Papers by Francisco Javier Diez
Distributed reasoning and learning in Bayesian expert systems
This paper presents Bayesian networks as a framework for distributed reasoning in expert systems.... more This paper presents Bayesian networks as a framework for distributed reasoning in expert systems. We discuss methods for evidence propagation, for learning, with emphasis on sequential learning, and for generating linguistic explanations. When a parallel implementation is possible, we describe the computational power, i.e. the information that must be stored and the local calculations that must be performed at every node, in order to get a distributed expert system. Finally, a brief comparison to neural networks is ooered.
Criteria for Combining Knowledge from Different Sources in ProbabilisticModels
Building probabilistic and decision-analytic models requires a considerable knowledge engineering... more Building probabilistic and decision-analytic models requires a considerable knowledge engineering effort in which obtaining numerical parameters is especially daunting. Often knowledge engineers combine various sources of information, such as information reported in textbooks and professional literature, available statistics, and data collected in practical settings. We show that combining probabilistic knowledge that originates from different sources requires utmost care. In particular, we demonstrate that even such seemingly population-independent characteristics as sensitivity and specificity of medical symptoms can vary within a population, depending purely on how the data are collected. We offer guidelines for detecting when different sources of data can be safely combined. Our analysis shows that a knowledge engineer should exercise much care in building practical models.
Distributed Inference in Bayesian Networks
Cybernetics and Systems, 1994
Bayesian networks originated as a framework for distributed reasoning. In singly connected networ... more Bayesian networks originated as a framework for distributed reasoning. In singly connected networks, there exists an elegant inference algorithm that can be implemented in parallel having a processor for every node. It can be extended to take advantage of the OR-gate, a model of interaction among causes that simplifies knowledge acquisition and evidence propagation. We also discuss two exact and
Causal Bayesian Reasoning in Medicine
Cybernetics and Systems, 1992
Departamento de InformAtica, Ciencias, UNED, Senda del Rey, E-28040 Madrid, Spain ... We propose ... more Departamento de InformAtica, Ciencias, UNED, Senda del Rey, E-28040 Madrid, Spain ... We propose a knowledge representation architecture organized in three levels-a causal network (containing the domain knowledge), medical strategies and causal reasoning-and ...
Selecting treatment strategies with dynamic limited-memory influence diagrams
Artificial Intelligence in Medicine, 2007
The development of dynamic limited-memory influence diagrams as a framework for representing fact... more The development of dynamic limited-memory influence diagrams as a framework for representing factorized infinite-horizon partially observable Markov decision processes (POMDPs), the introduction of algorithms for their (approximate) solution, and the application to a dynamic decision problem in clinical oncology. A dynamic limited-memory influence diagram for high-grade carcinoid tumor pathophysiology was developed in collaboration with an expert physician. Three algorithms, known as single policy updating, single rule updating, and simulated annealing have been examined for approximating the optimal treatment strategy from a space of 10(19) possible strategies. Single policy updating proved intractable for finding a treatment strategy for carcinoid tumors. Single rule updating and simulated annealing both found the treatment strategy that is applied by physicians in practice. Dynamic limited-memory influence diagrams are a suitable framework for the representation of factorized infinite-horizon POMDPs, and the developed algorithms find acceptable solutions under the assumption of limited memory about past observations. The framework allows for finding reasonable treatment strategies for complex dynamic decision problems in medicine.
DIAVAL, a Bayesian expert system for echocardiography
Artificial Intelligence in Medicine, 1997
DIAVAL is an expert system for the diagnosis of heart diseases, including several kinds of data, ... more DIAVAL is an expert system for the diagnosis of heart diseases, including several kinds of data, mainly from echocardiography. The first part of this paper is devoted to the causal probabilistic model which constitutes the knowledge base of the expert system in the form of a Bayesian network, emphasizing the importance of the OR gate. The second part deals with the process of diagnosis, which consists of computing the a posteriori probabilities, selecting the most probable and most relevant diagnoses, and generating a written report. It also describes the results of the evaluation of the program.
Dynamic limited-memory influence diagrams (DLIMIDs) have been developed as a framework for decisi... more Dynamic limited-memory influence diagrams (DLIMIDs) have been developed as a framework for decision-making under uncertainty over time. We show that DLIMIDs constructed from twostage temporal LIMIDs can represent infinitehorizon decision processes. Given a treatment strategy supplied by the physician, DLIMIDs may be used as prognostic models. The theory is applied to determine the prognosis of patients that suffer from an aggressive type of neuroendocrine tumor. * This research was sponsored by the Dutch Institute Madrid and by the Dutch Science Foundation under grant number 612.066.201. 1 Much recent POMDP research has been concerned with taking advantage of such factorizations [Boutilier et al., 1996a].
The hardest task in knowledge engineering for probabilistic graphical models, such as Bayesian ne... more The hardest task in knowledge engineering for probabilistic graphical models, such as Bayesian networks and influence diagrams, is obtaining their numerical parameters. Models based on acyclic directed graphs and composed of discrete variables, currently most common in practice, require for every variable a number of parameters that is exponential in the number of its parents in the graph, which makes elicitation from experts or learning from databases a daunting task. In this paper, we review the so called canonical models, whose main advantage is that they require much fewer parameters. We propose a general framework for them, based on three categories: deterministic models, ICI models, and simple canonical models. ICI models rely on the concept of independence of causal influence and can be subdivided into noisy and leaky. We then analyze the most common families of canonical models (the OR/MAX, the AND/MIN, and the noisy XOR), generalizing them and offering criteria for applying them in practice. We also briefly review temporal canonical models.
Mathematics
OpenMarkov is an open-source software tool for probabilistic graphical models. It has been develo... more OpenMarkov is an open-source software tool for probabilistic graphical models. It has been developed especially for medicine, but has also been used to build applications in other fields and for tuition, in more than 30 countries. In this paper we explain how to use it as a pedagogical tool to teach the main concepts of Bayesian networks and influence diagrams, such as conditional dependence and independence, d-separation, Markov blankets, explaining away, optimal policies, expected utilities, etc., and some inference algorithms: logic sampling, likelihood weighting, and arc reversal. The facilities for learning Bayesian networks interactively can be used to illustrate step by step the performance of the two basic algorithms: search-and-score and PC.
Uncertainty in Artificial Intelligence, 1993
Artificial Intelligence, 1996
Local conditioning (LC) is an exact algorithm for computing probability in Bayesian networks, dev... more Local conditioning (LC) is an exact algorithm for computing probability in Bayesian networks, developed as an extension of Kim and Pearl's algorithm for singly-connected networks. A list of variables associated to each node guarantees that only the nodes inside a loop are conditioned on the variable which breaks it. The main advantage of this algorithm is that it computes the probability directly on the original network instead of building a cluster tree, and this can save time when debugging a model and when the sparsity of evidence allows a pruning of the network. The algorithm is also advantageous when some families in the network interact through AND/OR gates. A parallel implementation of the algorithm with a processor for each node is possible even in the case of multiply-connected networks.
Análisis de decisiones en medicina con modelos gráficos probabilistas
The naïve-Bayes method. Bayesian networks. Influence diagrams. Decision analysis networks. Cost-... more The naïve-Bayes method. Bayesian networks. Influence diagrams. Decision analysis networks. Cost-effectiveness analysis. Markov models. ConclusionN
Parameter adjustement in Bayes networks. The generalized noisy OR--gate
Uai, 1993
The usual way of applying Bayesian n etworks to the modelling of temporal processes consists in d... more The usual way of applying Bayesian n etworks to the modelling of temporal processes consists in d iscretizing time a nd creating an instance of each random variable for each point in time. This method leads to large a nd complex networks. We present a new approach called Net of Irreversible Events in Discrete Time (NIEDT), for temporal reasoning in
Lecture Notes in Computer Science, 2013
Severe and profound hearing losses can be treated with cochlear implants (CI). Given that a CI ma... more Severe and profound hearing losses can be treated with cochlear implants (CI). Given that a CI may have up to 150 tunable parameters, adjusting them is a highly complex task. For this reason, we decided to build a decision support system based on a new type of probabilistic graphical model (PGM) that we call tuning networks. Given the results of a set of audiological tests and the current status of the parameter set, the system looks for the set of changes in the parameters of the CI that will lead to the biggest improvement in the user's hearing ability. Because of the high number of variables involved in the problem we have used an object-oriented approach to build the network. The prototype has been informally evaluated comparing its advice with those of the expert and of a previous decision support system based on deterministic rules. Tuning networks can be used to adjust other electrical or mechanical devices, not only in medicine.
Algorithms for learning Bayesian networks (BNs) behave as a black box that takes a database as an... more Algorithms for learning Bayesian networks (BNs) behave as a black box that takes a database as an input and returns a network as the output. In contrast, OpenMarkov, our tool for probabilistic graphical models, includes the option to run the algorithms in a step-by-step fashion, presenting a ranked list of operations (such as adding, removing, or inverting links) the user can select, while allowing live edition of the BN throughout the learning process. The application offers some data preprocessing options and the possibility to use a model network to guide the learning process. This functionality in OpenMarkov can be employed to learn BNs with partial expert knowledge, to debug new algorithms, and as a pedagogical tool.
Explanation of reasoning in expert systems is necessary for debugging the knowledge base, for fac... more Explanation of reasoning in expert systems is necessary for debugging the knowledge base, for facilitating their acceptance by human users, and for using them as tutoring systems. Influence diagrams have proved to be effective tools for building decision-support systems, but explanation of their reasoning is difficult, because inference in probabilistic graphical models seems to have little relation with human thinking. The current paper describes some explanation capabilities for influence diagrams and how they have been implemented in Elvira, a public software tool.
Lung cancer is a very frequent tumor in the developed world and the leading cause of cancer death... more Lung cancer is a very frequent tumor in the developed world and the leading cause of cancer death, with non-small cell lung cancer being the most prevalent type and with most difficult prognosis. In this paper we present a decision support system built for finding the optimal selection of tests and therapy for each patient. The system basically consists of an influence diagram with super value nodes. The parameter λ, which in costeffectiveness analyses represents the amount of money that the decision maker is willing to pay to obtain a unit of effectiveness, has been included in the influence diagram, and has allowed us to find a trade-off between cost and effectiveness. Finally, given the uncertainty on the values of the parameters, we have assigned, with the expert's help, a probability distribution to each parameter of the model and have performed a probabilistic sensitivity analysis.
Encyclopedia of Cognitive Science, 2006
International Journal of Approximate Reasoning, 2010
In the original formulation of influence diagrams, each model contained exacly one utility node. ... more In the original formulation of influence diagrams, each model contained exacly one utility node. Tatman and Shachter (1990) introduced the possibility of having super-value nodes that represent the sum or the product of their parents' utility functions. However the algorithm they proposed for dealing with super-value nodes has two shortcomings: it requires dividing potentials when reversing arcs, and it tends to introduce unnecessary variables in the resulting policies. In this paper we propose a new algorithm for influence diagrams with super-value nodes that avoids these shortcomings and will be in general much more efficient than their arc-reversal algorithm.
Uploads
Papers by Francisco Javier Diez