Papers by Linda C. Van Der Gaag
arXiv (Cornell University), Jul 4, 2012
We present a method for learning the parameters of a Bayesian network with prior knowledge about ... more We present a method for learning the parameters of a Bayesian network with prior knowledge about the signs of influences between variables. Our method accommodates not just the standard signs, but provides for context-specific signs as well. We show how the various signs translate into order constraints on the network parameters and how isotonic regression can be used to compute order-constrained estimates from the available data. Our experimental results show that taking prior knowledge about the signs of influences into account leads to an improved fit of the true distribution, especially when only a small sample of data is available. Moreover, the computed estimates are guaranteed to be consistent with the specified signs, thereby resulting in a network that is more likely to be accepted by experts in its domain of application.
Probabilistic relational models (PRMs) extend Bayesian networks beyond propositional expressivene... more Probabilistic relational models (PRMs) extend Bayesian networks beyond propositional expressiveness by allowing the representation of multiple interacting classes. For a specific instance of sets of concrete objects per class, a ground Bayesian network is composed by replicating parts of the PRM. The interactions between the objects that are thereby induced, are not always obvious from the PRM. We demonstrate in this paper that the replicative structure of the ground network in fact constrains the space of possible probability distributions and thereby the possible patterns of intercausal interaction.
Despite their popularity, naive Bayesian classifiers are not well suited for real-world applicati... more Despite their popularity, naive Bayesian classifiers are not well suited for real-world applications involving extreme probability features. As will be demonstrated in this paper, methods used to forestall the inclusion of zero probability parameters in naive classifiers have quite counterintuitive effects. An elegant, principled solution for handling extreme probability events is available however, from coherent conditional probability theory. We will show how this theory can be integrated in standard naive Bayesian classifiers, and then present a computational framework that retains the classifiers’ efficiency in the presence of a limited number of extreme probability features.
Strahlentherapie und Onkologie, 2020
European Conference on Artificial Intelligence, Aug 18, 2014
Monitoring applications of Bayesian networks require computing a sequence of most probable explan... more Monitoring applications of Bayesian networks require computing a sequence of most probable explanations for the observations from a monitored entity at consecutive time steps. Such applications rapidly become impracticable, especially when computations are performed in real time. In this paper, we argue that a sequence of explanations can often be feasibly computed if consecutive time steps share large numbers of observed features. We show more specifically that we can conclude persistence of an explanation at an early stage of propagation. We present an algorithm that exploits this result to forestall unnecessary re-computation of explanations.
Decision-support systems often include a strategy for selecting tests in their domain of applicat... more Decision-support systems often include a strategy for selecting tests in their domain of application. Such a strategy serves to provide support for the reasoning processes in the domain. Generally a test-selection strategy is o#ered in which tests are selected sequentially. Upon building a system for the domain of oesophageal cancer, however, we felt that a sequential strategy would be an oversimplification of daily practice. To design a test-selection strategy for our system, we decided therefore to acquire knowledge about the actual strategy used by the experts in the domain and, more specifically, about the arguments underlying their strategy.
For many real-life Bayesian networks, common knowledge dictates that the output established for t... more For many real-life Bayesian networks, common knowledge dictates that the output established for the main variable of interest increases with higher values for the observable variables. We define two concepts of monotonicity to capture this type of knowledge. We say that a network is isotone in distribution if the probability distribution computed for the output variable given specific observations is stochastically dominated by any such distribution given higher-ordered observations; a network is isotone in mode if a probability distribution given higher observations has a higher mode. We show that establishing whether a network exhibits any of these properties of monotonicity is coNP PP-complete in general, and remains coNP-complete for polytrees. We present an approximate algorithm for deciding whether a network is monotone in distribution and illustrate its application to a real-life network in oncology. 1
Communications in Computer and Information Science, 2010
The use of the noisy-OR model is advocated throughout the literature as an approach to lightening... more The use of the noisy-OR model is advocated throughout the literature as an approach to lightening the task of obtaining all probabilities required for a Bayesian network. Little evidence is available, however, as to the effects of using the model on a network’s performance. In this paper, we construct a noisy-OR version of a real-life hand-built Bayesian network of moderate size, and compare the performance of the original network with that of the constructed noisy-OR version. Empirical results from using the two networks on real-life data show that the performance of the original network does not degrade by using the noisy-OR model.
Lecture Notes in Computer Science, 2003
In the medical domain, establishing a diagnosis typically amounts to reasoning about the unobserv... more In the medical domain, establishing a diagnosis typically amounts to reasoning about the unobservable truth, based upon a set of indirect observations from diagnostic tests. A diagnostic test may not be perfectly reliable, however. To avoid misdiagnosis, therefore, the reliability characteristics of the test should be taken into account upon reasoning. In this paper, we address the issue of modelling such characteristics in a probabilistic network. We argue that the standard reliability characteristics that are generally available from the literature have to be further detailed, for example by experts, before they can be included in a network. We illustrate this and related modelling issues by means of a real-life probabilistic network in oncology.
Computational Intelligence, 1993
The basic algorithms involved in reason maintenance in the standard ATMS is known to have a compu... more The basic algorithms involved in reason maintenance in the standard ATMS is known to have a computational complexity that is exponential in the worst case. Yet, also in average‐case problem solving, the ATMS often lays claim to a major part of the computational effort spent by a problem solver/ATMS system. In this paper, we argue that within the limits of the worst‐case computational complexity, it is possible to improve on the average‐case complexity of reason maintenance and query processing by eliminating computation that is of no relevance to the problem solver's performance. To this purpose, we present a set of algorithms designed to control the effort spent by the ATMS on label updating. The basic idea underlying these algorithms is that of lazy evaluation: labels are not automatically maintained on all datums but are computed only when needed (either directly or indirectly) by the problem solver. The algorithms have been implemented in the LazyRMS with which we have exper...
ECAI, 2004
Page 1. A New MDL-based Function for Feature Selection for Bayesian Network Classifiers M˘ad˘alin... more Page 1. A New MDL-based Function for Feature Selection for Bayesian Network Classifiers M˘ad˘alina M. Drugan and Linda C. van der Gaag 1 Abstract. Upon constructing a Bayesian network classifier from data, the accuracy ...
ABSTRACT This is the Proceedings of the Twenty-Third Conference on Uncertainty in Artificial Inte... more ABSTRACT This is the Proceedings of the Twenty-Third Conference on Uncertainty in Artificial Intelligence, which was held in Vancouver, British Columbia, July 19 - 22 2007.
Communications in Computer and Information Science, 2018
Recently, a heuristic was proposed for constructing Bayesian networks (BNs) from structured argum... more Recently, a heuristic was proposed for constructing Bayesian networks (BNs) from structured arguments. This heuristic helps domain experts who are accustomed to argumentation to transform their reasoning into a BN and subsequently weigh their case evidence in a probabilistic manner. While the underlying undirected graph of the BN is automatically constructed by following the heuristic, the arc directions are to be set manually by a BN engineer in consultation with the domain expert. As the knowledge elicitation involved is known to be time-consuming, it is of value to (partly) automate this step. We propose a refinement of the heuristic to this end, which specifies the directions in which arcs are to be set given specific conditions on structured arguments.
Uploads
Papers by Linda C. Van Der Gaag