
Numerické optimalizační metody
Lukšan, Ladislav
Tato zpráva popisuje teoretické i praktické vlastnosti numerických metod pro nepodmíněnou optimalizaci. Studují se metody pro obecné i speciální optimalizační úlohy, mezi které patří minimalizace součtu čtverců, součtu absolutních hodnot, maximní hodnoty a dalších nehladkých funkcí. Kromě metod pro standardní úlohy středních rozměrů jsou studovány i metody pro rozsáhlé řídké a strukturované úlohy. Velká pozornost je věnována soustavám nelineárních rovnic.\n

 

The Equation x  Ax = b
Rohn, Jiří
We formulate conditions on A and b under which the double absolute value equation x  Ax = b possesses in each orthant a unique solution which, moreover, belongs to the interior of that orthant.
Plný tet: PDF


Globální implicitní funkce
Rohn, Jiří
Tento text pochází z roku 1973 a nebyl dosud zveřejněn. Jeho hlavním výsledkem je věta o existenci a jednoznačnosti globální implicitní funkce v Rn. Tomuto výsledku předchází řada pomocných tvrzení.
Plný tet: PDF


Some modiﬁcations of the limitedmemory variable metric optimization methods
Vlček, Jan ; Lukšan, Ladislav
Several modiﬁcations of the limitedmemory variable metric (or quasiNewton) line search methods for large scale unconstrained optimization are investigated. First the block version of the symmetric rankone (SR1) update formula is derived in a similar way as for the block BFGS update in Vlˇcek and Lukˇsan (Numerical Algorithms 2019). The block SR1 formula is then modiﬁed to obtain an update which can reduce the required number of arithmetic operations per iteration. Since it usually violates the corresponding secant conditions, this update is combined with the shifting investigated in Vlˇcek and Lukˇsan (J. Comput. Appl. Math. 2006). Moreover, a new eﬃcient way how to realize the limitedmemory shifted BFGS method is proposed. For a class of methods based on the generalized shifted economy BFGS update, global convergence is established. A numerical comparison with the standard LBFGS and BNS methods is given.
Plný tet: PDF


Introduction to statistical inference based on scalarvalued scores
Fabián, Zdeněk
In the report we maintain consistently the following point of view: Given a continuous model, there are not the observed values, which are to be used in probabilistic and statistical considerations, but their ”treated forms”,the values of the scalarvalued score function corresponding to the model. Based on this modiﬁed concept of the score function, we develop theory of score random variables, study their geometry and deﬁne their new characteristics, ﬁnite even in cases of heavytailed models. A generalization for parametric families provides a new approach to parametric point estimation.


A New Look to Information and Uncertainty of Continuous Distributions
Fabián, Zdeněk
We deﬁne information and uncertainty function of a family of continuous distributions. Their values are relative information and uncertainty of an observation from the given parametric family, their mean values are the generalized Fisher information and a new measure of variability, the score variance. In a series of examples we show why to use new concepts instead of the diﬀerential entropy.


Score correlation for skewed distributions
Fabián, Zdeněk
Based on the new concept of the scalarvalued score function of continuous distributions we introduce the score correlation coeﬃcient ”tailored” to the assumed probabilistic model and study its properties by means of simulation experiments. It appeared that the new correlation method is useful for enormously skewed distributions.


ScalarValued Score Functions and their use in Parametric Estimation
Fabián, Zdeněk
In the paper we describe and explain a new direction in probabilistic and statistical reasoning, the approach based on scalarvalued score functions of continuous random variables. We show basic properties of score functions of standard distributions, generalize the approach for parametric families and show how to use them for solutions of problems of parametric statistics.


SpatioSpectral EEG Patterns in the SourceReconstructed Space and Relation to RestingState Networks: An EEGfMRI Study
Jiříček, Stanislav ; Koudelka, V. ; Mantini, D. ; Mareček, R. ; Hlinka, Jaroslav
In this work, we present and evaluate a novel EEGfMRI integration approach combining a spatiospectral decomposition method and a reliable source localization technique. On the large 72 subjects resting state hdEEGfMRI data set we tested the stability of the proposed method in terms of both extracted spatiospectral patterns(SSPs) as well as their correspondence to the BOLD signal. We also compared the proposed method with the spatiospectral decomposition in the electrode space as well as wellknown occipital alpha correlate in terms of the explained variance of BOLD signal. We showed that the proposed method is stable in terms of extracted patterns and where they correlate with the BOLD signal. Furthermore, we show that the proposed method explains a very similar level of the BOLD signal with the other methods and that the BOLD signal in areas of typical BOLD functional networks is explained significantly more than by a chance. Nevertheless, we didn’t observe a significant relation between our sourcespace SSPs and the BOLD ICs when spatiotemporally comparing them. Finally, we report several the most stable source space EEGfMRI patterns together with their interpretation and comparison to the electrode space patterns.
