Abstract: As an extension of partial least squares (PLS), kernel partial least squares (KPLS) is an very important methods to find nonlinear patterns from data. However, the application of KPLS to large-scale problems remains a big challenge, due to storage and computation issues in the number of examples. To address this limitation, we utilize randomness to design scalable new variants of the kernel matrix to solve KPLS. Specifically, we consider the spectral properties of low-rank kernel matrices constructed as sums of random feature dot-products and present a new method called randomized kernel partial least squares (RKPLS) to approximate KPLS. RKPLS can alleviate the computation requirements of approximate KPLS with linear space and computation in the sample size. Theoretical analysis and experimental results show that the solution of our algorithm converges to exact kernel matrix in expectation. PubDate: 2020-03-27

Abstract: The architecture described in this paper encodes a theory of intentions based on the key principles of non-procrastination, persistence, and automatically limiting reasoning to relevant knowledge and observations. The architecture reasons with transition diagrams of any given domain at two different resolutions, with the fine-resolution description defined as a refinement of, and hence tightly-coupled to, a coarse-resolution description. For any given goal, nonmonotonic logical reasoning with the coarse-resolution description computes an activity, i.e., a plan, comprising a sequence of abstract actions to be executed to achieve the goal. Each abstract action is implemented as a sequence of concrete actions by automatically zooming to and reasoning with the part of the fine-resolution transition diagram relevant to the current coarse-resolution transition and the goal. Each concrete action in this sequence is executed using probabilistic models of the uncertainty in sensing and actuation, and the corresponding fine-resolution outcomes are used to infer coarse-resolution observations that are added to the coarse-resolution history. The architecture’s capabilities are evaluated in the context of a simulated robot assisting humans in an office domain, on a physical robot (Baxter) manipulating tabletop objects, and on a wheeled robot (Turtlebot) moving objects to particular places or people. The experimental results indicate improvements in reliability and computational efficiency compared with an architecture that does not include the theory of intentions, and an architecture that does not include zooming for fine-resolution reasoning. PubDate: 2020-03-24

Abstract: Perceivable objects are customarily termed as concepts and their representations (localist-distributed, modality-specific, or experience-dependent) are ingrained in our lives. Despite a considerable amount of computational modeling research focuses on concrete concepts, no comprehensible method for abstract concepts has hitherto been considered. concepts can be viewed as a blend of concrete concepts. We use this view in our proposed model, Regulated Activation Network (RAN), by learning representations of non-convex abstract concepts without supervision via a hybrid model that has an evolving topology. First, we describe the RAN’s modeling process through a Toy-data problem yielding a performance of 98.5%(ca.) in a classification task. Second, RAN’s model is used to infer psychological and physiological biomarkers from students’ active and inactive states using sleep-detection data. The RAN’s capability of performing classification is shown using five UCI benchmarks, with the best outcome of 96.5% (ca.) for Human Activity recognition data. We empirically demonstrate the proposed model using standard performance measures for classification and establish RAN’s competency with five classifiers. We show that the RAN adeptly performs classification with a small amount of data and simulate cognitive functions like activation propagation and learning. PubDate: 2020-03-21

Abstract: This paper develops and examines an innovative methodology for training an artificial neural network to identify and tag target visual objects in a given database. While the field of Artificial Intelligence in general, and computer vision in particular, has greatly advanced in recent years, fast and efficient methods for tagging (i.e., labeling) visual targets are still lacking. Tagging data is important to train, as it allow to train supervised learning models. However, this is a tiresome task that often creates bottlenecks in academic and industrial research projects. In order to develop an algorithm that improves data tagging processes, this study utilizes the advantages of human cognition and machine learning by combining Brain Computer Interface, Human-In-The-Loop, and Deep Learning. Combining these three fields into one algorithm could enable the rapid annotation of large visual databases that have no prior references and cannot be described as a mathematical optimization function. Human-In-The-Loop is an increasingly researched area that refers to the integration of human feedback in computation processes. At present, computer-based deep learning can only be incorporated in the process of identifying and tagging target objects of interest if a predefined database exists – one that has already been defined by a human user. To reduce the scope of this timely and costly process, our algorithm uses machine learning techniques (i.e., active learning) to minimize the number of target objects a human user needs to identify before the computer can successfully carry out the task independently. In our method, users are connected to electroencephalograms electrodes and shown images using rapid serial visual presentation – a fast method for presenting users with images. Some images are target objects, while others are not. Based on users’ brainwave activity when target objects are shown, the computer learns to identify and tag target objects – already in the learning stage (unlike naïve uniform sampling methods that first require human input, and only then begin the learning stage). As such, our work is proof of concept for the effectiveness of involving humans in the computer’s learning stage, i.e., human-in-the-loop as opposed to the traditional method of humans first tagging the data and the machines then learning and creating a model. PubDate: 2020-03-16

Abstract: In this paper, we analyze the fundamental conditions for low-rank tensor completion given the separation or tensor-train (TT) rank, i.e., ranks of TT unfoldings. We exploit the algebraic structure of the TT decomposition to obtain the deterministic necessary and sufficient conditions on the locations of the samples to ensure finite completability. Specifically, we propose an algebraic geometric analysis on the TT manifold that can incorporate the whole rank vector simultaneously in contrast to the existing approach based on the Grassmannian manifold that can only incorporate one rank component. Our proposed technique characterizes the algebraic independence of a set of polynomials defined based on the sampling pattern and the TT decomposition, which is instrumental to obtaining the deterministic condition on the sampling pattern for finite completability. In addition, based on the proposed analysis, assuming that the entries of the tensor are sampled independently with probability p, we derive a lower bound on the sampling probability p, or equivalently, the number of sampled entries that ensures finite completability with high probability. Moreover, we also provide the deterministic and probabilistic conditions for unique completability. PubDate: 2020-03-16

Abstract: Abstract In this paper we investigate the problem of scheduling a set of jobs on a single-machine. The jobs are classified in families and setup times are required between the processing of two jobs of different families. Each job requires a certain amount of a common resource that is supplied through upstream processes. The total resource consumed must not exceed the resource supply up. Therefore, jobs may have to wait and the machine has to be idle due to an insufficient availability of the resource. To minimize the total tardiness, simple and effective iterated greedy (IG) heuristics are proposed. Different neighborhood operators are used in the local search phase. To choose the right neighborhood operators, we propose an adaptive selecting strategy. The heuristics are tested over an extensive computational experience on benchmark of instances from the literature and instances randomly generated in this work. Experimental results and statistical tests show that the proposed heuristics are able to obtain high-quality solutions within reasonable computational effort, and they outperform the state-of-the-art heuristic. PubDate: 2020-03-01

Abstract: Abstract In the past few decades, design theory has grown to encompass a wide variety of research directions. It comes as no surprise that applications in coding theory and communications continue to arise, and also that designs have found applications in new areas. Computer science has provided a new source of applications of designs, and simultaneously a field of new and challenging problems in design theory. In this paper, we revisit a construction for orthogonal designs using the multiplication tables of Cayley-Dixon algebras of dimension 2n. The desired orthogonal designs can be described by a system of equations with the aid of a Gröbner basis computation. For orders greater than 16 the combinatorial explosion of the problem gives rise to equations that are unfeasible to be handled by traditional search algorithms. However, the structural properties of the designs make this problem possible to be tackled in terms of rewriting techniques, by equational unification. We establish connections between central concepts of design theory and equational unification where equivalence operations of designs point to the computation of a minimal complete set of unifiers. These connections make viable the computation of some types of orthogonal designs that have not been found before with the aforementioned algebraic modeling. PubDate: 2020-03-01

Abstract: Abstract Multi-objective optimization aims at finding trade-off solutions to conflicting objectives. These constitute the Pareto optimal set. In the context of expensive-to-evaluate functions, it is impossible and often non-informative to look for the entire set. As an end-user would typically prefer a certain part of the objective space, we modify the Bayesian multi-objective optimization algorithm which uses Gaussian Processes and works by maximizing the Expected Hypervolume Improvement, to focus the search in the preferred region. The cumulated effects of the Gaussian Processes and the targeting strategy lead to a particularly efficient convergence to the desired part of the Pareto set. To take advantage of parallel computing, a multi-point extension of the targeting criterion is proposed and analyzed. PubDate: 2020-03-01

Abstract: Abstract Revenue Management uses data-driven modelling and optimization methods to decide what to sell, when to sell, to whom to sell, and for which price, in order to increase revenue and profit. Hotel Revenue Management is a very complex context characterized by nonlinearities, many parameters and constraints, and stochasticity, in particular in the demand by customers. It suffers from the curse of dimensionality (Bellman 2015): when the number of variables increases (number of rooms, number possible prices and capacities, number of reservation rules and constraints) exact solutions by dynamic programming or by alternative global optimization techniques cannot be used and one has to resort to intelligent heuristics, i.e., methods which can improve current solutions but without formal guarantees of optimality. Effective heuristics can incorporate “learning” (“reactive” schemes) that update strategies based on the past history of the process, the past reservations received up to a certain time and the previous steps in the iterative optimization process. Different approaches can be classified according to the specific model considered (stochastic demand and hotel rules), the control mechanism (the pricing policy) and the optimization technique used to determine improving or optimal solutions. In some cases, model definitions, control mechanism and solution techniques are strongly interrelated: this is the case of dynamic programming, which demands suitably simplified problem formulations. We design a flexible discrete-event simulator for the hotel reservation process and experiment different approaches though measurements of the expected effect on profit (obtained by carefully separating a “training” phase from the final “validation” phase obtained from different simulations). The experimental results show the effectiveness of intelligent heuristics with respect to exact optimization methods like dynamic programming, in particular for more constrained situations (cases when demand tends to saturate hotel room availability), when the simplifying assumptions needed to make the problem analytically treatable do not hold. PubDate: 2020-03-01

Abstract: Abstract The following two strongly NP-hard problems are considered. In the first problem, we need to find in the given finite set of points in Euclidean space the subset of largest size. The sum of squared distances between the elements of this subset and its unknown centroid (geometrical center) must not exceed a given value. This value is defined as percentage of the sum of squared distances between the elements of the input set and its centroid. In the second problem, the input is a sequence (not a set) and we have some additional constraints on the indices of the elements of the chosen subsequence. The restriction on the sum of squared distances is the same as in the first problem. Both problems can be treated as data editing problems aimed to find similar elements and removal of extraneous (dissimilar) elements. We propose exact algorithms for the cases of both problems in which the input points have integer-valued coordinates. If the space dimension is bounded by some constant, our algorithms run in a pseudopolynomial time. Some results of numerical experiments illustrating the performance of the algorithms are presented. PubDate: 2020-03-01

Abstract: Abstract A support vector machine (SVM) classifier corresponds in its most basic form to a quadratic programming problem. Various linear variations of support vector classification have been investigated such as minimizing the L1-norm of the weight-vector instead of the L2-norm. In this paper we introduce a classifier where we minimize the boundary (lower envelope) of the epigraph that is generated over a set of functions, which can be interpreted as a measure of distance or slack from the origin. The resulting classifier appears to provide a generalization performance similar to SVMs while displaying a more advantageous computational complexity. The discussed formulation can also be extended to allow for cases with imbalanced data. PubDate: 2020-03-01

Abstract: Abstract This article describes the application of soft computing methods for solving the problem of locating garbage accumulation points in urban scenarios. This is a relevant problem in modern smart cities, in order to reduce negative environmental and social impacts in the waste management process, and also to optimize the available budget from the city administration to install waste bins. A specific problem model is presented, which accounts for reducing the investment costs, enhance the number of citizens served by the installed bins, and the accessibility to the system. A family of single- and multi-objective heuristics based on the PageRank method and two mutiobjective evolutionary algorithms are proposed. Experimental evaluation performed on real scenarios on the cities of Montevideo (Uruguay) and Bahía Blanca (Argentina) demonstrates the effectiveness of the proposed approaches. The methods allow computing plannings with different trade-off between the problem objectives. The computed results improve over the current planning in Montevideo and provide a reasonable budget cost and quality of service for Bahía Blanca. PubDate: 2020-03-01

Abstract: Abstract We consider the geometric version of the well-known Generalized Traveling Salesman Problem introduced in 2015 by Bhattacharya et al. that is called the Euclidean Generalized Traveling Salesman Problem in Grid Clusters (EGTSP-GC). They proved the intractability of the problem and proposed first polynomial time algorithms with fixed approximation factors. The extension of these results in the field of constructing the polynomial time approximation schemes (PTAS) and the description of non-trivial polynomial time solvable subclasses for the EGTSP-GC appear to be relevant in the light of the classic C. Papadimitriou result on the intractability of the Euclidean TSP and recent inapproximability results for the Traveling Salesman Problem with Neighborhoods (TSPN) in the case of discrete neighborhoods. In this paper, we propose Efficient Polynomial Time Approximation Schemes (EPTAS) for two special cases of the EGTSP-GC, when the number of clusters \(k=O(\log n)\) and \(k=n-O(\log n)\). Also, we show that any time, when one of the grid dimensions (height or width) is fixed, the EGTSP-GC can be solved to optimality in polynomial time. As a consequence, we specify a novel non-trivial polynomially solvable subclass of the Euclidean TSP in the plane. PubDate: 2020-03-01

Abstract: Abstract A detailed review of the dynamic search trajectory methods for global optimization is given. In addition, a family of dynamic search trajectories methods that are created using numerical methods for solving autonomous ordinary differential equations is presented. Furthermore, a strategy for developing globally convergent methods that is applicable to the proposed family of methods is given and the corresponding theorem is proved. Finally, theoretical results for obtaining nonmonotone convergent methods that exploit the accumulated information with regard to the most recent values of the objective function are given. PubDate: 2020-03-01

Abstract: Abstract This paper introduces a parallel-batching scheduling problem with deterioration and learning effects on parallel machines, where the actual processing time of a job is subject to the phenomena of deterioration and learning. All jobs are first divided into different parallel batches, and the processing time of the batches is equal to the largest processing time of their belonged jobs. Then, the generated batches are assigned to parallel machines to be processed. Motivated by the characteristics of machine maintenance activities in a semiconductor manufacturing process, we take the machine preventive maintenance into account, i.e., the machine should be maintained after a fixed number of batches have been completed. In order to solve the problem, we analyze several structural properties with respect to the batch formation and sequencing. Based on these properties, a hybrid BRKGA-DE algorithm combining biased random-key genetic algorithm (BRKGA) and Differential Evolution (DE) is proposed to solve the parallel-batching scheduling problem. A series of computational experiments is conducted to demonstrate the effectiveness and efficiency of the proposed algorithm. PubDate: 2020-03-01

Abstract: Abstract Many real applications require the representation of complex entities and their relations. Frequently, networks are the chosen data structures, due to their ability to highlight topological and qualitative characteristics. In this work, we are interested in supervised classification models for data in the form of networks. Given two or more classes whose members are networks, we build mathematical models to classify them, based on various graph distances. Due to the complexity of the models, made of tens of thousands of nodes and edges, we focus on model simplification solutions to reduce execution times, still maintaining high accuracy. Experimental results on three datasets of biological interest show the achieved performance improvements. PubDate: 2020-03-01

Abstract: Nowadays, energy represents the most important resource; however, we need to face several energy-related rising issues, one main concern is how energy is consumed. In particular, how we can stimulate consumers on a specific behaviour. In this work, we present a model facing energy allocation and payment. Thus, we start with the explanation of the first step of our work concerning a mechanism design approach for energy allocation among consumers. More in details, we go deep into the formal description of the energy model and users’ consumption profiles. We aim to select the optimal consumption profile for every user avoiding consumption peaks when the total required energy could exceed the energy production. The mechanism will be able to drive users in shifting energy consumptions in different hours of the day. The next step concerns a payment estimation problem which involves a community of users and an energy distributor (or producer). Our aim is to compute payments for every user in the community according to the single user’s consumption, the community’s consumption and the available energy. By computing community-dependent energy bills, our model stimulates a users’ virtuous behaviour, so that everyone approaches the production threshold as close as possible. Our payment function distributes incentives if the consumption is lower than the produced energy and penalties when the consumption exceeds the resources threshold, satisfying efficiency and fairness properties both from the community (efficiency as an economic equilibrium among sellers and buyers) and the single user (fairness as an economic measure of energy good-behaving) points of view. PubDate: 2020-02-18

Abstract: Moser and Tardos (J. ACM (JACM) 57(2), 11 2010) gave an algorithmic proof of the lopsided Lovász local lemma (LLL) in the variable framework, where each of the undesirable events is assumed to depend on a subset of a collection of independent random variables. For the proof, they define a notion of a lopsided dependency between the events suitable for this framework. In this work, we strengthen this notion, defining a novel directed notion of dependency and prove the LLL for the corresponding graph. We show that this graph can be strictly sparser (thus the sufficient condition for the LLL weaker) compared with graphs that correspond to other extant lopsided versions of dependency. Thus, in a sense, we address the problem “find other simple local conditions for the constraints (in the variable framework) that advantageously translate to some abstract lopsided condition” posed by Szegedy (2013). We also give an example where our notion of dependency graph gives better results than the classical Shearer lemma. Finally, we prove Shearer’s lemma for the dependency graph we define. For the proofs, we perform a direct probabilistic analysis that yields an exponentially small upper bound for the probability of the algorithm that searches for the desired assignment to the variables not to return a correct answer within n steps. In contrast, the method of proof that became known as the entropic method, gives an estimate of only the expectation of the number of steps until the algorithm returns a correct answer, unless the probabilities are tinkered with. PubDate: 2019-12-02

Abstract: Abstract We consider the binary classification problem when data are large and subject to unknown but bounded uncertainties. We address the problem by formulating the nonlinear support vector machine training problem with robust optimization. To do so, we analyze and propose two bounding schemes for uncertainties associated to random approximate features in low dimensional spaces. The proposed bound calculations are based on Random Fourier Features and the Nyström methods. Numerical experiments are conducted to illustrate the benefit of the technique. We also emphasize the decomposable structure of the proposed robust nonlinear formulation that allows the use of efficient stochastic approximation techniques when datasets are large. PubDate: 2019-11-15