A common applied statistics task involves building regression models to characterize non-linear relationships between variables. It is possible to fit such models by assuming a particular non-linear functional form, such as a sinusoidal, exponential, or polynomial function, to describe one variable's response to the variation in another. Unless this relationship is obvious from the outset, however, it involves possibly extensive model selection procedures to ensure the most appropriate model is retained. Alternatively, a non-parametric approach can be adopted by defining a set of knots across the variable space and use a spline or kernel regression to describe arbitrary non-linear relationships. However, knot layout procedures are somewhat ad hoc and can also involve variable selection. A third alternative is to adopt a Bayesian non-parametric strategy, and directly model the unknown underlying function.

Active search (AS) on graphs focuses on collecting certain labeled nodes (targets) given global knowledge of the network topology and its edge weights under a query budget. However, in most networks, nodes, topology and edge weights are all initially unknown. We introduce selective harvesting, a variant of AS where the next node to be queried must be chosen among the neighbors of the current queried node set; the available training data for deciding which node to query is restricted to the subgraph induced by the queried set (and their node attributes) and their neighbors (without any node or edge attributes). Therefore, selective harvesting is a sequential decision problem, where we must decide which node to query at each step. A classifier trained in this scenario suffers from a tunnel vision effect: without recourse to independent sampling, the urge to query promising nodes forces classifiers to gather increasingly biased training data, which we show significantly hurts the performance of AS methods and standard classifiers. We find that it is possible to collect a much larger set of targets by using multiple classifiers, not by combining their predictions as an ensemble, but switching between classifiers used at each step, as a way to ease the tunnel vision effect. We discover that switching classifiers collects more targets by (a) diversifying the training data and (b) broadening the choices of nodes that can be queried next. This highlights an exploration, exploitation, and diversification trade-off in our problem that goes beyond the exploration and exploitation duality found in classic sequential decision problems. From these observations we propose D3TS, a method based on multi-armed bandits for non-stationary stochastic processes that enforces classifier diversity, matching or exceeding the performance of competing methods on seven real network datasets in our evaluation.

Sejun Park Eunho Y ang † Jinwoo Shin November 4, 2017 Abstract Learning parameters of latent graphical models (GM) is inherently much harder than that of no-latent ones since the latent variables make the corresponding log-likelihood non-concave. Nevertheless, expectation-maximization schemes are popularly used in practice, but they are typically stuck in local optima. In the recent years, the method of moments have provided a refreshing angle for resolving the non-convex issue, but it is applicable to a quite limited class of latent GMs. In this paper, we aim for enhancing its power via enlarging such a class of latent GMs. To this end, we introduce two novel concepts, coined marginalization and conditioning, which can reduce the problem of learning a larger GM to that of a smaller one. More importantly, they lead to a sequential learning framework that repeatedly increases the learning portion of given latent GM, and thus covers a significantly broader and more complicated class of loopy latent GMs which include convolutional and random regular models. 1 Introduction Graphical models (GM) are succinct representation of a joint distribution on a graph where each node corresponds to a random variable and each edge represents the conditional independence between random variables. GM have been successfully applied for various fields including information theory [12, 19], physics [24] and machine learning [18, 11]. Introducing latent variables to GM has been popular approaches for enhancing their representation powers in recent deep models, e.g., convolutional/restricted/deep Boltzmann machines [20, 27]. Furthermore, they are inevitable in certain scenarios when a part of samples is missing, e.g., see [10]. However, learning parameters of latent GMs is significantly harder than that of no-latent ones since the latent variables make the corresponding negative log-likelihood non-convex.

It is a truism that the knowledge of groups of people, particularly experts, outperforms that of individuals [43] and there is increasing call to use the dispersed judgments of the crowd in policy making [42]. There is a large literature spanning multiple disciplines on methods for aggregating beliefs (for reviews see [9, 6, 7]), and previous applications have included political and economic forecasting [3, 27], evaluating nuclear safety [10] and public policy [28], and assessing the quality of chemical probes [31]. However, previous approaches to aggregating beliefs have implicitly assumed'kind' (as opposed to'wicked') environments [16]. In a previous paper, [35] we proposed an algorithm for aggregating beliefs using not only respondent's answers but also their prediction of the answer distribution, and proved that for an infinite number of non-noisy Bayesian respondents, it would always determine the correct answer if sufficient evidence was available in the world. 1 Here, we build on this approach but treat the aggregation problem as one of statistical inference. We propose a model of how people formulate their own judgments and predict the distribution of the judgments of others, and use this model to infer the most probable world state giving rise to the observed data from people. The model can be applied at the level of a single question but also across multiple questions, to infer the domain expertise of respondents. The model is thus broader in scope than other machine learning models for aggregation in that it accepts unique questions, but can also be compared to their performance across multiple questions. We do not assume that the aggregation model has access to correct answers or to historical data about the performance of respondents on similar questions. By using a simple model of how people make such judgments, we are able to increase the accuracy of the group's aggregate answer in domains ranging from estimating art prices to diagnosing skin lesions.

Bayesian network structure learning is often performed in a Bayesian setting, by evaluating candidate structures using their posterior probabilities for a given data set. Score-based algorithms then use those posterior probabilities as an objective function and return the maximum a posteriori network as the learned model. For discrete Bayesian networks, the canonical choice for a posterior score is the Bayesian Dirichlet equivalent uniform (BDeu) marginal likelihood with a uniform (U) graph prior (Heckerman et al., 1995). Its favourable theoretical properties descend from assuming a uniform prior both on the space of the network structures and on the space of the parameters of the network. In this paper, we revisit the limitations of these assumptions; and we introduce an alternative set of assumptions and the resulting score: the Bayesian Dirichlet sparse (BDs) empirical Bayes marginal likelihood with a marginal uniform (MU) graph prior. We evaluate its performance in an extensive simulation study, showing that MU+BDs is more accurate than U+BDeu both in learning the structure of the network and in predicting new observations, while not being computationally more complex to estimate.

Context: One of the black arts of data mining is learning the magic parameters which control the learners. In software analytics, at least for defect prediction, several methods, like grid search and differential evolution (DE), have been proposed to learn these parameters, which has been proved to be able to improve the performance scores of learners. Objective: We want to evaluate which method can find better parameters in terms of performance score and runtime cost. Methods: This paper compares grid search to differential evolution, which is an evolutionary algorithm that makes extensive use of stochastic jumps around the search space. Results: We find that the seemingly complete approach of grid search does no better, and sometimes worse, than the stochastic search. When repeated 20 times to check for conclusion validity, DE was over 210 times faster than grid search to tune Random Forests on 17 testing data sets with F-Measure Conclusions: These results are puzzling: why does a quick partial search be just as effective as a much slower, and much more, extensive search? To answer that question, we turned to the theoretical optimization literature. Bergstra and Bengio conjecture that grid search is not more effective than more randomized searchers if the underlying search space is inherently low dimensional. This is significant since recent results show that defect prediction exhibits very low intrinsic dimensionality-- an observation that explains why a fast method like DE may work as well as a seemingly more thorough grid search. This suggests, as a future research direction, that it might be possible to peek at data sets before doing any optimization in order to match the optimization algorithm to the problem at hand.

High signal to noise ratio (SNR) consistency of model selection criteria in linear regression models has attracted a lot of attention recently. However, most of the existing literature on high SNR consistency deals with model order selection. Further, the limited literature available on the high SNR consistency of subset selection procedures (SSPs) is applicable to linear regression with full rank measurement matrices only. Hence, the performance of SSPs used in underdetermined linear models (a.k.a compressive sensing (CS) algorithms) at high SNR is largely unknown. This paper fills this gap by deriving necessary and sufficient conditions for the high SNR consistency of popular CS algorithms like $l_0$-minimization, basis pursuit de-noising or LASSO, orthogonal matching pursuit and Dantzig selector. Necessary conditions analytically establish the high SNR inconsistency of CS algorithms when used with the tuning parameters discussed in literature. Novel tuning parameters with SNR adaptations are developed using the sufficient conditions and the choice of SNR adaptations are discussed analytically using convergence rate analysis. CS algorithms with the proposed tuning parameters are numerically shown to be high SNR consistent and outperform existing tuning parameters in the moderate to high SNR regime.

All the regression theory developed by statisticians over the last 200 years (related to the general linear model) is useless. Regression can be performed as accurately without statistical models, including the computation of confidence intervals (for estimates, predicted values or regression parameters). The non-statistical approach is also more robust than theory described in all statistics textbooks and taught in all statistical courses. It does not require Map-Reduce when data is really big, nor any matrix inversion, maximum likelihood estimation, or mathematical optimization (Newton algorithm). It is indeed incredibly simple, robust, easy to interpret, and easy to code (no statistical libraries required).

The emergence and development of cancer is a consequence of the accumulation over time of genomic mutations involving a specific set of genes, which provides the cancer clones with a functional selective advantage. In this work, we model the order of accumulation of such mutations during the progression, which eventually leads to the disease, by means of probabilistic graphic models, i.e., Bayesian Networks (BNs). We investigate how to perform the task of learning the structure of such BNs, according to experimental evidence, adopting a global optimization meta-heuristics. In particular, in this work we rely on Genetic Algorithms, and to strongly reduce the execution time of the inference -- which can also involve multiple repetitions to collect statistically significant assessments of the data -- we distribute the calculations using both multi-threading and a multi-node architecture. The results show that our approach is characterized by good accuracy and specificity; we also demonstrate its feasibility, thanks to a 84x reduction of the overall execution time with respect to a traditional sequential implementation.

To obtain uncertainty estimates with real-world Bayesian deep learning models, practical inference approximations are needed. Dropout variational inference (VI) for example has been used for machine vision and medical applications, but VI can severely underestimates model uncertainty. Alpha-divergences are alternative divergences to VI's KL objective, which are able to avoid VI's uncertainty underestimation. But these are hard to use in practice: existing techniques can only use Gaussian approximating distributions, and require existing models to be changed radically, thus are of limited use for practitioners. We propose a re-parametrisation of the alpha-divergence objectives, deriving a simple inference technique which, together with dropout, can be easily implemented with existing models by simply changing the loss of the model. We demonstrate improved uncertainty estimates and accuracy compared to VI in dropout networks. We study our model's epistemic uncertainty far away from the data using adversarial images, showing that these can be distinguished from non-adversarial images by examining our model's uncertainty.