Refine
Has Fulltext
- no (27)
Year of publication
- 2004 (27) (remove)
Document Type
- Article (27) (remove)
Is part of the Bibliography
- yes (27)
Institute
- Institut für Informatik und Computational Science (27) (remove)
Untitled
(2004)
A well-known result by Stein (1956) shows that in particular situations, biased estimators can yield better parameter estimates than their generally preferred unbiased counterparts. This letter follows the same spirit, as we will stabilize the unbiased generalization error estimates by regularization and finally obtain more robust model selection criteria for learning. We trade a small bias against a larger variance reduction, which has the beneficial effect of being more precise on a single training set. We focus on the subspace information criterion (SIC), which is an unbiased estimator of the expected generalization error measured by the reproducing kernel Hilbert space norm. SIC can be applied to the kernel regression, and it was shown in earlier experiments that a small regularization of SIC has a stabilization effect. However, it remained open how to appropriately determine the degree of regularization in SIC. In this article, we derive an unbiased estimator of the expected squared error, between SIC and the expected generalization error and propose determining the degree of regularization of SIC such that the estimator of the expected squared error is minimized. Computer simulations with artificial and real data sets illustrate that the proposed method works effectively for improving the precision of SIC, especially in the high-noise-level cases. We furthermore compare the proposed method to the original SIC, the cross-validation, and an empirical Bayesian method in ridge parameter selection, with good results
Motivation: Continued development of analytical techniques based on gas chromatography and mass spectrometry now facilitates the generation of larger sets of metabolite concentration data. An important step towards the understanding of metabolite dynamics is the recognition of stable states where metabolite concentrations exhibit a simple behaviour. Such states can be characterized through the identification of significant thresholds in the concentrations. But general techniques for finding discretization thresholds in continuous data prove to be practically insufficient for detecting states due to the weak conditional dependences in concentration data. Results: We introduce a method of recognizing states in the framework of decision tree induction. It is based upon a global analysis of decision forests where stability and quality are evaluated. It leads to the detection of thresholds that are both comprehensible and robust. Applied to metabolite concentration data, this method has led to the discovery of hidden states in the corresponding variables. Some of these reflect known properties of the biological experiments, and others point to putative new states
Interest in developing a new method of man-to-machine communication-a brain-computer interface (BCI)-has grown steadily over the past few decades. BCIs create a new communication channel between the brain and an output device by bypassing conventional motor output pathways of nerves and muscles. These systems use signals recorded from the scalp, the surface of the cortex, or from inside the brain to enable users to control a variety of applications including simple word-processing software and orthotics. BCI technology could therefore provide a new communication and control option for individuals who cannot otherwise express their wishes to the outside world. Signal processing and classification methods are essential tools in the development of improved BCI technology. We organized the BCI Competition 2003 to evaluate the current state of the art of these tools. Four laboratories well versed in EEG-based BCI research provided six data sets in a documented format. We made these data sets (i.e., labeled training sets and unlabeled test sets) and their descriptions available on the Internet. The goal in the competition was to maximize the performance measure for the test labels. Researchers worldwide tested their algorithms and competed for the best classification results. This paper describes the six data sets and the results and function of the most successful algorithms
We present a general approach for representing and reasoning with sets of defaults in default logic, focusing on reasoning about preferences among sets of defaults. First, we consider how to control the application of a set of defaults so that either all apply (if possible) or none do (if not). From this, an approach to dealing with preferences among sets of default rules is developed. We begin with an ordered default theory, consisting of a standard default theory, but with possible preferences on sets of rules. This theory is transformed into a second, standard default theory wherein the preferences are respected. The approach differs from other work, in that we obtain standard default theories and do not rely on prioritized versions of default logic. In practical terms this means we can immediately use existing default logic theorem provers for an implementation. Also, we directly generate just those extensions containing the most preferred applied rules; in contrast, most previous approaches generate all extensions, then select the most preferred. In a major application of the approach, we show how semimonotonic default theories can be encoded so that reasoning can be carried out at the object level. With this, we can reason about default extensions from within the framework of a standard default logic. Hence one can encode notions such as skeptical and credulous conclusions, and can reason about such conclusions within a single extension
In this paper, we show how an approach to belief revision and belief contraction can be axiomatized by means of quantified Boolean formulas. Specifically, we consider the approach of belief change scenarios, a general framework that has been introduced for expressing different forms of belief change. The essential idea is that for a belief change scenario (K, R, C), the set of formulas K, representing the knowledge base, is modified so that the sets of formulas R and C are respectively true in, and consistent with the result. By restricting the form of a belief change scenario, one obtains specific belief change operators including belief revision, contraction, update, and merging. For both the general approach and for specific operators, we give a quantified Boolean formula such that satisfying truth assignments to the free variables correspond to belief change extensions in the original approach. Hence, we reduce the problem of determining the results of a belief change operation to that of satisfiability. This approach has several benefits. First, it furnishes an axiomatic specification of belief change with respect to belief change scenarios. This then leads to further insight into the belief change framework. Second, this axiomatization allows us to identify strict complexity bounds for the considered reasoning tasks. Third, we have implemented these different forms of belief change by means of existing solvers for quantified Boolean formulas. As well, it appears that this approach may be straightforwardly applied to other specific approaches to belief change