Authors:D. Magos; I. Mourtos Pages: 267 - 289 Abstract: Abstract Submodularity defines a general framework rich of important theoretical properties while accommodating numerous applications. Although the notion has been present in the literature of Combinatorial Optimization for several decades, it has been overlooked in the analysis of global constraints. The current work illustrates the potential of submodularity as a powerful tool for such an analysis. In particular, we show that the cumulative constraint, when all tasks are identical, has the submodular/supermodular representation property, i.e., it can be represented by a submodular/supermodular system of linear inequalities. Motivated by that representation, we show that the system of any two (global) constraints not necessarily of the same type, each bearing the above-mentioned property, has an integral relaxation given by the conjunction of the linear inequalities representing each individual constraint. This result is obtained through the use of the celebrated polymatroid intersection theorem. PubDate: 2017-04-01 DOI: 10.1007/s10472-016-9522-x Issue No:Vol. 79, No. 4 (2017)

Authors:Tomás de la Rosa; Raquel Fuentetaja Pages: 291 - 305 Abstract: Abstract In this paper we describe ENSEMBLE-ROLLER, a learning-based automated planner that uses a bagging approach to enhance existing techniques for learning planning policies. Previous policy-style planning and learning systems sort state successors based on action predictions from a relational classifier. However, these learning-based planners can produce several plans of bad quality, since it is very difficult to encode in a single classifier all possible situations occurring in a planning domain. We propose to use ensembles of relational classifiers to generate more robust policies. As in other applications of machine learning, the idea of the ensembles of classifiers consists of providing accuracy for particular scenarios and diversity to cover a wide range of situations. In particular, ENSEMBLE-ROLLER learns ensembles of relational decision trees for each planning domain. The control knowledge from different sets of trees is aggregated as a single prediction or applied separately in a multiple-queue search algorithm. Experimental results show that both ways of using new policies produce on average plans of better quality. PubDate: 2017-04-01 DOI: 10.1007/s10472-016-9523-9 Issue No:Vol. 79, No. 4 (2017)

Authors:Jan Triska; Vilem Vychodil Pages: 307 - 335 Abstract: Abstract We study logic for reasoning with if-then formulas describing dependencies between attributes of objects which are observed in consecutive points in time. We introduce semantic entailment of the formulas, show its fixed-point characterization, investigate closure properties of model classes, present an axiomatization and prove its completeness, and investigate alternative axiomatizations and normalized proofs. We investigate decidability and complexity issues of the logic and prove that the entailment problem is NP-hard and belongs to EXPSPACE. We show that by restricting to predictive formulas, the entailment problem is decidable in pseudo-linear time. PubDate: 2017-04-01 DOI: 10.1007/s10472-016-9526-6 Issue No:Vol. 79, No. 4 (2017)

Authors:Mikko Lauri; Aino Ropponen; Risto Ritala Pages: 337 - 370 Abstract: Abstract We consider the problem of an agent traversing a directed graph with the objective of maximizing the probability of reaching a goal node before a given deadline. Only the probability of the travel times of edges is known to the agent. The agent must balance between traversal actions towards the goal, and delays due to actions improving information about graph edge travel times. We describe the relationship of the problem to the more general partially observable Markov decision process. Further, we show that if edge travel times are independent and the underlying directed graph is acyclic, a closed loop solution can be computed. The solution specifies whether to execute a traversal or information-gathering action as a function of the current node, the time remaining until the deadline, and the information about edge travel times. We present results from two case studies, quantifying the usefulness of information-gathering as opposed to applying only traversal actions. PubDate: 2017-04-01 DOI: 10.1007/s10472-016-9527-5 Issue No:Vol. 79, No. 4 (2017)

Authors:Piotr Wojciechowski; Pavlos Eirinakis; K. Subramani Pages: 371 - 392 Abstract: Erratum to: Ann Math Artif Intell (2017) 79:245-266 DOI 10.1007/s10472-016-9525-7 Owing to an error in the production process, the following article has been published incorrectly online. The article including the illustrations is presented hereafter. PubDate: 2017-04-01 Issue No:Vol. 79, No. 4 (2017)

Authors:Dan Gutfreund; Aryeh Kontorovich; Ran Levy; Michal Rosen-Zvi Pages: 129 - 144 Abstract: Abstract In the standard agnostic multiclass model, <instance, label > pairs are sampled independently from some underlying distribution. This distribution induces a conditional probability over the labels given an instance, and our goal in this paper is to learn this conditional distribution. Since even unconditional densities are quite challenging to learn, we give our learner access to <instance, conditional distribution > pairs. Assuming a base learner oracle in this model, we might seek a boosting algorithm for constructing a strong learner. Unfortunately, without further assumptions, this is provably impossible. However, we give a new boosting algorithm that succeeds in the following sense: given a base learner guaranteed to achieve some average accuracy (i.e., risk), we efficiently construct a learner that achieves the same level of accuracy with arbitrarily high probability. We give generalization guarantees of several different kinds, including distribution-free accuracy and risk bounds. None of our estimates depend on the number of boosting rounds and some of them admit dimension-free formulations. PubDate: 2017-03-01 DOI: 10.1007/s10472-015-9465-7 Issue No:Vol. 79, No. 1-3 (2017)

Authors:Seishi Ouchi; Tomohiko Okayama; Keisuke Otaki; Ryo Yoshinaka; Akihiro Yamamoto Pages: 181 - 203 Abstract: Abstract This paper is concerned with a sufficient condition under which a concept class is learnable in Gold’s classical model of identification in the limit from positive data. The standard principle of learning algorithms working under this model is called the MINL strategy, which is to conjecture a hypothesis representing a minimal concept among the ones consistent with the given positive data. The minimality of a concept is defined with respect to the set-inclusion relation – the strategy is semantics-based. On the other hand, refinement operators have been developed in the field of learning logic programs, where a learner constructs logic programs as hypotheses consistent with given logical formulae. Refinement operators have syntax-based definitions – they are defined based on inference rules in first-order logic. This paper investigates the relation between the MINL strategy and refinement operators in inductive inference. We first show that if a hypothesis space admits a refinement operator with certain properties, the concept class will be learnable by an algorithm based on the MINL strategy. We then present an additional condition that ensures the learnability of the class of unbounded finite unions of concepts. Furthermore, we show that under certain assumptions a learning algorithm runs in polynomial time. PubDate: 2017-03-01 DOI: 10.1007/s10472-015-9458-6 Issue No:Vol. 79, No. 1-3 (2017)

Authors:Xia Qu; Prashant Doshi Pages: 205 - 227 Abstract: Abstract Experiments show that in sequential bargaining games ( \(\mathcal {SBG}\) ), subjects usually deviate from game-theoretic predictions. Previous explanations have focused on considerations of fairness in the offers, and social utility functions have been formulated to model the data. However, a recent explanation by Ho and Su (Manag. Sci. 59(2), 452–469 2013) for observed deviations from game-theoretic predictions in sequential games such as the Centipede game is that players engage in limited backward induction. In this article, a suite of new and existing computational models that integrate different choice models with utility functions are comprehensively evaluated on \(\mathcal {SBG}\) data. These include DeBruyn and Bolton’s recursive quantal response with social utility functions, those based on Ho and Su’s dynamic level-k, and analogous extensions of the cognitive hierarchy with dynamic components. Our comprehensive analysis reveals that in extended \(\mathcal {SBG}\) with 5 rounds, models that capture violations of backward induction perform better than those that model fairness. However, we did not observe this result for \(\mathcal {SBG}\) with less rounds, and fairness of the offer remains a key consideration in these games. These findings contribute to the broader observation that non-social factors play a significant role in non-equilibrium play of sequential games. PubDate: 2017-03-01 DOI: 10.1007/s10472-015-9481-7 Issue No:Vol. 79, No. 1-3 (2017)

Authors:Takahisa Toda Pages: 229 - 244 Abstract: Abstract Dualization of Boolean functions is a fundamental problem that appears in various fields such as artificial intelligence, logic, data mining, etc. For monotone Boolean functions, many empirical researches that focus on practical efficiency have recently been done. We extend our previous work for monotone dualization and present a novel method for dualization that allows us to handle any Boolean function, including non-monotone Boolean functions. We furthermore present a variant of this method in cooperation with all solutions solver. By experiments we evaluate efficiency and characteristics of our methods. PubDate: 2017-03-01 DOI: 10.1007/s10472-016-9520-z Issue No:Vol. 79, No. 1-3 (2017)

Authors:Lior Aronshtam; Havazelet Cohen; Tammar Shrot Abstract: Abstract This article focuses on the question of whether a certain candidate’s (player’s) chance to advance further in a tennis tournament can be increased when the ordering of the tournament can be controlled (manipulated by the organizers) according to his own preferences. Is it possible to increase the number of ranking points a player will receive? And most importantly, can it be done in reasonable computational time? The answers to these questions differ for different settings. e.g., the information available on the outcome of each game, the significance of the number of points gained or of the number of games won. We analyzed five different variations of these tournament questions. First the complexity hardness of trying to control the tournaments is shown. Then, the tools of parametrized complexity are used to investigate the source of the problems’ hardness. Specifically, we check whether this hardness holds when the size of the problem is bounded. The findings of this analysis show that it is possible under certain circumstances to control the tournament in favor of a specific candidate in order to help him advance further in the tournament. PubDate: 2017-04-24 DOI: 10.1007/s10472-017-9549-7

Authors:John McCabe-Dansted; Mark Reynolds Abstract: Abstract Attempts to manage the reasoning about systems with fairness properties are long running. The popular but restricted Computational Tree Logic (CTL) is amenable to automated reasoning but has difficulty expressing some fairness properties. More expressive languages such as CTL* and CTL+ are computationally complex. The main contribution of this paper is to show the usefulness and practicality of employing the bundled variants of these languages to handle fairness. In particular we present a tableau for a bundled variant of CTL that still has the similar computational complexity to the CTL tableau and a simpler implementation. We further show that the decision problem remains in EXPTIME even if a bounded number of CTL* fairness constraints are allowed in the input formulas. By abandoning limit closure the bundled logics can simultaneously be easier to automate and express many typical fairness constraints. PubDate: 2017-04-19 DOI: 10.1007/s10472-017-9546-x

Authors:Vladimir Vovk; Dusko Pavlovic Abstract: Abstract We construct universal prediction systems in the spirit of Popper’s falsifiability and Kolmogorov complexity and randomness. These prediction systems do not depend on any statistical assumptions (but under the IID assumption they dominate, to within the usual accuracy, conformal prediction). Our constructions give rise to a theory of algorithmic complexity and randomness of time containing analogues of several notions and results of the classical theory of Kolmogorov complexity and randomness. PubDate: 2017-04-19 DOI: 10.1007/s10472-017-9547-9

Authors:A. Zaytsev; E. Burnaev Abstract: Abstract Engineers widely use Gaussian process regression framework to construct surrogate models aimed to replace computationally expensive physical models while exploring design space. Thanks to Gaussian process properties we can use both samples generated by a high fidelity function (an expensive and accurate representation of a physical phenomenon) and a low fidelity function (a cheap and coarse approximation of the same physical phenomenon) while constructing a surrogate model. However, if samples sizes are more than few thousands of points, computational costs of the Gaussian process regression become prohibitive both in case of learning and in case of prediction calculation. We propose two approaches to circumvent this computational burden: one approach is based on the Nyström approximation of sample covariance matrices and another is based on an intelligent usage of a blackbox that can evaluate a low fidelity function on the fly at any point of a design space. We examine performance of the proposed approaches using a number of artificial and real problems, including engineering optimization of a rotating disk shape. PubDate: 2017-04-05 DOI: 10.1007/s10472-017-9545-y

Authors:Evgeny Burnaev; Ivan Panin; Bruno Sudret Abstract: Abstract Global sensitivity analysis aims at quantifying respective effects of input random variables (or combinations thereof) onto variance of a physical or mathematical model response. Among the abundant literature on sensitivity measures, Sobol indices have received much attention since they provide accurate information for most of models. We consider a problem of experimental design points selection for Sobol’ indices estimation. Based on the concept of D-optimality, we propose a method for constructing an adaptive design of experiments, effective for calculation of Sobol’ indices based on Polynomial Chaos Expansions. We provide a set of applications that demonstrate the efficiency of the proposed approach. PubDate: 2017-03-30 DOI: 10.1007/s10472-017-9542-1

Authors:Weifu Ding; Jiangshe Zhang Abstract: Abstract This paper presents a novel online object tracking algorithm with sparse representation for learning effective appearance models under a particle filtering framework. Compared with the state-of-the-art ℓ 1 sparse tracker, which simply assumes that the image pixels are corrupted by independent Gaussian noise, our proposed method is based on information theoretical Learning and is much less sensitive to corruptions; it achieves this by assigning small weights to occluded pixels and outliers. The most appealing aspect of this approach is that it can yield robust estimations without using the trivial templates adopted by the previous sparse tracker. By using a weighted linear least squares with non-negativity constraints at each iteration, a sparse representation of the target candidate is learned; to further improve the tracking performance, target templates are dynamically updated to capture appearance changes. In our template update mechanism, the similarity between the templates and the target candidates is measured by the earth movers’ distance(EMD). Using the largest open benchmark for visual tracking, we empirically compare two ensemble methods constructed from six state-of-the-art trackers, against the individual trackers. The proposed tracking algorithm runs in real-time, and using challenging sequences performs favorably in terms of efficiency, accuracy and robustness against state-of-the-art algorithms. PubDate: 2017-03-23 DOI: 10.1007/s10472-017-9543-0

Authors:Federico Bergenti; Stefania Monica Abstract: Abstract This paper describes an algorithm to enforce hyper-arc consistency of polynomial constraints defined over finite domains. First, the paper describes the language of so called polynomial constraints over finite domains, and it introduces a canonical form for such constraints. Then, the canonical form is used to transform the problem of testing the satisfiability of a constraint in a box into the problem of studying the sign of a related polynomial function in the same box, a problem which is effectively solved by using the modified Bernstein form of polynomials. The modified Bernstein form of polynomials is briefly discussed, and the proposed hyper-arc consistency algorithm is finally detailed. The proposed algorithm is a subdivision procedure which, starting from an initial approximation of the domains of variables, removes values from domains to enforce hyper-arc consistency. PubDate: 2017-03-20 DOI: 10.1007/s10472-017-9544-z

Authors:Claus Bendtsen; Andrea Degasperi; Ernst Ahlberg; Lars Carlsson Abstract: Abstract The high cost for new medicines is hindering their development and machine learning is therefore being used to avoid carrying out physical experiments. Here, we present a comparison between three different machine learning approaches in a classification setting where learning and prediction follow a teaching schedule to mimic the drug discovery process. The approaches are standard SVM classification, SVM based multi-kernel classification and SVM classification based on learning using privileged information. Our two main conclusions are derived using experimental in-vitro data and compound structure descriptors. The in-vitro data is assumed to i) be completely absent in the standard SVM setting, ii) be available at all times when applying multi-kernel learning, or iii) be available as privileged information during training only. The structure descriptors are always available. One conclusion is that multi-kernel learning has higher odds than standard SVM in producing higher accuracy. The second is that learning using privileged information does not have higher odds than the standard SVM, although it may improve accuracy when the training sets are small. PubDate: 2017-03-18 DOI: 10.1007/s10472-017-9541-2

Authors:Vladimir Vovk; Ilia Nouretdinov; Valentina Fedorova; Ivan Petej; Alex Gammerman Abstract: Abstract We study optimal conformity measures for various criteria of efficiency of set-valued classification in an idealised setting. This leads to an important class of criteria of efficiency that we call probabilistic and argue for; it turns out that the most standard criteria of efficiency used in literature on conformal prediction are not probabilistic unless the problem of classification is binary. We consider both unconditional and label-conditional conformal prediction. PubDate: 2017-03-14 DOI: 10.1007/s10472-017-9540-3