We show that the multi-class support vector machine (MSVM) proposed by Lee et al. (2004) can be viewed as a MAP estimation procedure under an appropriate probabilistic interpretation of the classifier. We also show that this interpretation can be extended to a hierarchical Bayesian architecture and to a fully-Bayesian inference procedure for multiclass classification based on data augmentation. We present empirical results that show that the advantages of the Bayesian formalism are obtained without a loss in classification accuracy.

We show how to reduce the process of predicting general order statistics (and the median in particular) to solving classification. The accompanying theoretical statement shows that the regret of the classifier bounds the regret of the quantile regression under a quantile loss. We also test this reduction empirically against existing quantile regression methods on large real-world datasets and discover that it provides state-of-the-art performance.

We provide faster algorithms for the problem of Gaussian summation, which occurs in many machine learning methods. We develop two new extensions - an O(Dp) Taylor expansion for the Gaussian kernel with rigorous error bounds and a new error control scheme integrating any arbitrary approximation method - within the best discretealgorithmic framework using adaptive hierarchical data structures. We rigorously evaluate these techniques empirically in the context of optimal bandwidth selection in kernel density estimation, revealing the strengths and weaknesses of current state-of-the-art approaches for the first time. Our results demonstrate that the new error control scheme yields improved performance, whereas the series expansion approach is only effective in low dimensions (five or less).

Nonparametric Bayesian approaches to clustering, information retrieval, language modeling and object recognition have recently shown great promise as a new paradigm for unsupervised data analysis. Most contributions have focused on the Dirichlet process mixture models or extensions thereof for which efficient Gibbs samplers exist. In this paper we explore Gibbs samplers for infinite complexity mixture models in the stick breaking representation. The advantage of this representation is improved modeling flexibility. For instance, one can design the prior distribution over cluster sizes or couple multiple infinite mixture models (e.g. over time) at the level of their parameters (i.e. the dependent Dirichlet process model). However, Gibbs samplers for infinite mixture models (as recently introduced in the statistics literature) seem to mix poorly over cluster labels. Among others issues, this can have the adverse effect that labels for the same cluster in coupled mixture models are mixed up. We introduce additional moves in these samplers to improve mixing over cluster labels and to bring clusters into correspondence. An application to modeling of storm trajectories is used to illustrate these ideas.

Many tasks require finding groups of elements in a matrix of numbers, symbols or class likelihoods. One approach is to use efficient bi- or tri-linear factorization techniques including PCA, ICA, sparse matrix factorization and plaid analysis. These techniques are not appropriate when addition and multiplication of matrix elements are not sensibly defined. More directly, methods like bi-clustering can be used to classify matrix elements, but these methods make the overly-restrictive assumption that the class of each element is a function of a row class and a column class. We introduce a general computational problem, `matrix tile analysis' (MTA), which consists of decomposing a matrix into a set of non-overlapping tiles, each of which is defined by a subset of usually nonadjacent rows and columns. MTA does not require an algebra for combining tiles, but must search over discrete combinations of tile assignments. Exact MTA is a computationally intractable integer programming problem, but we describe an approximate iterative technique and a computationally efficient sum-product relaxation of the integer program. We compare the effectiveness of these methods to PCA and plaid on hundreds of randomly generated tasks. Using double-gene-knockout data, we show that MTA finds groups of interacting yeast genes that have biologically-related functions.

It is said that genes that cluster with similar expression-- that is, are co-expressed--serve similar functional roles in a process (see, for example, Eisen et al. 1998). Bioin-formaticians have more recently had access to sets of time-series measurements of genes' expression over the duration of an experiment, and have desired therefore to learn not just co-expression, but causal relationships that may help elucidate co-regulation as well. Two problematic issues hamper practical methods for clustering gene expression time course data: first, if deriving a model-based clustering metric, it is often unclear what the appropriate model complexity should be; second, the current clustering algorithms available cannot handle, and therefore disregard, the temporal information. This usually occurs when constructing a metric for the distance between any two such genes. The common practice for an experiment having T measurements of a gene's expression over time is to consider the expression as positioned in a T -dimensional space, and to perform (at worse spherical metric) clustering in that space. The result is that the clustering algorithm is invariant to arbitrary permutations of the time points, which is highly undesirable since we would like to take into account the correlations between all the genes' expression at nearby or adjacent time points.

Discriminative linear models are a popular tool in machine learning. These can be generally divided into two types: The first is linear classifiers, such as support vector machines, which are well studied and provide state-of-the-art results. One shortcoming of these models is that their output (known as the 'margin') is not calibrated, and cannot be translated naturally into a distribution over the labels. Thus, it is difficult to incorporate such models as components of larger systems, unlike probabilistic based approaches. The second type of approach constructs class conditional distributions using a nonlinearity (e.g. log-linear models), but is occasionally worse in terms of classification error. We propose a supervised learning method which combines the best of both approaches. Specifically, our method provides a distribution over the labels, which is a linear function of the model parameters. As a consequence, differences between probabilities are linear functions, a property which most probabilistic models (e.g. log-linear) do not have. Our model assumes that classes correspond to linear subspaces (rather than to half spaces). Using a relaxed projection operator, we construct a measure which evaluates the degree to which a given vector 'belongs' to a subspace, resulting in a distribution over labels. Interestingly, this view is closely related to similar concepts in quantum detection theory. The resulting models can be trained either to maximize the margin or to optimize average likelihood measures. The corresponding optimization problems are semidefinite programs which can be solved efficiently. We illustrate the performance of our algorithm on real world datasets, and show that it outperforms 2nd order kernel methods.

X in R^D has mean zero and finite second moments. We show that there is a precise sense in which almost all linear projections of X into R^d (for d < D) look like a scale-mixture of spherical Gaussians -- specifically, a mixture of distributions N(0, sigma^2 I_d) where the weight of the particular sigma component is P (| X |^2 = sigma^2 D). The extent of this effect depends upon the ratio of d to D, and upon a particular coefficient of eccentricity of X's distribution. We explore this result in a variety of experiments.

To date, testing interactions in high dimensions has been a challenging task. Existing methods often have issues with sensitivity to modeling assumptions and heavily asymptotic nominal p-values. To help alleviate these issues, we propose a permutation-based method for testing marginal interactions with a binary response. Our method searches for pairwise correlations which differ between classes. In this manuscript, we compare our method on real and simulated data to the standard approach of running many pairwise logistic models. On simulated data our method finds more significant interactions at a lower false discovery rate (especially in the presence of main effects). On real genomic data, although there is no gold standard, our method finds apparent signal and tells a believable story, while logistic regression does not. We also give asymptotic consistency results under not too restrictive assumptions.

We propose a semiparametric approach, named nonparanormal skeptic, for estimating high dimensional undirected graphical models. In terms of modeling, we consider the nonparanormal family proposed by Liu et al (2009). In terms of estimation, we exploit nonparametric rank-based correlation coefficient estimators including the Spearman's rho and Kendall's tau. In high dimensional settings, we prove that the nonparanormal skeptic achieves the optimal parametric rate of convergence in both graph and parameter estimation. This result suggests that the nonparanormal graphical models are a safe replacement of the Gaussian graphical models, even when the data are Gaussian.