Structured probabilistic inference

Probabilistic inference is among the main topics with reasoning in uncertainty in AI. For this purpose, Bayesian Networks (BNs) is one of the most successful and efficient Probabilistic Graphical Model (PGM) so far. Since the mid-90s, a growing number of BNs extensions have been proposed. Object-oriented, entity-relationship and first-order logic are the main representation paradigms used to extend BNs. While entity-relationship and first-order models have been successfully used for machine learning in defining lifted probabilistic inference, object-oriented models have been mostly underused. Structured inference, which exploits the structural knowledge encoded in an object-oriented PGM, is a surprisingly unstudied technique. In this paper we propose a full object-oriented framework for PRM and propose two extensions of the state-of-the-art structured inference algorithm: SPI which removes the major flaws of existing algorithms and SPISBB which largely enhances SPI by using d-separation.     

Anytime anyspace probabilistic inference

This paper investigates methods that balance time and space constraints against the quality of Bayesian network inferences––we explore the three-dimensional spectrum of “time × space × quality” trade-offs. The main result of our investigation is the adaptive conditioning algorithm, an inference algorithm that works by dividing a Bayesian network into sub-networks and processing each sub-network with a combination of exact and anytime strategies. The algorithm seeks a balanced synthesis of probabilistic techniques for bounded systems. Adaptive conditioning can produce inferences in situations that defy existing algorithms, and is particularly suited as a component of bounded agents and embedded devices.     

A continuous approach to inductive inference

In this paper we describe an interior point mathematical programming approach to inductive inference. We list several versions of this problem and study in detail the formulation based on hidden Boolean logic. We consider the problem of identifying a hidden Boolean function:{0, 1} ⁿ {0, 1} using outputs obtained by applying a limited number of random inputs to the hidden function. Given this input—output sample, we give a method to synthesize a Boolean function that describes the sample. We pose the Boolean Function Synthesis Problem as a particular type of Satisfiability Problem. The Satisfiability Problem is translated into an integer programming feasibility problem, that is solved with an interior point algorithm for integer programming. A similar integer programming implementation has been used in a previous study to solve randomly generated instances of the Satisfiability Problem. In this paper we introduce a new variant of this algorithm, where the Riemannian metric used for defining the search region is dynamically modified. Computational results on 8-, 16- and 32-input, 1-output functions are presented. Our implementation successfully identified the majority of hidden functions in the experiment.     

On cautious probabilistic inference and default detachment

Conditional probabilities are one promising and widely used approach to model uncertainty in information systems. This paper discusses the DUCK-calculus, which is founded on the cautious approach to uncertain probabilistic inference. Based on a set of sound inference rules, derived probabilistic information is gained by local bounds propagation techniques. Precision being always a central point of criticism to such systems, we demonstrate that DUCK need not necessarily suffer from these problems. We can show that the popular Bayesian networks are subsumed by DUCK, implying that precise probabilities can be deduced by local propagation techniques, even in the multiply connected case. A comparative study with INFERNO and with inference techniques based on global operations-research techniques yields quite favorable results for our approach. Since conditional probabilities are also suited to model nonmonotonic situations by considering different contexts, we investigate the problems of maximal and relevant contexts, needed to draw default conclusions about individuals.     

Speeding-up structured probabilistic inference using pattern mining

In many domains where experts are the main source of knowledge, e.g., in reliability and risk management, a framework well suited for modeling, maintenance and exploitation of complex probabilistic systems is essential. In these domains, models usually define closed-world systems and result from the aggregation of multiple patterns repeated many times. Object Oriented-based Frameworks such as Probabilistic Relational Models (PRM) thus offer an effective way to represent such systems. They define patterns as classes and substitute large Bayesian networks (BN) by graphs of instances of these classes. In this framework, Structured Inference avoids many computation redundancies by exploiting class knowledge, hence reducing BN inference times by orders of magnitude. However, to keep modeling and maintenance costs low, object oriented-based framework’s classes often encode only generic situations. More complex situations, even those repeated many times, are only represented by combinations of instances. In this paper, we propose to determine online such combination patterns and exploit them as classes to speed-up Structured Inference. We prove that determining an optimal set of patterns is NP-hard. We also provide an efficient algorithm to approximate this set and show numerical experiments that highlight its practical efficiency.     

Generalizing inference rules in a coherence-based probabilistic default reasoning

In this paper we first recall some notions and results on the coherence-based probabilistic treatment of uncertainty. Then, we deepen some probabilistic aspects in nonmonotonic reasoning, by generalizing OR, CM, and Cut rules. We also illustrate the degradation of these inference rules when the number of premises increases. Finally, we show that the lower bounds obtained when applying OR and Quasi-Conjunction inference rules coincide, respectively, with Hamacher and Lukasiewicz t-norms; the upper bounds in both rules coincide with Hamacher t-conorm.     

Randomized statistical inference: A unified statistical inference frame of frequentist,fiducial, and Bayesian inference

We propose randomized inference(RI), a new statistical inference approach. RI may be realized through a randomized estimate(RE) of a parameter vector, which is a random vector that takes values in the parameter space with a probability density function(PDF) that depends on the sample or sufficient statistics,such as the posterior distributions in Bayesian inference. Based on the PDF of an RE of an unknown parameter,we propose a framework for both the vertical density representation(VDR) test and the construction of a confidence region. This approach is explained with the aid of examples. For the equality hypothesis of multiple normal means without the condition of variance homogeneity, we present an exact VDR test, which is shown as an extension of one-way analysis of variance(ANOVA). In the case of two populations, the PDF of the Welch statistics is given by using the RE. Furthermore, through simulations, we show that the empirical distribution function, the approximated t, and the RE distribution function of Welch statistics are almost equal. The VDR test of the homogeneity of variance is shown to be more efficient than both the Bartlett test and the revised Bartlett test. Finally, we discuss the prospects of RI.     

Inductive and deductive science

An important unfinished problem is the theory of transition from deflagration to detonation DDT. The understanding of this problem is critical for the safe handling of explosives. Some of the successes of deflagration and detonation theory are outlined. For detonation to occur in a cylindrical explosive, for example, one must remember that as a shock wave moves into cold explosive the shock loses energy and is dying out. Behind a developed wave front there is a succession of increasingly exothermic reaction surfaces rising to a maximum at the Chapman-Jouguet surface where rarefaction begins. A sufficiently sharp blow or an intense beam of laser light falling on unreacted explosive generates this type of structure which travels through the material as a self supporting shock wave. How to avoid forming such structures which will grow into travelling self supporting waves has many aspects. Here we consider: 1. Starvation kinetics which arise from a change of reaction mechanism due to the sudden introduction of unburned material into a hot environment. As the gradient from unburned to burned material rises a situation develops where the fastest way to get energy into the bond that breaks is not directly from translation to this bond but through a reservoir which readily communicates energy. This shifts the developing bottle neck of energizing the molecule (starvation kinetics) to a new pathway with interesting consequences; 2. Turbulence promotes initiation of detonation by sharpening the concentration and temperature gradients between unburned and burned materials; 3. The dependence of shock wave velocity, due to momentum loss out the side is related to the diameter of the explosive; and 4. The curvature of the wave front is related to drag of telescoping cylinders on the central axis of the cylindrical explosive.     

Finite mixture-of-gamma distributions: estimation,inference, and model-based clustering

Advances in Data Analysis and Classification - Finite mixtures of (multivariate) Gaussian distributions have broad utility, including their usage for model-based clustering. There is increasing...     

Reject inference,augmentation, and sample selection

Many researchers see the need for reject inference in credit scoring models to come from a sample selection problem whereby a missing variable results in omitted variable bias. Alternatively, practitioners often see the problem as one of missing data where the relationship in the new model is biased because the behaviour of the omitted cases differs from that of those who make up the sample for a new model. To attempt to correct for this, differential weights are applied to the new cases. The aim of this paper is to see if the use of both a Heckman style sample selection model and the use of sampling weights, together, will improve predictive performance compared with either technique used alone. This paper will use a sample of applicants in which virtually every applicant was accepted. This allows us to compare the actual performance of each model with the performance of models which are based only on accepted cases.     

Pretopologies, preuniformities and probabilistic metric spaces

Summary A pretopology on a given set can be generated from a filter of reflexive relations on that set (we call such a structure a preuniformity). We show that the familly of filters inducing a given pretopology on Xform a complete lattice in the lattice of filters on X. The smallest and largest elements of that lattice are explicitly given. The largest element is characterized by a condition which is formally equivalent to a property introduced by Knaster--Kuratowski--Mazurkiewicz in their well known proof of Brouwer's fixed point theorem. Menger spaces and probabilistic metric spaces also generate pretopologies. Semi-uniformities and pretopologies associated to a possibly nonseparated Menger space are completely characterized.     

Singular electromagnetic fields: inductive approach

In a non-convex polyhedral domain, we describe the local trace (i.e. defined on a face) of the normal derivative of an $L^2$ function, with $L^2$ Laplacian. We then provide generalized integration by parts formulae for the Laplace, divergence and curl operators. Finally, these results allow us to split electromagnetic fields into regular and singular parts, which can be characterized. To cite this article: F. Assous et al., C. R. Acad. Sci. Paris, Ser. I 341 (2005).     

Ternary and quaternary deductive terms for Nelson algebras

We exhibit explicit ternary and quaternary deductive terms for the variety of Nelson algebras. The results partially answer a question of Blok and Pigozzi.     

Categorical abstract algebraic logic: The criterion for deductive equivalence

Equivalent deductive systems were introduced in [4] with the goal of treating 1‐deductive systems and algebraic 2‐deductive systems in a uniform way. Results of [3], appropriately translated and strengthened, show that two deductive systems over the same language type are equivalent if and only if their lattices of theories are isomorphic via an isomorphism that commutes with substitutions. Deductive equivalence of π‐institutions [14, 15] generalizes the notion of equivalence of deductive systems. In [15, Theorem 10.26] this criterion for the equivalence of deductive systems was generalized to a criterion for the deductive equivalence of term π‐institutions, forming a subclass of all π‐institutions that contains those π‐institutions directly corresponding to deductive systems. This criterion is generalized here to cover the case of arbitrary π‐institutions.     

Normal projection: deterministic and probabilistic algorithms

We consider the following problem： for a collection of points in an n-dimensional space, find a linear projection mapping the points to the ground field such that different points are mapped to different values. Such projections are called normal and are useful for making algebraic varieties into normal positions. The points may be given explicitly or implicitly and the coefficients of the projection come from a subset S of the ground field. If the subset S is small, this problem may be hard. This paper deals with relatively large S, a deterministic algorithm is given when the points are given explicitly, and a lower bound for success probability is given for a probabilistic algorithm from in the literature.     

Mean,variance, and probabilistic criteria in finite Markov decision processes: A review

This paper is a survey of papers which make use of nonstandard Markov decision process criteria (i.e., those which do not seek simply to optimize expected returns per unit time or expected discounted return). It covers infinite-horizon nondiscounted formulations, infinite-horizon discounted formulations, and finite-horizon formulations. For problem formulations in terms solely of the probabilities of being in each state and taking each action, policy equivalence results are given which allow policies to be restricted to the class of Markov policies or to the randomizations of deterministic Markov policies. For problems which cannot be stated in such terms, in terms of the primitive state setI, formulations involving a redefinition of the states are examined.The author would like to thank two referees for a very thorough and helpful referceing of the original article and for the extra references (Refs. 47–52) now added to the original reference list.