function and a component capturing individual variation, in addition to each task being phase shifted. We develop an efficient \textsc{em} algorithm to learn the parameters of the model. As a special case we obtain the Gaussian mixture model and \textsc{em} algorithm for phased-shifted periodic time series. Furthermore, we extend the proposed model by using a Dirichlet Process prior and thereby leading to an infinite mixture model that is capable of doing automatic model selection. A Variational Bayesian approach is developed for inference in this model. Experiments in regression, classification and class discovery demonstrate the performance of the proposed models using both synthetic data and real-world time series data from astrophysics. Our methods are particularly useful when the time series are sparsely and non-synchronously sampled.

Instance-based regression methods generate solutions from prior solutions within a neighborhood of the input query. Their performance depends on both neighborhood selection criteria and on the method for generating new solutions from the values of prior instances. This paper proposes a new approach to addressing both problems, in which solutions are generated by an ensemble of solutions of local linear regression models built for a collection of "stretched" neighborhoods of the query. Each neighborhood is generated by relaxing a different dimension of the problem space. The rationale is to enable major change trends along that dimension to have increased influence on the corresponding model. The approach is evaluated for two candidate relaxation approaches, gradient-based and based on fixed profiles, and compared to baselines of k-NN and using a radius-based spherical neighborhood in n-dimensional space. Results in four test domains show up to 15 percent improvement over baselines, and suggest that the approach could be particularly useful in domains for which the space of prior instances is sparse.

In the present paper we propose an unsupervised attribute ranking method based on evaluating the quality of clustering that each attribute produces by partitioning the data into subsets according to its values. We use the Minimum Description Length (MDL) principle to evaluate the quality of clustering and describe an algorithm for attribute ranking and a related clustering algorithm. Both algorithms are empirically evaluated on benchmark data sets. The experiments show that the MDL-based ranking performs closely to the supervised information gain ranking and thus improves the performance of the EM and k-means clustering algorithms in purely unsupervised setting.

Probabilistic conditional logics offer a rich and well-founded framework for designing expert systems. The factorization of their maximum entropy models has several interesting applications. In this paper a general factorization is derived providing a more rigorous proof than in previous work. It yields an approach to extend Iterative Scaling variants to deterministic knowledge bases. Subsequently the connection to Markov Random Fields is revisited.

High-dimensional indexing is an important area of current research, especially for range and kNN queries. This work introduces clustering for the sake of indexing. The goal is to develop new clustering methods designed to optimize the data partitioning for an indexing-specific tree structure instead of finding data distribution-based clusters. We focus on iDistance, a state-of-the-art high-dimensional indexing method, and take a basic approach to solving this new problem. By utilizing spherical clusters in an unsupervised Expectation Maximization algorithm dependent upon local density and cluster overlap, we create a partitioning of the space providing balanced segmentation for a B+-tree. We also look at the novel idea of reclustering for a specific indexing method by taking the output of one clustering method and reclustering it for use in an index. The algorithms are then tested and evaluated based on our error metric and iDistance query performance.

K-means is undoubtedly the most widely used partitional clustering algorithm. Unfortunately, the nearest neighbor search step of this algorithm can be computationally expensive, as the distance between each input vector and all cluster centers need to be calculated. To accelerate this step, a computationally inexpensive distance estimation method can be tried first, resulting in the rejection of candidate centers that cannot possibly be the nearest center to the input vector under consideration. This way, the computational requirements of the search can be reduced as most of the full distance computations become unnecessary. In this paper, a fast nearest neighbor search method that rejects impossible centers to accelerate the k-means clustering algorithm is presented. Our method uses geometrical relations among the input vectors and the cluster centers to reject many unlikely centers that are not typically rejected by similar approaches. Experimental results show that the method can reduce the number of distance computations significantly without degrading the clustering accuracy.

In student modeling, the concept of "mastery learning" i.e. that a student continues to learn a skill till mastery is attained is important. Usually, mastery is defined in terms of most recent student performance. This is also the case with models such as Knowledge Tracing which estimate knowledge solely based on patterns of questions a student gets correct and the task usually is to predict immediate next action of the student. In retrospect however, it is not clear if this is a good definition of mastery since it is perhaps more useful to focus more on student retention over a longer period of time. This paper improves a recently introduced model by Wang and Beck that predicts long term student performance by clustering the students and generating multiple predictions by using a recently developed ensemble technique. Another contribution is that we introduce a novel clustering algorithm we call "Regularity Clustering" and show that it is superior in the task of predicting student retention over more popular techniques such as k-means and Spectral Clustering.

In this work, a methodology is developed to detect sentient actors in spoken stories. Meta-tags are then saved to XML files associated with the audio files. A recursive approach is used to find actor candidates and features which are then classified using machine learning approaches. Results of the study indicate that the methodology performed well on a narrative based corpus of children’s stories. Using Support Vector Machines for classification, an F-measure accuracy score of 86% was achieved for both named and unnamed entities. Additionally, feature analysis indicated that speech features were very useful when detecting unnamed actors.

Currently, signature based Intrusion Detection Systems (IDS) approaches are inadequate to address threats posed to networked systems by zero-day exploits. Statistical machine learning techniques offer a great opportunity to mitigate these threats. However, at this point, statistical based IDS systems are not mature enough to be implemented in realtime systems and the techniques to be used are not sufficiently understood. This study focuses on a recently expanded corpus for IDS analysis. Feature analysis and Support Vector Machines classification are performed to obtain a better understanding of the corpus and to establish a baseline set of results which can be used by other studies for comparison. Results of the classification and feature analysis are discussed.