We describe and evaluate experimentally a method for clustering words according to their dis- tribution in particular syntactic contexts. Words are represented by the relative frequency distributions of contexts in which they appear, and relative entropy between those distributions is used as the similarity measure for clustering. Clusters are represented by average context distributions derived from the given words according to their probabilities of cluster membership. In many cases, the clusters can be thought of as encoding coarse sense distinctions. Deterministic annealing is used to find lowest distortion sets of clusters: as the an- nealing parameter increases, existing clusters become unstable and subdivide, yielding a hierarchi- cal "soft" clustering of the data. Clusters are used as the basis for class models of word coocurrence, and the models evaluated with respect to held-out test data.

Describing a video sequence in terms of a small number of coherently moving segments is useful for tasks ranging from video compression to event perception. A promising approach is to view the motion segmentation problem in a mixture estimation framework. However, existing formulations generally use only the motion data and thus fail to make use of static cues when segmenting the sequence. Furthermore, the number of models is either specified in advance or estimated outside the mixturemodel framework. In this work we address both of these issues. We show how to add spatial constraints to the mixture formulations and present a variant of the EM algorithm that makes use of both the form and the motion constraints. Moreover this algorithm estimates the number of segments given knowledge about the level of model failure expected in the sequence. The algorithm's performance is illustrated on synthetic and real image sequences.

Dyadic data refers to a domain with two finite sets of objects in which observations are made for dyads, i.e., pairs with one element from either set. This includes event co-occurrences, histogram data, and single stimulus preference data as special cases. Dyadic data arises naturally in many applications ranging from computational linguistics and information retrieval to preference analysis and computer vision. In this paper, we present a systematic, domain-independent framework for unsupervised learning from dyadic data by statistical mixture models. Our approach covers different models with flat and hierarchical latent class structures and unifies probabilistic modeling and structure discovery. Mixture models provide both, a parsimonious yet flexible parameterization of probability distributions with good generalization performance on sparse data, as well as structural information about data-inherent grouping structure. We propose an annealed version of the standard Expectation Maximization algorithm for model fitting which is empirically evaluated on a variety of data sets from different domains.

this paper: ftp://ftp.cs.colorado.edu/pub/Time-Series/MyPapers/experts.ps.Z,

We present a method for the unsupervised segmentation of data streams originating from different unknown sources which alternate in time. We use an architecture consisting of competing neural networks. Memory is included in order to resolve ambiguities of input-output relations. In order to obtain maximal specialization, the competition is adiabatically increased during training. Our method achieves almost perfect identification and segmentation in the case of switching chaotic dynamics where input manifolds overlap and input-output relations are ambiguous. Only a small dataset is needed for the training proceedure. Applications to time series from complex systems demonstrate the potential relevance of our approach for time series analysis and short-term prediction. 1 Introduction Neural networks provide frameworks for the representation of relations present in data. Especially in the fields of classification and time series prediction, neural networks Corresponding author, email:k...

We introduce a method for validation of results obtained by clustering analysis of data. The method is based on resampling the available data. A figure of merit that measures the stability of clustering solutions against resampling is introduced. Clusters which are stable against resampling give rise to local maxima of this figure of merit. This is presented first for a one-dimensional data set, for which an analytic approximation for the figure of merit is derived and compared with numerical measurements. Next, the applicability of the method is demonstrated for higher dimensional data, including gene microarray expression data.

The Self-Organizing Map (SOM) is an efficient tool for visualization of multidimensional numerical data. In this paper, an overview and categorization of both old and new methods for the visualization of SOM is presented. The purpose is to give an idea of what kind of information can be acquired from different presentations and how the SOM can best be utilized in exploratory data visualization. Most of the presented methods can also be applied in the more general case of first making a vector quantization (e.g. k-means) and then a vector projection (e.g. Sammon's mapping).

Vector quantization is a data compression method where a set of data points is encoded by a reduced set of reference vectors, the codebook. We discuss a vector quantization strategy which jointly optimizes distortion errors and the codebook complexity, thereby, determining the size of the codebook. A maximum entropy estimation of the cost function yields an optimal number of reference vectors, their positions and their assignment probabilities. The dependence of the codebook density on the data density for different complexity functions is investigated in the limit of asymptotic quantization levels. How different complexity measures influence the efficiency of vector quantizers is studied for the task of image compression, i.e., we quantize the wavelet coefficients of gray level images and measure the reconstruction error. Our approach establishes a unifying framework for different quantization methods like K-means clustering and its fuzzy version, entropy constrained vector quantizati...

We present a new approach to clustering, based on the physical properties of an inhomogeneous ferromagnet. No assumption is made regarding the underlying distribution of the data. We assign a Potts spin to each data point and introduce an interaction between neighboring points, whose strength is a decreasing function of the distance between the neighbors. This magnetic system exhibits three phases. At very low temperatures, it is completely ordered; all spins are aligned. At very high temperatures, the system does not exhibit any ordering, and in an intermediate regime, clusters of relatively strongly coupled spins become ordered, whereas different clusters remain uncorrelated. This intermediate phase is identified by a jump in the order parameters. The spin-spin correlation function is used to partition the spins and the corresponding data points into clusters. We demonstrate on three synthetic and three real data sets how the method works. Detailed comparison to the performance of other techniques clearly indicates the relative success of our method. 1

One key difficulty with text classification learning algorithms is that they require many hand-labeled examples to learn accurately. This dissertation demonstrates that supervised learning algorithms that use a small number of labeled examples and many inexpensive unlabeled examples can create high-accuracy text classifiers. By assuming that documents are created by a parametric generative model, Expectation-Maximization (EM) finds local maximum a posteriori models and classifiers from all the data -- labeled and unlabeled. These generative models do not capture all the intricacies of text; however on some domains this technique substantially improves classification accuracy, especially when labeled data are sparse. Two problems arise from this basic approach. First, unlabeled data can hurt performance in domains where the generative modeling assumptions are too strongly violated. In this case the assumptions can be made more representative in two ways: by modeling sub-topic class structure, and by modeling super-topic hierarchical class relationships. By doing so, model probability and classification accuracy come into correspondence, allowing unlabeled data to improve classification performance. The second problem is that even with a representative model, the improvements given by unlabeled data do not sufficiently compensate for a paucity of labeled data. Here, limited labeled data provide EM initializations that lead to low-probability models. Performance can be significantly improved by using active learning to select high-quality initializations, and by using alternatives to EM that avoid low-probability local maxima.