We present a graphical criterion for reading dependencies from the minimal directed independence map G of a graphoid p, under the assumption that G is a polytree and p satisfies weak transitivity. We prove that the criterion is sound and complete. We argue that assuming weak transitivity is not too restrictive. 1

expressed in this material are those of the author and do not necessarily reflect the views of

This paper presents an extension to the Conservative PC algorithm which is able to detect violations of adjacency faithfulness under causal sufficiency and triangle faithfulness. Violations can be characterized by pseudo-independent relations and equivalent edges, both generating a pattern of conditional independencies that cannot be modeled faithfully. Both cases lead to uncertainty about specific parts of the skeleton of the causal graph. This is modeled by an f-pattern. We proved that our Very Conservative PC algorithm is able to correctly learn the f-pattern. We argue that the solution also applies for the finite sample case if we accept that only strong edges can be identified. Experiments based on simulations show that the rate of false edge removals is significantly reduced, at the expense of uncertainty on the skeleton and a higher sensitivity for accidental correlations. 1

We prove that perfect distributions exist when using a finite number of bits to represent the parameters of a Bayesian network. In addition, we provide an upper bound on the probability of sampling a non-perfect distribution when using a fixed number of bits for the parameters and that the upper bound approaches zero exponentially fast as one increases the number of bits. We also provide an upper bound on the number of bits needed to guarantee that a distribution sampled from a uniform Dirichlet distribution is perfect with probability greater than 1/2. 1

We present a sound and complete graphical criterion for reading dependencies from the minimal undirected independence map G of a graphoid M that satisfies weak transitivity. Here, complete means that it is able to read all the dependencies in M that can be derived by applying the graphoid properties and weak transitivity to the dependencies used in the construction of G and the independencies obtained from G by vertex separation. We argue that assuming weak transitivity is not too restrictive. As an intermediate step in the derivation of the graphical criterion, we prove that for any undirected graph G there exists a strictly positive discrete probability distribution with the prescribed sample spaces that is faithful to G. We also report an algorithm that implements the graphical criterion and whose running time is considered to be at most O(n 2 (e+n)) for n nodes and e edges. Finally, we illustrate how the graphical criterion can be used within bioinformatics to identify biologically meaningful gene dependencies.

We introduce a novel, sound, sample-efficient, and highly-scalable algorithm for variable selection for classification, regression and prediction called HITON. The algorithm works by inducing the Markov Blanket of the variable to be classified or predicted. A wide variety of biomedical tasks with different characteristics were used for an empirical evaluation. Namely, (i) bioactivity prediction for drug discovery, (ii) clinical diagnosis of arrhythmias, (iii) bibliographic text categorization, (iv) lung cancer diagnosis from gene expression array data, and (v) proteomics-based prostate cancer detection. State-of-the-art algorithms for each domain were selected for baseline comparison. Results: (1) HITON reduces the number of variables in the prediction models by three orders of magnitude relative to the original variable set while improving or maintaining accuracy. (2) HITON outperforms the baseline algorithms by selecting more than two orders-ofmagnitude smaller variable sets than the baselines, in the selected tasks and datasets.

This paper introduces design principles for modular Bayesian fusion systems which can (i) cope with large quantities of heterogeneous information and (ii) can adapt to changing constellations of information sources on the fly. The presented approach exploits the locality of relations in causal probabilistic processes, which facilitates decentralized modeling and information fusion. Observed events resulting from stochastic causal processes can be modeled with the help of Bayesian networks, compact and mathematically rigorous probabilistic models. With the help of the theory of Bayesian networks and factor graphs we derive design and organization rules for modular fusion systems which implement exact belief propagation without centralized configuration and fusion control. These rules are applied in distributed perception networks (DPN), a multi agent systems approach to distributed Bayesian information fusion. While each DPN agent has limited fusion capabilities, multiple DPN agents can autonomously collaborate to form complex modular fusion systems. Such self-organizing systems of agents can adapt to the available information sources at runtime and can infer critical hidden events through interpretation of complex patterns consisting of many heterogeneous observations.

Docteur en Sciences de l’Ingénieur

ABSTRACT: In this paper, we study sequences of regressions in joint or single responses given a set of context variables, where a dependence structure of interest is captured by a regression graph. These graphs have nodes representing random variables and three types of edge. Their set of missing edges defines the independence structure of the graph provided two properties are used that are not common to all probability distributions, named the intersection and the composition property. We derive the additionally needed properties for tracing the effects of single active paths and for excluding any canceling of effects due to several paths connecting the same pair of nodes. For this, we use the notion of a generating process for the joint distribution and derive new properties of an edge matrix calculus for transforming graphs. One key is the M-matrix property of each regularized square edge matrix, others are the proposed notions of traceable regressions and of singleton transitivity.

Over the past two decades, several consistent procedures have been designed to infer causal conclusions from observational data. We prove that if the true causal network might be an arbitrary, linear Gaussian network or a discrete Bayes network, then every unambiguous causal conclusion produced by a consistent method from non-experimental data is subject to reversal as the sample size increases any finite number of times. That result, called the causal flipping theorem, extends prior results to the effect that causal discovery cannot be reliable on a given sample size. We argue that since repeated flipping of causal conclusions is unavoidable in principle for consistent methods, the best possible discovery methods are consistent methods that retract their earlier conclusions no more than necessary. A series of simulations of various methods across a wide range of sample sizes illustrates concretely both the theorem and the principle of comparing methods in terms of retractions. 1