Machine Learning models are getting more accurate, yet their complexity and opacity are also increasing. Explainable AI improved their general interpretability by quantifying the contributions of input features to predictions. Despite these advancements, practitioners still seek to gain causal insights into the underlying data-generating mechanisms. To this end, a possible solution is to rely on classical probabilistic causal analysis, which offers tools to quantify causal effects. However, causal analysis assumes a sufficient knowledge of causal structure, which is often unreachable from data alone. Indeed, causal discovery algorithms produce, at most, partial causal structures, namely Completed Partially Directed Acyclic Graphs (CPDAG). The conventional approach involves fully orienting the structure with exogenous causal knowledge through expert interaction or real-world experiments. In this paper, we focus on quantifying a specific total causal effect. Within this context, we emphasize that a partial structure can be sufficient to answer the query, and can be reached by different sequences of additional causal knowledge. Whether coming from an expert or an experiment, each addition has a cost difficult to assess a priori. The contribution of this paper is twofold: given a CPDAG and a specific query, we identify a set of irrelevant edges, and we introduce an algorithm for ranking the remaining informative edges, providing a guide to iteratively obtain a partial structure sufficient for resolving the query. Simulations show that these two contributions significantly reduce the number of requests for exogenous causal information, corroborating the feasibility of a causal impact quantification with very limited exogenous information.

Nested Rollout Policy Adaptation (NRPA) is an approach using online learning policies in a nested structure. It has achieved a great result in a variety of difficult combinatorial optimization problems. In this paper, we propose Meta-NRPA, which combines optimal stopping theory with NRPA for warm-starting and significantly improves the performance of NRPA. We also present several exploratory techniques for NRPA which enable it to perform better exploration. We establish this for three notoriously difficult problems ranging from telecommunication, transportation and coding theory namely Minimum Congestion Shortest Path Routing, Traveling Salesman Problem with Time Windows and Snake-in-the-Box. We also improve the lower bounds of the Snake-in-the-Box problem for multiple dimensions.

This paper presents a workflow for the design of transformation processes using different kinds of expert’s knowledge. It introduces POND (Process and observation ONtology Discovery), a workflow dedicated to answer expert’s questions about processes. It addresses two main issues: (1) how to represent the processes inner complexity, and (2) how to reason about processes taking into account uncertainty and causality. First, we show how to use a semantic model, an ontology, and its associated data to answer some of the expert’s questions concerning the processes, using semantic web languages and technologies. Then, we describe how to learn a predictive model, to discover new knowledge and provide explicative models by integrating the semantic model into a probabilistic relational model. The result is a complete workflow able to extensively analyze transformation processes through all their granularity levels and answer expert’s questions about their domains. An example of this workflow is given on biocomposites manufacturing for food packaging.

We propose a new framework to learn non-parametric graphical models from continuous observational data. Our method is based on concepts from information theory in order to discover independences and causality between variables: the conditional and multivariate mutual information (such as \cite{verny2017learning} for discrete models). To estimate these quantities, we propose non-parametric estimators relying on the Bernstein copula and that are constructed by exploiting the relation between the mutual information and the copula entropy \cite{ma2011mutual, belalia2017testing}. To our knowledge, this relation is only documented for the bivariate case and, for the need of our algorithms, is here extended to the conditional and multivariate mutual information. This framework leads to a new algorithm to learn continuous non-parametric Bayesian network. Moreover, we use this estimator to speed up the BIC algorithm proposed in \cite{elidan2010copula} by taking advantage of the decomposition of the likelihood function in a sum of mutual information \cite{koller2009probabilistic}. Finally, our method is compared in terms of performances and complexity with other state of the art techniques to learn Copula Bayesian Networks and shows superior results. In particular, it needs less data to recover the true structure and generalizes better on data that are not sampled from Gaussian distributions.

International audience

While Service Oriented Architectures (SOAs) are for many parts deployed online, and today often in a cloud, the testing of the systems still happens mostly locally. In this paper, we want to present the MIDAS Testing as a Service (TaaS), a cloud platform for the testing of SOAs. We focus on the testing of whole SOA orchestrations, a complex task due to the number of potential service interactions and the increasing complexity with each service that joins an orchestration. Since traditional testing does not scale well with such a complex setup, we employ a Model-based Testing (MBT) approach based on the Unified Modeling Language (UML) and the UML Testing Profile (UTP) within MIDAS. Through this, we provide methods for functional testing, security testing, and usage-based testing of service orchestrations. Through harnessing the computational power of the cloud, MIDAS is able to generate and execute complex test scenarios which would be infeasible to run in a local environment.

An Intelligent Tutoring Systems (ITS) is concerned with the construction of intelligent softwares helping students overcoming different problems in their learning process. We present in this work a novel Multicriteria Bayesian Intelligent Tutoring System MBITS used to help students overcoming their their lack of comprehension of concepts in a course. It is based on a Bayesian Network (BN) to model a course and a multicriteria approach to evaluate different solution and information gathering actions using the same set of criteria. MBITS is an automated troubleshooter presented as an interactive and easy to use Web application.

We propose a cooperative-coevolution - Parisian trend - algorithm, IMPEA (Independence Model based Parisian EA), to the problem of Bayesian networks structure estimation. It is based on an intermediate stage which consists of evaluating an independence model of the data to be modelled. The Parisian cooperative coevolution is particularly well suited to the structure of this intermediate problem, and allows to represent an independence model with help of a whole population, each individual being an independence statement, i.e. a component of the independence model. Once an independence model is estimated, a Bayesian network can be built. This two level resolution of the complex problem of Bayesian network structure estimation has the major advantage to avoid the difficult problem of direct acyclic graph representation within an evolutionary algorithm, which causes many troubles related to constraints handling and slows down algorithms. Comparative results with a deterministic algorithm, PC, on two test cases (including the Insurance BN benchmark), prove the efficiency of IMPEA, which provides better results than PC in a comparable computation time, and which is able to tackle more complex issues than PC.

Recent decision-theoric planning algorithms are able to find optimal solutions in large problems, using Factored Markov Decision Processes (FMDPs). However, these algorithms need a perfect knowledge of the structure of the problem. In this paper, we propose SDYNA, a general framework for addressing large reinforcement learning problems by trial-and-error and with no initial knowledge of their structure. SDYNA integrates incremental planning algorithms based on FMDPs with supervised learning techniques building structured representations of the problem. We describe SPITI, an instantiation of SDYNA, that uses incremental decision tree induction to learn the structure of a problem combined with an incremental version of the Structured Value Iteration algorithm. We show that SPITI can build a factored representation of a reinforcement learning problem and may improve the policy faster than tabular reinforcement learning algorithms by exploiting the generalization property of decision tree induction algorithms.