One of the limiting factors of using support vector machines (SVMs) in large scale applications are their super-linear computational requirements in terms of the number of training samples. To address this issue, several approaches that train SVMs on many small chunks of large data sets separately have been proposed in the literature. So far, however, almost all these approaches have only been empirically investigated. In addition, their motivation was always based on computational requirements. In this work, we consider a localized SVM approach based upon a partition of the input space. For this local SVM, we derive a general oracle inequality. Then we apply this oracle inequality to least squares regression using Gaussian kernels and deduce local learning rates that are essentially minimax optimal under some standard smoothness assumptions on the regression function. This gives the first motivation for using local SVMs that is not based on computational requirements but on theoretical predictions on the generalization performance. We further introduce a data-dependent parameter selection method for our local SVM approach and show that this method achieves the same learning rates as before. Finally, we present some larger scale experiments for our localized SVM showing that it achieves essentially the same test performance as a global SVM for a fraction of the computational requirements. In addition, it turns out that the computational requirements for the local SVMs are similar to those of a vanilla random chunk approach, while the achieved test errors are significantly better.

Most of machine learning deals with vector parameters. Ideally we would like to take higher order information into account and make use of matrix or even tensor parameters. However the resulting algorithms are usually inefficient. Here we address on-line learning with matrix parameters. It is often easy to obtain online algorithm with good generalization performance if you eigendecompose the current parameter matrix in each trial (at a cost of $O(n^3)$ per trial). Ideally we want to avoid the decompositions and spend $O(n^2)$ per trial, i.e. linear time in the size of the matrix data. There is a core trade-off between the running time and the generalization performance, here measured by the regret of the on-line algorithm (total gain of the best off-line predictor minus the total gain of the on-line algorithm). We focus on the key matrix problem of rank $k$ Principal Component Analysis in $\mathbb{R}^n$ where $k \ll n$. There are $O(n^3)$ algorithms that achieve the optimum regret but require eigendecompositions. We develop a simple algorithm that needs $O(kn^2)$ per trial whose regret is off by a small factor of $O(n^{1/4})$. The algorithm is based on the Follow the Perturbed Leader paradigm. It replaces full eigendecompositions at each trial by the problem finding $k$ principal components of the current covariance matrix that is perturbed by Gaussian noise.

For many election systems, bribery (and related) attacks have been shown NP-hard using constructions on combinatorially rich structures such as partitions and covers. This paper shows that for voters who follow the most central political-science model of electorates---single-peaked preferences---those hardness protections vanish. By using single-peaked preferences to simplify combinatorial covering challenges, we for the first time show that NP-hard bribery problems---including those for Kemeny and Llull elections---fall to polynomial time for single-peaked electorates. By using single-peaked preferences to simplify combinatorial partition challenges, we for the first time show that NP-hard partition-of-voters problems fall to polynomial time for single-peaked electorates. We show that for single-peaked electorates, the winner problems for Dodgson and Kemeny elections, though Theta-two-complete in the general case, fall to polynomial time. And we completely classify the complexity of weighted coalition manipulation for scoring protocols in single-peaked electorates.

Temporally extended actions have proven useful for reinforcement learning, but their duration also makes them valuable for efficient planning. The options framework provides a concrete way to implement and reason about temporally extended actions. Existing literature has demonstrated the value of planning with options empirically, but there is a lack of theoretical analysis formalizing when planning with options is more efficient than planning with primitive actions. We provide a general analysis of the convergence rate of a popular Approximate Value Iteration (AVI) algorithm called Fitted Value Iteration (FVI) with options. Our analysis reveals that longer duration options and a pessimistic estimate of the value function both lead to faster convergence. Furthermore, options can improve convergence even when they are suboptimal and sparsely distributed throughout the state-space. Next we consider the problem of generating useful options for planning based on a subset of landmark states. This suggests a new algorithm, Landmark-based AVI (LAVI), that represents the value function only at the landmark states. We analyze both FVI and LAVI using the proposed landmark-based options and compare the two algorithms. Our experimental results in three different domains demonstrate the key properties from the analysis. Our theoretical and experimental results demonstrate that options can play an important role in AVI by decreasing approximation error and inducing fast convergence.

Conjunctive regular path queries are an expressive extension of the well-known class of conjunctive queries. Such queries have been extensively studied in the (graph) database community, since they support a controlled form of recursion and enable sophisticated path navigation. Somewhat surprisingly, there has been little work aimed at using such queries in the context of description logic (DL) knowledge bases, particularly for the lightweight DLs that are considered best suited for data-intensive applications. This paper aims to bridge this gap by providing algorithms and tight complexity bounds for answering two-way conjunctive regular path queries over DL knowledge bases formulated in lightweight DLs of the DL-Lite and EL families. Our results demonstrate that in data complexity, the cost of moving to this richer query language is as low as one could wish for: the problem is NL-complete for DL-Lite and P-complete for EL. The combined complexity of query answering increases from NP- to PSpace-complete, but for two-way regular path queries (without conjunction), we show that query answering is tractable even with respect to combined complexity. Our results reveal two-way conjunctive regular path queries as a promising language for querying data enriched by ontologies formulated in DLs of the DL-Lite and EL families or the corresponding OWL 2 QL and EL profiles.

Hidden Markov models have successfully been applied as models of discrete time series in many fields. Often, when applied in practice, the parameters of these models have to be estimated. The currently predominating identification methods, such as maximum-likelihood estimation and especially expectation-maximization, are iterative and prone to have problems with local minima. A non-iterative method employing a spectral subspace-like approach has recently been proposed in the machine learning literature. This paper evaluates the performance of this algorithm, and compares it to the performance of the expectation-maximization algorithm, on a number of numerical examples. We find that the performance is mixed; it successfully identifies some systems with relatively few available observations, but fails completely for some systems even when a large amount of observations is available. An open question is how this discrepancy can be explained. We provide some indications that it could be related to how well-conditioned some system parameters are.

Random forests are a type of ensemble method which makes predictions by combining the results of several independent trees. However, the theory of random forests has long been outpaced by their application. In this paper, we propose a novel random forests algorithm based on cooperative game theory. Banzhaf power index is employed to evaluate the power of each feature by traversing possible feature coalitions. Unlike the previously used information gain rate of information theory, which simply chooses the most informative feature, the Banzhaf power index can be considered as a metric of the importance of each feature on the dependency among a group of features. More importantly, we have proved the consistency of the proposed algorithm, named Banzhaf random forests (BRF). This theoretical analysis takes a step towards narrowing the gap between the theory and practice of random forests for classification problems. Experiments on several UCI benchmark data sets show that BRF is competitive with state-of-the-art classifiers and dramatically outperforms previous consistent random forests. Particularly, it is much more efficient than previous consistent random forests.

Variable selection plays an important role in the high-dimensional data analysis. However the high-dimensional data often induces the strongly correlated variables problem. In this paper, we propose Elastic Net procedure for partially linear models and prove the group effect of its estimate. By a simulation study, we show that the strongly correlated variables problem can be better handled by the Elastic Net procedure than Lasso, ALasso and Ridge. Based on an empirical analysis, we can get that the Elastic Net procedure is particularly useful when the number of predictors $p$ is much bigger than the sample size $n$.

In this study we present a kernel based convolution model to characterize neural responses to natural sounds by decoding their time-varying acoustic features. The model allows to decode natural sounds from high-dimensional neural recordings, such as magnetoencephalography (MEG), that track timing and location of human cortical signalling noninvasively across multiple channels. We used the MEG responses recorded from subjects listening to acoustically different environmental sounds. By decoding the stimulus frequencies from the responses, our model was able to accurately distinguish between two different sounds that it had never encountered before with 70% accuracy. Convolution models typically decode frequencies that appear at a certain time point in the sound signal by using neural responses from that time point until a certain fixed duration of the response. Using our model, we evaluated several fixed durations (time-lags) of the neural responses and observed auditory MEG responses to be most sensitive to spectral content of the sounds at time-lags of 250 ms to 500 ms. The proposed model should be useful for determining what aspects of natural sounds are represented by high-dimensional neural responses and may reveal novel properties of neural signals.

With the wealth of high-throughput sequencing data generated by recent large-scale consortia, predictive gene expression modelling has become an important tool for integrative analysis of transcriptomic and epigenetic data. However, sequencing data-sets are characteristically large, and previously modelling frameworks are typically inefficient and unable to leverage multi-core or distributed processing architectures. In this study, we detail an efficient and parallelised MapReduce implementation of gene expression modelling. We leverage the computational efficiency of this framework to provide an integrative analysis of over fifty histone modification data-sets across a variety of cancerous and non-cancerous cell-lines. Our results demonstrate that the genome-wide relationships between histone modifications and mRNA transcription are lineage, tissue and karyotype-invariant, and that models trained on matched epigenetic/transcriptomic data from non-cancerous cell-lines are able to predict cancerous expression with equivalent genome-wide fidelity.