Expressive power of binary relevance and chain classifiers based on Bayesian Networks for multi-label classification

Bayesian network classifiers are widely used in machine learning because they intuitively represent causal relations. Multi-label classification problems require each instance to be assigned a subset of a defined set of h labels. This problem is equivalent to finding a multi-valued decision function that predicts a vector of h binary classes. In this paper we obtain the decision boundaries of two widely used Bayesian network approaches for building multi-label classifiers: Multi-label Bayesian network classifiers built using the binary relevance method and Bayesian network chain classifiers. We extend our previous single-label results to multi-label chain classifiers, and we prove that, as expected, chain classifiers provide a more expressive model than the binary relevance method.

Decision analysis for the genotype designation in low-template-DNA profiles

What genotype should the scientist specify for conducting a database search to try to find the source of a low-template-DNA (lt-DNA) trace? When the scientist answers this question, he or she makes a decision. Here, we approach this decision problem from a normative point of view by defining a decision-theoretic framework for answering this question for one locus. This framework combines the probability distribution describing the uncertainty over the trace's donor's possible genotypes with a loss function describing the scientist's preferences concerning false exclusions and false inclusions that may result from the database search. According to this approach, the scientist should choose the genotype designation that minimizes the expected loss. To illustrate the results produced by this approach, we apply it to two hypothetical cases: (1) the case of observing one peak for allele xi on a single electropherogram, and (2) the case of observing one peak for allele xi on one replicate, and a pair of peaks for alleles xi and xj, i ≠ j, on a second replicate. Given that the probabilities of allele drop-out are defined as functions of the observed peak heights, the threshold values marking the turning points when the scientist should switch from one designation to another are derived in terms of the observed peak heights. For each case, sensitivity analyses show the impact of the model's parameters on these threshold values. The results support the conclusion that the procedure should not focus on a single threshold value for making this decision for all alleles, all loci and in all laboratories.

The database search problem: a question of rational decision making

This paper applies probability and decision theory in the graphical interface of an influence diagram to study the formal requirements of rationality which justify the individualization of a person found through a database search. The decision-theoretic part of the analysis studies the parameters that a rational decision maker would use to individualize the selected person. The modeling part (in the form of an influence diagram) clarifies the relationships between this decision and the ingredients that make up the database search problem, i.e., the results of the database search and the different pairs of propositions describing whether an individual is at the source of the crime stain. These analyses evaluate the desirability associated with the decision of 'individualizing' (and 'not individualizing'). They point out that this decision is a function of (i) the probability that the individual in question is, in fact, at the source of the crime stain (i.e., the state of nature), and (ii) the decision maker's preferences among the possible consequences of the decision (i.e., the decision maker's loss function). We discuss the relevance and argumentative implications of these insights with respect to recent comments in specialized literature, which suggest points of view that are opposed to the results of our study.

Decision-theoretic reflections on processing a fingermark

A recent publication in this journal [Neumann et al., Forensic Sci. Int. 212 (2011) 32-46] presented the results of a field study that revealed the data provided by the fingermarks not processed in a forensic science laboratory. In their study, the authors were interested in the usefulness of this additional data in order to determine whether such fingermarks would have been worth submitting to the fingermark processing workflow. Taking these ideas as a starting point, this communication here places the fingermark in its context of a case brought before a court, and examines the question of processing or not processing a fingermark from a decision-theoretic point of view. The decision-theoretic framework presented provides an answer to this question in the form of a quantified expression of the expected value of information (EVOI) associated with the processed fingermark, which can then be compared with the cost of processing the mark.

An evaluation of Bayesian designs for dose-escalation studies in healthy volunteers

In this paper, Bayesian decision procedures previously proposed for dose-escalation studies in healthy volunteers are reviewed and evaluated. Modifications are made to the expression of the prior distribution in order to make the procedure simpler to implement and a more relevant criterion for optimality is introduced. The results of an extensive simulation exercise to establish the proper-ties of the procedure and to aid choice between designs are summarized, and the way in which readers can use simulation to choose a design for their own trials is described. The influence of the value of the within-subject correlation on the procedure is investigated and the use of a simple prior to reflect uncertainty about the correlation is explored. Copyright (c) 2005 John Wiley & Sons, Ltd.

Decision-theoretic designs for phase II clinical trials allowing for competing studies

This article describes an approach to optimal design of phase II clinical trials using Bayesian decision theory. The method proposed extends that suggested by Stallard (1998, Biometrics54, 279–294) in which designs were obtained to maximize a gain function including the cost of drug development and the benefit from a successful therapy. Here, the approach is extended by the consideration of other potential therapies, the development of which is competing for the same limited resources. The resulting optimal designs are shown to have frequentist properties much more similar to those traditionally used in phase II trials.

Implementation of a Bayesian design in a dose-escalation study of an experimental agent in healthy volunteers

Bayesian decision procedures have recently been developed for dose escalation in phase I clinical trials concerning pharmacokinetic responses observed in healthy volunteers. This article describes how that general methodology was extended and evaluated for implementation in a specific phase I trial of a novel compound. At the time of writing, the study is ongoing, and it will be some time before the sponsor will wish to put the results into the public domain. This article is an account of how the study was designed in a way that should prove to be safe, accurate, and efficient whatever the true nature of the compound. The study involves the observation of two pharmacokinetic endpoints relating to the plasma concentration of the compound itself and of a metabolite as well as a safety endpoint relating to the occurrence of adverse events. Construction of the design and its evaluation via simulation are presented.

A Bayesian approach for dose-escalation in a phase I clinical trial incorporating pharmacodynamic endpoints

Bayesian decision procedures have already been proposed for and implemented in Phase I dose-escalation studies in healthy volunteers. The procedures have been based on pharmacokinetic responses reflecting the concentration of the drug in blood plasma and are conducted to learn about the dose-response relationship while avoiding excessive concentrations. However, in many dose-escalation studies, pharmacodynamic endpoints such as heart rate or blood pressure are observed, and it is these that should be used to control dose-escalation. These endpoints introduce additional complexity into the modeling of the problem relative to pharmacokinetic responses. Firstly, there are responses available following placebo administrations. Secondly, the pharmacodynamic responses are related directly to measurable plasma concentrations, which in turn are related to dose. Motivated by experience of data from a real study conducted in a conventional manner, this paper presents and evaluates a Bayesian procedure devised for the simultaneous monitoring of pharmacodynamic and pharmacokinetic responses. Account is also taken of the incidence of adverse events. Following logarithmic transformations, a linear model is used to relate dose to the pharmacokinetic endpoint and a quadratic model to relate the latter to the pharmacodynamic endpoint. A logistic model is used to relate the pharmacokinetic endpoint to the risk of an adverse event.

The Bayesian Foundations of solution concepts of games

We transform a non co-operati ve game into a -Bayesian decision problem for each player where the uncertainty faced by a player is the strategy choices of the other players, the pr iors of other players on the choice of other players, the priors over priors and so on.We provide a complete characterization between the extent of knowledge about the rationality of players and their ability to successfulIy eliminate strategies which are not best responses. This paper therefore provides the informational foundations of iteratively unàominated strategies and rationalizable strategic behavior (Bernheim (1984) and Pearce (1984». Moreover, sufficient condi tions are also found for Nash equilibrium behavior. We also provide Aumann's (1985) results on correlated equilibria .

Bayesian invariant measurements of generalisation for continuous distributions

A family of measurements of generalisation is proposed for estimators of continuous distributions. In particular, they apply to neural network learning rules associated with continuous neural networks. The optimal estimators (learning rules) in this sense are Bayesian decision methods with information divergence as loss function. The Bayesian framework guarantees internal coherence of such measurements, while the information geometric loss function guarantees invariance. The theoretical solution for the optimal estimator is derived by a variational method. It is applied to the family of Gaussian distributions and the implications are discussed. This is one in a series of technical reports on this topic; it generalises the results of ¸iteZhu95:prob.discrete to continuous distributions and serve as a concrete example of a larger picture ¸iteZhu95:generalisation.

Bayesian invariant measurements of generalisation

The problem of evaluating different learning rules and other statistical estimators is analysed. A new general theory of statistical inference is developed by combining Bayesian decision theory with information geometry. It is coherent and invariant. For each sample a unique ideal estimate exists and is given by an average over the posterior. An optimal estimate within a model is given by a projection of the ideal estimate. The ideal estimate is a sufficient statistic of the posterior, so practical learning rules are functions of the ideal estimator. If the sole purpose of learning is to extract information from the data, the learning rule must also approximate the ideal estimator. This framework is applicable to both Bayesian and non-Bayesian methods, with arbitrary statistical models, and to supervised, unsupervised and reinforcement learning schemes.

Dynamic decision networks for decision-making in self-adaptive systems:a case study

Bayesian decision theory is increasingly applied to support decision-making processes under environmental variability and uncertainty. Researchers from application areas like psychology and biomedicine have applied these techniques successfully. However, in the area of software engineering and speci?cally in the area of self-adaptive systems (SASs), little progress has been made in the application of Bayesian decision theory. We believe that techniques based on Bayesian Networks (BNs) are useful for systems that dynamically adapt themselves at runtime to a changing environment, which is usually uncertain. In this paper, we discuss the case for the use of BNs, speci?cally Dynamic Decision Networks (DDNs), to support the decision-making of self-adaptive systems. We present how such a probabilistic model can be used to support the decision making in SASs and justify its applicability. We have applied our DDN-based approach to the case of an adaptive remote data mirroring system. We discuss results, implications and potential bene?ts of the DDN to enhance the development and operation of self-adaptive systems, by providing mechanisms to cope with uncertainty and automatically make the best decision.

Understanding information use in multiattribute decision making

An inference task in one in which some known set of information is used to produce an estimate about an unknown quantity. Existing theories of how humans make inferences include specialized heuristics that allow people to make these inferences in familiar environments quickly and without unnecessarily complex computation. Specialized heuristic processing may be unnecessary, however; other research suggests that the same patterns in judgment can be explained by existing patterns in encoding and retrieving memories. This dissertation compares and attempts to reconcile three alternate explanations of human inference. After justifying three hierarchical Bayesian version of existing inference models, the three models are com- pared on simulated, observed, and experimental data. The results suggest that the three models capture different patterns in human behavior but, based on posterior prediction using laboratory data, potentially ignore important determinants of the decision process.

Detection of Optimal Models in Parameter Space with Support Vector Machines

The paper proposes an approach aimed at detecting optimal model parameter combinations to achieve the most representative description of uncertainty in the model performance. A classification problem is posed to find the regions of good fitting models according to the values of a cost function. Support Vector Machine (SVM) classification in the parameter space is applied to decide if a forward model simulation is to be computed for a particular generated model. SVM is particularly designed to tackle classification problems in high-dimensional space in a non-parametric and non-linear way. SVM decision boundaries determine the regions that are subject to the largest uncertainty in the cost function classification, and, therefore, provide guidelines for further iterative exploration of the model space. The proposed approach is illustrated by a synthetic example of fluid flow through porous media, which features highly variable response due to the parameter values' combination.

Statistical learning theory for geospatial data. Case study: Aral sea

In recent years there has been an explosive growth in the development of adaptive and data driven methods. One of the efficient and data-driven approaches is based on statistical learning theory (Vapnik 1998). The theory is based on Structural Risk Minimisation (SRM) principle and has a solid statistical background. When applying SRM we are trying not only to reduce training error ? to fit the available data with a model, but also to reduce the complexity of the model and to reduce generalisation error. Many nonlinear learning procedures recently developed in neural networks and statistics can be understood and interpreted in terms of the structural risk minimisation inductive principle. A recent methodology based on SRM is called Support Vector Machines (SVM). At present SLT is still under intensive development and SVM find new areas of application (www.kernel-machines.org). SVM develop robust and non linear data models with excellent generalisation abilities that is very important both for monitoring and forecasting. SVM are extremely good when input space is high dimensional and training data set i not big enough to develop corresponding nonlinear model. Moreover, SVM use only support vectors to derive decision boundaries. It opens a way to sampling optimization, estimation of noise in data, quantification of data redundancy etc. Presentation of SVM for spatially distributed data is given in (Kanevski and Maignan 2004).