Bayesian optimization (BO) is a flexible and powerful framework that is suitable for computationally expensive simulation-based applications and guarantees statistical convergence to the global optimum. While remaining as one of the most popular optimization methods, its capability is hindered by the size of data, the dimensionality of the considered problem, and the nature of sequential optimization. These scalability issues are intertwined with each other and must be tackled simultaneously. In this work, we propose the Scalable$^3$-BO framework, which employs sparse GP as the underlying surrogate model to scope with Big Data and is equipped with a random embedding to efficiently optimize high-dimensional problems with low effective dimensionality. The Scalable$^3$-BO framework is further leveraged with asynchronous parallelization feature, which fully exploits the computational resource on HPC within a computational budget. As a result, the proposed Scalable$^3$-BO framework is scalable in three independent perspectives: with respect to data size, dimensionality, and computational resource on HPC. The goal of this work is to push the frontiers of BO beyond its well-known scalability issues and minimize the wall-clock waiting time for optimizing high-dimensional computationally expensive applications. We demonstrate the capability of Scalable$^3$-BO with 1 million data points, 10,000-dimensional problems, with 20 concurrent workers in an HPC environment.

The generalization error of a classifier is related to the complexity of the set of functions among which the classifier is chosen. Roughly speaking, the more complex the family, the greater the potential disparity between the training error and the population error of the classifier. This principle is embodied in layman's terms by Occam's razor principle, which suggests favoring low-complexity hypotheses over complex ones. We study a family of low-complexity classifiers consisting of thresholding the one-dimensional feature obtained by projecting the data on a random line after embedding it into a higher dimensional space parametrized by monomials of order up to k. More specifically, the extended data is projected n-times and the best classifier among those n (based on its performance on training data) is chosen. We obtain a bound on the generalization error of these low-complexity classifiers. The bound is less than that of any classifier with a non-trivial VC dimension, and thus less than that of a linear classifier. We also show that, given full knowledge of the class conditional densities, the error of the classifiers would converge to the optimal (Bayes) error as k and n go to infinity; if only a training dataset is given, we show that the classifiers will perfectly classify all the training points as k and n go to infinity.

Many scientific problems require identifying a small set of covariates that are associated with a target response and estimating their effects. Often, these effects are nonlinear and include interactions, so linear and additive methods can lead to poor estimation and variable selection. The Bayesian framework makes it straightforward to simultaneously express sparsity, nonlinearity, and interactions in a hierarchical model. But, as for the few other methods that handle this trifecta, inference is computationally intractable - with runtime at least quadratic in the number of covariates, and often worse. In the present work, we solve this computational bottleneck. We first show that suitable Bayesian models can be represented as Gaussian processes (GPs). We then demonstrate how a kernel trick can reduce computation with these GPs to O(# covariates) time for both variable selection and estimation. Our resulting fit corresponds to a sparse orthogonal decomposition of the regression function in a Hilbert space (i.e., a functional ANOVA decomposition), where interaction effects represent all variation that cannot be explained by lower-order effects. On a variety of synthetic and real datasets, our approach outperforms existing methods used for large, high-dimensional datasets while remaining competitive (or being orders of magnitude faster) in runtime.

The'planning as inference' paradigm extends Bayesian inference to future observations. The agent in the environment is modelled as a Bayesian generative model, but the belief about the distribution of agent's actions is updated based on future goals rather than on past facts. This allows to use common modelling and inference tools, notably probabilistic programming, to represent computer agents and explore their behavior. Representing agents as general programs provides flexibility compared to restricted approaches, such as Markov decision processes and their variants and extensions, and allows to model a broad range of complex behaviors in a unified and natural way. Planning as inference models agent preferences through conditioning agents on preferred future behaviors. Often, the conditioning is achieved through the Boltzmann distribution: the probability of a realization of agent's behavior is proportional to the exponent of the agent's reward. The motivation of using the Boltzmann distribution is not clear though. A'rational' agent should behave in a way that maximizes the agent's expected utility, shouldn't it? One argument is that the Boltzmann distribution models human errors and irrationality.

Given a combinatorial search problem, it may be highly useful to enumerate its (all) solutions besides just finding one solution, or showing that none exists. The same can be stated about optimal solutions if an objective function is provided. This work goes beyond the bare enumeration of optimal solutions and addresses the computational task of solution enumeration by optimality (SEO). This task is studied in the context of Answer Set Programming (ASP) where (optimal) solutions of a problem are captured with the answer sets of a logic program encoding the problem. Existing answer-set solvers already support the enumeration of all (optimal) answer sets. However, in this work, we generalize the enumeration of optimal answer sets beyond strictly optimal ones, giving rise to the idea of answer set enumeration in the order of optimality (ASEO). This approach is applicable up to the best k answer sets or in an unlimited setting, which amounts to a process of sorting answer sets based on the objective function. As the main contribution of this work, we present the first general algorithms for the aforementioned tasks of answer set enumeration. Moreover, we illustrate the potential use cases of ASEO. First, we study how efficiently access to the next-best solutions can be achieved in a number of optimization problems that have been formalized and solved in ASP. Second, we show that ASEO provides us with an effective sampling technique for Bayesian networks.

We study the problem of estimating potential revenue or demand at business facilities and understanding its generating mechanism. This problem arises in different fields such as operation research or urban science, and more generally, it is crucial for businesses' planning and decision making. We develop a Bayesian spatial interaction model, henceforth BSIM, which provides probabilistic predictions about revenues generated by a particular business location provided their features and the potential customers' characteristics in a given region. BSIM explicitly accounts for the competition among the competitive facilities through a probability value determined by evaluating a store-specific Gaussian distribution at a given customer location. We propose a scalable variational inference framework that, while being significantly faster than competing Markov Chain Monte Carlo inference schemes, exhibits comparable performances in terms of parameters identification and uncertainty quantification. We demonstrate the benefits of BSIM in various synthetic settings characterised by an increasing number of stores and customers. Finally, we construct a real-world, large spatial dataset for pub activities in London, UK, which includes over 1,500 pubs and 150,000 customer regions. We demonstrate how BSIM outperforms competing approaches on this large dataset in terms of prediction performances while providing results that are both interpretable and consistent with related indicators observed for the London region.

Shape modelling (with methods that output shapes) is a new and important task in Bayesian nonparametrics and bioinformatics. In this work, we focus on Bayesian nonparametric methods for capturing shapes by partitioning a space using curves. In related work, the classical Mondrian process is used to partition spaces recursively with axis-aligned cuts, and is widely applied in multi-dimensional and relational data. The Mondrian process outputs hyper-rectangles. Recently, the random tessellation process was introduced as a generalization of the Mondrian process, partitioning a domain with non-axis aligned cuts in an arbitrary dimensional space, and outputting polytopes. Motivated by these processes, in this work, we propose a novel parallelized Bayesian nonparametric approach to partition a domain with curves, enabling complex data-shapes to be acquired. We apply our method to HIV-1-infected human macrophage image dataset, and also simulated datasets sets to illustrate our approach. We compare to support vector machines, random forests and state-of-the-art computer vision methods such as simple linear iterative clustering super pixel image segmentation. We develop an R package that is available at \url{https://github.com/ShufeiGe/Shape-Modeling-with-Spline-Partitions}.

In this paper, we present a Stochastic Deep Neural Network-based Model Reference Adaptive Control. Building on our work "Deep Model Reference Adaptive Control", we extend the controller capability by using Bayesian deep neural networks (DNN) to represent uncertainties and model non-linearities. Stochastic Deep Model Reference Adaptive Control uses a Lyapunov-based method to adapt the output-layer weights of the DNN model in real-time, while a data-driven supervised learning algorithm is used to update the inner-layers parameters. This asynchronous network update ensures boundedness and guaranteed tracking performance with a learning-based real-time feedback controller. A Bayesian approach to DNN learning helped avoid over-fitting the data and provide confidence intervals over the predictions. The controller's stochastic nature also ensured "Induced Persistency of excitation," leading to convergence of the overall system signal.

Bayesian networks are a widely-used class of probabilistic graphical models capable of representing symmetric conditional independence between variables of interest using the topology of the underlying graph. They can be seen as a special case of the much more general class of models called staged trees, which can represent any type of non-symmetric conditional independence. Here we formalize the relationship between these two models and introduce a minimal Bayesian network representation of the staged tree, which can be used to read conditional independences in an intuitive way. Furthermore, we define a new labeled graph, termed asymmetry-labeled directed acyclic graph, whose edges are labeled to denote the type of dependence existing between any two random variables. Various datasets are used to illustrate the methodology, highlighting the need to construct models which more flexibly encode and represent non-symmetric structures.

Exponential families are widely used in machine learning; they include many distributions in continuous and discrete domains (e.g., Gaussian, Dirichlet, Poisson, and categorical distributions via the softmax transformation). Distributions in each of these families have fixed support. In contrast, for finite domains, there has been recent works on sparse alternatives to softmax (e.g. sparsemax, $\alpha$-entmax, and fusedmax) and corresponding losses, which have varying support. This paper expands that line of work in several directions: first, it extends $\Omega$-regularized prediction maps and Fenchel-Young losses to arbitrary domains (possibly countably infinite or continuous). For linearly parametrized families, we show that minimization of Fenchel-Young losses is equivalent to moment matching of the statistics, generalizing a fundamental property of exponential families. When $\Omega$ is a Tsallis negentropy with parameter $\alpha$, we obtain "deformed exponential families," which include $\alpha$-entmax and sparsemax ($\alpha$ = 2) as particular cases. For quadratic energy functions in continuous domains, the resulting densities are $\beta$-Gaussians, an instance of elliptical distributions that contain as particular cases the Gaussian, biweight, triweight and Epanechnikov densities, and for which we derive closed-form expressions for the variance, Tsallis entropy, and Fenchel-Young loss. When $\Omega$ is a total variation or Sobolev regularizer, we obtain a continuous version of the fusedmax. Finally, we introduce continuous-domain attention mechanisms, deriving efficient gradient backpropagation algorithms for $\alpha \in \{1, 4/3, 3/2, 2\}$. Using them, we demonstrate our sparse continuous distributions for attention-based audio classification and visual question answering, showing that they allow attending to time intervals and compact regions.