Aspekte der linearen Minimax-Schätzung

Es werde das lineare Regressionsmodell y = X b + e mit den ueblichen Bedingungen betrachtet. Weiter werde angenommen, dass der Parametervektor aus einem Ellipsoid stammt. Ein optimaler Schaetzer fuer den Parametervektor ist durch den Minimax-Schaetzer gegeben. Nach der entscheidungstheoretischen Formulierung des Minimax-Schaetzproblems werden mit dem Bayesschen Ansatz, Spektralen Methoden und der Darstellung von Hoffmann und Laeuter Wege zur Bestimmung des Minimax- Schaetzers dargestellt und in Beziehung gebracht. Eine Betrachtung von Modellen mit drei Einflussgroeßen und gemeinsamen Eigenvektor fuehrt zu einer Strukturierung des Problems nach der Vielfachheit des maximalen Eigenwerts. Die Bestimmung des Minimax-Schaetzers in einem noch nicht geloesten Fall kann auf die Bestimmung einer Nullstelle einer nichtlinearen reellwertigen Funktion gefuehrt werden. Es wird ein Beispiel gefunden, in dem die Nullstelle nicht durch Radikale angegeben werden kann. Durch das Intervallschachtelungs-Prinzip oder Newton-Verfahren ist die numerische Bestimmung der Nullstelle moeglich. Durch Entwicklung einer Fixpunktgleichung aus der Darstellung von Hoffmann und Laeuter war es in einer Simulation moeglich die angestrebten Loesungen zu finden.

Inference with the lognormal distribution

Several estimators of the expectation, median and mode of the lognormal distribution are derived. They aim to be approximately unbiased, efficient, or have a minimax property in the class of estimators we introduce. The small-sample properties of these estimators are assessed by simulations and, when possible, analytically. Some of these estimators of the expectation are far more efficient than the maximum likelihood or the minimum-variance unbiased estimator, even for substantial samplesizes.

Suitability of relative humidity as an estimator of leaf wetness duration

Leaf wetness duration (LWD) is a key parameter in agricultural meteorology since it is related to epidemiology of many important crops, controlling pathogen infection and development rates. Because LWD is not widely measured, several methods have been developed to estimate it from weather data. Among the models used to estimate LWD, those that use physical principles of dew formation and dew and/or rain evaporation have shown good portability and sufficiently accurate results, but their complexity is a disadvantage for operational use. Alternatively, empirical models have been used despite their limitations. The simplest empirical models use only relative humidity data. The objective of this study was to evaluate the performance of three RH-based empirical models to estimate LWD in four regions around the world that have different climate conditions. Hourly LWD, air temperature, and relative humidity data were obtained from Ames, IA (USA), Elora, Ontario (Canada), Florence, Toscany (Italy), and Piracicaba, Sao Paulo State (Brazil). These data were used to evaluate the performance of the following empirical LWD estimation models: constant RH threshold (RH >= 90%); dew point depression (DPD); and extended RH threshold (EXT_RH). Different performance of the models was observed in the four locations. In Ames, Elora and Piracicaba, the RH >= 90% and DPD models underestimated LWD, whereas in Florence these methods overestimated LWD, especially for shorter wet periods. When the EXT_RH model was used, LWD was overestimated for all locations, with a significant increase in the errors. In general, the RH >= 90% model performed best, presenting the highest general fraction of correct estimates (F(C)), between 0.87 and 0.92, and the lowest false alarm ratio (F(AR)), between 0.02 and 0.31. The use of specific thresholds for each location improved accuracy of the RH model substantially, even when independent data were used; MAE ranged from 1.23 to 1.89 h, which is very similar to errors obtained with published physical models for LWD estimation. Based on these results, we concluded that, if calibrated locally, LWD can be estimated with acceptable accuracy by RH above a specific threshold, and that the EXT_RH method was unsuitable for estimating LWD at the locations used in this study. (C) 2007 Elsevier B.V. All rights reserved.

A Novel Nonparametric Density Estimator

We present a novel nonparametric density estimator and a new data-driven bandwidth selection method with excellent properties. The approach is in- spired by the principles of the generalized cross entropy method. The pro- posed density estimation procedure has numerous advantages over the tra- ditional kernel density estimator methods. Firstly, for the first time in the nonparametric literature, the proposed estimator allows for a genuine incor- poration of prior information in the density estimation procedure. Secondly, the approach provides the first data-driven bandwidth selection method that is guaranteed to provide a unique bandwidth for any data. Lastly, simulation examples suggest the proposed approach outperforms the current state of the art in nonparametric density estimation in terms of accuracy and reliability.

F-LQE: a Fuzzy Link Quality Estimator for wireless sensor networks

Radio Link Quality Estimation (LQE) is a fundamental building block for Wireless Sensor Networks, namely for a reliable deployment, resource management and routing. Existing LQEs (e.g. PRR, ETX, Fourbit, and LQI ) are based on a single link property, thus leading to inaccurate estimation. In this paper, we propose F-LQE, that estimates link quality on the basis of four link quality properties: packet delivery, asymmetry, stability, and channel quality. Each of these properties is defined in linguistic terms, the natural language of Fuzzy Logic. The overall quality of the link is specified as a fuzzy rule whose evaluation returns the membership of the link in the fuzzy subset of good links. Values of the membership function are smoothed using EWMA filter to improve stability. An extensive experimental analysis shows that F-LQE outperforms existing estimators.

On the Hill estimator: a comparison of methods

Extreme value theory (EVT) deals with the occurrence of extreme phenomena. The tail index is a very important parameter appearing in the estimation of the probability of rare events. Under a semiparametric framework, inference requires the choice of a number k of upper order statistics to be considered. This is the crux of the matter and there is no definite formula to do it, since a small k leads to high variance and large values of k tend to increase the bias. Several methodologies have emerged in literature, specially concerning the most popular Hill estimator (Hill, 1975). In this work we compare through simulation well-known procedures presented in Drees and Kaufmann (1998), Matthys and Beirlant (2000), Beirlant et al. (2002) and de Sousa and Michailidis (2004), with a heuristic scheme considered in Frahm et al. (2005) within the estimation of a different tail measure but with a similar context. We will see that the new method may be an interesting alternative.

On convergence of the maximum likelihood estimator in adaptive designs

Otto-von-Guericke-Universität Magdeburg, Fakultät für Mathematik, Dissertation, 2016

Imaging EEG synchronization in schizophrenia patients with S-estimator

Recently a new measure of the cooperative behavior of simultaneous time series was introduced (Carmeli et al. NeuroImage 2005). This measure called S-estimator is defined from the embedding dimension in a state space. S-estimator quantifies the amount of synchronization within a data set by comparing the actual dimensionality of the set with the expected full dimensionality of the asynchronous set. It has the advantage of being a multivariate measure over traditionally used in systems neuroscience bivariate measures of synchronization. Multivariate measures of synchronization are of particular interest for applications in the field of modern multichannel EEG research, since they easily allow mapping of local and/or regional synchronization and are compatible with other imaging techniques. We applied Sestimator to the analysis of EEG synchronization in schizophrenia patients vs. matched controls. The whole-head mapping with S-estimator revealed a specific pattern of local synchronization in schizophrenia patients. The differences in the landscape of synchronization included decreased local synchronization in the territories over occipital and midline areas and increased synchronization over temporal areas. In frontal areas, the S-estimator revealed a tendency for an asymmetry: decreased S-values over the left hemisphere were adjacent to increased values over the right hemisphere. Separate calculations showed reproducibility of this pattern across the main EEG frequency bands. The maintenance of the same synchronization landscape across EEG frequencies probably implies the structural changes in the cortical circuitry of schizophrenia patients. These changes are regionally specific and suggest that schizophrenia is a misconnectivity rather than hypo- or hyper-connectivity disorder.

Strong minimax lower bounds for learning

Minimax lower bounds for concept learning state, for example, thatfor each sample size $n$ and learning rule $g_n$, there exists a distributionof the observation $X$ and a concept $C$ to be learnt such that the expectederror of $g_n$ is at least a constant times $V/n$, where $V$ is the VC dimensionof the concept class. However, these bounds do not tell anything about therate of decrease of the error for a {\sl fixed} distribution--concept pair.\\In this paper we investigate minimax lower bounds in such a--stronger--sense.We show that for several natural $k$--parameter concept classes, includingthe class of linear halfspaces, the class of balls, the class of polyhedrawith a certain number of faces, and a class of neural networks, for any{\sl sequence} of learning rules $\{g_n\}$, there exists a fixed distributionof $X$ and a fixed concept $C$ such that the expected error is larger thana constant times $k/n$ for {\sl infinitely many n}. We also obtain suchstrong minimax lower bounds for the tail distribution of the probabilityof error, which extend the corresponding minimax lower bounds.

Minimax lower bounds for the two-armed bandit problem

We obtain minimax lower bounds on the regret for the classicaltwo--armed bandit problem. We provide a finite--sample minimax version of the well--known log $n$ asymptotic lower bound of Lai and Robbins. Also, in contrast to the log $n$ asymptotic results on the regret, we show that the minimax regret is achieved by mere random guessing under fairly mild conditions on the set of allowable configurations of the two arms. That is, we show that for {\sl every} allocation rule and for {\sl every} $n$, there is a configuration such that the regret at time $n$ is at least 1 -- $\epsilon$ times the regret of random guessing, where $\epsilon$ is any small positive constant.

Minimax regret and strategic uncertainty

This paper introduces a new solution concept, a minimax regret equilibrium, which allows for the possibility that players are uncertain about the rationality and conjectures of their opponents. We provide several applications of our concept. In particular, we consider pricesetting environments and show that optimal pricing policy follows a non-degenerate distribution. The induced price dispersion is consistent with experimental and empirical observations (Baye and Morgan (2004)).

The minimax distortion redundancy in empirical quantizer design

We obtain minimax lower and upper bounds for the expected distortionredundancy of empirically designed vector quantizers. We show that the meansquared distortion of a vector quantizer designed from $n$ i.i.d. datapoints using any design algorithm is at least $\Omega (n^{-1/2})$ awayfrom the optimal distortion for some distribution on a bounded subset of${\cal R}^d$. Together with existing upper bounds this result shows thatthe minimax distortion redundancy for empirical quantizer design, as afunction of the size of the training data, is asymptotically on the orderof $n^{1/2}$. We also derive a new upper bound for the performance of theempirically optimal quantizer.

Statistical analysis of maximum likelihood estimator images of human brain FDG PET studies

The work presented evaluates the statistical characteristics of regional bias and expected error in reconstructions of real positron emission tomography (PET) data of human brain fluoro-deoxiglucose (FDG) studies carried out by the maximum likelihood estimator (MLE) method with a robust stopping rule, and compares them with the results of filtered backprojection (FBP) reconstructions and with the method of sieves. The task of evaluating radioisotope uptake in regions-of-interest (ROIs) is investigated. An assessment of bias and variance in uptake measurements is carried out with simulated data. Then, by using three different transition matrices with different degrees of accuracy and a components of variance model for statistical analysis, it is shown that the characteristics obtained from real human FDG brain data are consistent with the results of the simulation studies.