Learning Nonlinear Visual Processing from Natural Images

The paradigm of computational vision hypothesizes that any visual function -- such as the recognition of your grandparent -- can be replicated by computational processing of the visual input. What are these computations that the brain performs? What should or could they be? Working on the latter question, this dissertation takes the statistical approach, where the suitable computations are attempted to be learned from the natural visual data itself. In particular, we empirically study the computational processing that emerges from the statistical properties of the visual world and the constraints and objectives specified for the learning process. This thesis consists of an introduction and 7 peer-reviewed publications, where the purpose of the introduction is to illustrate the area of study to a reader who is not familiar with computational vision research. In the scope of the introduction, we will briefly overview the primary challenges to visual processing, as well as recall some of the current opinions on visual processing in the early visual systems of animals. Next, we describe the methodology we have used in our research, and discuss the presented results. We have included some additional remarks, speculations and conclusions to this discussion that were not featured in the original publications. We present the following results in the publications of this thesis. First, we empirically demonstrate that luminance and contrast are strongly dependent in natural images, contradicting previous theories suggesting that luminance and contrast were processed separately in natural systems due to their independence in the visual data. Second, we show that simple cell -like receptive fields of the primary visual cortex can be learned in the nonlinear contrast domain by maximization of independence. Further, we provide first-time reports of the emergence of conjunctive (corner-detecting) and subtractive (opponent orientation) processing due to nonlinear projection pursuit with simple objective functions related to sparseness and response energy optimization. Then, we show that attempting to extract independent components of nonlinear histogram statistics of a biologically plausible representation leads to projection directions that appear to differentiate between visual contexts. Such processing might be applicable for priming, \ie the selection and tuning of later visual processing. We continue by showing that a different kind of thresholded low-frequency priming can be learned and used to make object detection faster with little loss in accuracy. Finally, we show that in a computational object detection setting, nonlinearly gain-controlled visual features of medium complexity can be acquired sequentially as images are encountered and discarded. We present two online algorithms to perform this feature selection, and propose the idea that for artificial systems, some processing mechanisms could be selectable from the environment without optimizing the mechanisms themselves. In summary, this thesis explores learning visual processing on several levels. The learning can be understood as interplay of input data, model structures, learning objectives, and estimation algorithms. The presented work adds to the growing body of evidence showing that statistical methods can be used to acquire intuitively meaningful visual processing mechanisms. The work also presents some predictions and ideas regarding biological visual processing.

Readers' resposes versus reader-response theories : An empirical study

This study examines both theoretically an empirically how well the theories of Norman Holland, David Bleich, Wolfgang Iser and Stanley Fish can explain readers' interpretations of literary texts. The theoretical analysis concentrates on their views on language from the point of view of Wittgenstein's Philosophical Investigations. This analysis shows that many of the assumptions related to language in these theories are problematic. The empirical data show that readers often form very similar interpretations. Thus the study challenges the common assumption that literary interpretations tend to be idiosyncratic. The empirical data consists of freely worded written answers to questions on three short stories. The interpretations were made by 27 Finnish university students. Some of the questions addressed issues that were discussed in large parts of the texts, some referred to issues that were mentioned only in passing or implied. The short stories were "The Witch à la Mode" by D. H. Lawrence, "Rain in the Heart" by Peter Taylor and "The Hitchhiking Game" by Milan Kundera. According to Fish, readers create both the formal features of a text and their interpretation of it according to an interpretive strategy. People who agree form an interpretive community. However, a typical answer usually contains ideas repeated by several readers as well as observations not mentioned by anyone else. Therefore it is very difficult to determine which readers belong to the same interpretive community. Moreover, readers with opposing opinions often seem to pay attention to the same textual features and even acknowledge the possibility of an opposing interpretation; therefore they do not seem to create the formal features of the text in different ways. Iser suggests that an interpretation emerges from the interaction between the text and the reader when the reader determines the implications of the text and in this way fills the "gaps" in the text. Iser believes that the text guides the reader, but as he also believes that meaning is on a level beyond words, he cannot explain how the text directs the reader. The similarity in the interpretations and the fact that the agreement is strongest when related to issues that are discussed broadly in the text do, however, support his assumption that readers are guided by the text. In Bleich's view, all interpretations have personal motives and each person has an idiosyncratic language system. The situation where a person learns a word determines the most important meaning it has for that person. In order to uncover the personal etymologies of words, Bleich asks his readers to associate freely on the basis of a text and note down all the personal memories and feelings that the reading experience evokes. Bleich's theory of the idiosyncratic language system seems to rely on a misconceived notion of the role that ostensive definitions have in language use. The readers' responses show that spontaneous associations to personal life seem to colour the readers' interpretations, but such instances are rather rare. According to Holland, an interpretation reflects the reader's identity theme. Language use is regulated by shared rules, but everyone follows the rules in his or her own way. Words mean different things to different people. The problem with this view is that if there is any basis for language use, it seems to be the shared way of following linguistic rules. Wittgenstein suggests that our understanding of words is related to the shared ways of using words and our understanding of human behaviour. This view seems to give better grounds for understanding similarity and differences in literary interpretations than the theories of Holland, Bleich, Fish and Iser.

Double-Edged Imitation : Theories and Practices of Pastiche in Literature

This study concentrates on the contested concept of pastiche in literary studies. It offers the first detailed examination of the history of the concept from its origins in the seventeenth century to the present, showing how pastiche emerged as a critical concept in interaction with the emerging conception of authorial originality and the copyright laws protecting it. One of the key results of this investigation is the contextualisation of the postmodern debate on pastiche. Even though postmodern critics often emphasise the radical novelty of pastiche, they in fact resuscitate older positions and arguments without necessarily reflecting on their historical conditions. This historical background is then used to analyse the distinction between the primarily French conception of pastiche as the imitation of style and the postmodern notion of it as the compilation of different elements. The latter s vagueness and inclusiveness detracts from its value as a critical concept. The study thus concentrates on the notion of stylistic pastiche, challenging the widespread prejudice that it is merely an indication of lack of talent. Because it is multiply based on repetition, pastiche is in fact a highly ambiguous or double-edged practice that calls into question the distinction between repetition and original, thereby undermining the received notion of individual unique authorship as a fundamental aesthetic value. Pastiche does not, however, constitute a radical upheaval of the basic assumptions on which the present institution of literature relies, since, in order to mark its difference, pastiche always refers to a source outside itself against which its difference is measured. Finally, the theoretical analysis of pastiche is applied to literary works. The pastiches written by Marcel Proust demonstrate how it can become an integral part of a writer s poetics: imitation of style is shown to provide Proust with a way of exploring the role of style as a connecting point between inner vision and reality. The pastiches of the Sherlock Holmes stories by Michael Dibdin, Nicholas Meyer and the duo Adrian Conan Doyle and John Dickson Carr illustrate the functions of pastiche within a genre detective fiction that is itself fundamentally repetitive. A.S. Byatt s Possession and D.M. Thomas s Charlotte use Victorian pastiches to investigate the conditions of literary creation in the age of postmodern suspicion of creativity and individuality. The study thus argues that the concept of pastiche has valuable insights to offer to literary criticism and theory, and that literary pastiches, though often dismissed in reviews and criticism, are a particularly interesting object of study precisely because of their characteristic ambiguity.

Noncommutative Quantum Field Theories and UV/IR Mixing

Our present-day understanding of fundamental constituents of matter and their interactions is based on the Standard Model of particle physics, which relies on quantum gauge field theories. On the other hand, the large scale dynamical behaviour of spacetime is understood via the general theory of relativity of Einstein. The merging of these two complementary aspects of nature, quantum and gravity, is one of the greatest goals of modern fundamental physics, the achievement of which would help us understand the short-distance structure of spacetime, thus shedding light on the events in the singular states of general relativity, such as black holes and the Big Bang, where our current models of nature break down. The formulation of quantum field theories in noncommutative spacetime is an attempt to realize the idea of nonlocality at short distances, which our present understanding of these different aspects of Nature suggests, and consequently to find testable hints of the underlying quantum behaviour of spacetime. The formulation of noncommutative theories encounters various unprecedented problems, which derive from their peculiar inherent nonlocality. Arguably the most serious of these is the so-called UV/IR mixing, which makes the derivation of observable predictions especially hard by causing new tedious divergencies, to which our previous well-developed renormalization methods for quantum field theories do not apply. In the thesis I review the basic mathematical concepts of noncommutative spacetime, different formulations of quantum field theories in the context, and the theoretical understanding of UV/IR mixing. In particular, I put forward new results to be published, which show that also the theory of quantum electrodynamics in noncommutative spacetime defined via Seiberg-Witten map suffers from UV/IR mixing. Finally, I review some of the most promising ways to overcome the problem. The final solution remains a challenge for the future.

Democratic Legitimacy and the Politics of Rights : A comparative analysis of the conceptual relationship between democracy and rights in contemporary political theories

Democratic Legitimacy and the Politics of Rights is a research in normative political theory, based on comparative analysis of contemporary democratic theories, classified roughly as conventional liberal, deliberative democratic and radical democratic. Its focus is on the conceptual relationship between alternative sources of democratic legitimacy: democratic inclusion and liberal rights. The relationship between rights and democracy is studied through the following questions: are rights to be seen as external constraints to democracy or as objects of democratic decision making processes? Are individual rights threatened by public participation in politics; do constitutionally protected rights limit the inclusiveness of democratic processes? Are liberal values such as individuality, autonomy and liberty; and democratic values such as equality, inclusion and popular sovereignty mutually conflictual or supportive? Analyzing feminist critique of liberal discourse, the dissertation also raises the question about Enlightenment ideals in current political debates: are the universal norms of liberal democracy inherently dependent on the rationalist grand narratives of modernity and incompatible with the ideal of diversity? Part I of the thesis introduces the sources of democratic legitimacy as presented in the alternative democratic models. Part II analyses how the relationship between rights and democracy is theorized in them. Part III contains arguments by feminists and radical democrats against the tenets of universalist liberal democratic models and responds to that critique by partly endorsing, partly rejecting it. The central argument promoted in the thesis is that while the deconstruction of modern rationalism indicates that rights are political constructions as opposed to externally given moral constraints to politics, this insight does not delegitimize the politics of universal rights as an inherent part of democratic institutions. The research indicates that democracy and universal individual rights are mutually interdependent rather than oppositional; and that democracy is more dependent on an unconditional protection of universal individual rights when it is conceived as inclusive, participatory and plural; as opposed to robust majoritarian rule. The central concepts are: liberalism, democracy, legitimacy, deliberation, inclusion, equality, diversity, conflict, public sphere, rights, individualism, universalism and contextuality. The authors discussed are e.g. John Rawls, Jürgen Habermas, Seyla Benhabib, Iris Young, Chantal Mouffe and Stephen Holmes. The research focuses on contemporary political theory, but the more classical work of John S. Mill, Benjamin Constant, Isaiah Berlin and Hannah Arendt is also included.

On gravitational fields of stars in f(R) gravity theories

Einstein's general relativity is a classical theory of gravitation: it is a postulate on the coupling between the four-dimensional, continuos spacetime and the matter fields in the universe, and it yields their dynamical evolution. It is believed that general relativity must be replaced by a quantum theory of gravity at least at extremely high energies of the early universe and at regions of strong curvature of spacetime, cf. black holes. Various attempts to quantize gravity, including conceptually new models such as string theory, have suggested that modification to general relativity might show up even at lower energy scales. On the other hand, also the late time acceleration of the expansion of the universe, known as the dark energy problem, might originate from new gravitational physics. Thus, although there has been no direct experimental evidence contradicting general relativity so far - on the contrary, it has passed a variety of observational tests - it is a question worth asking, why should the effective theory of gravity be of the exact form of general relativity? If general relativity is modified, how do the predictions of the theory change? Furthermore, how far can we go with the changes before we are face with contradictions with the experiments? Along with the changes, could there be new phenomena, which we could measure to find hints of the form of the quantum theory of gravity? This thesis is on a class of modified gravity theories called f(R) models, and in particular on the effects of changing the theory of gravity on stellar solutions. It is discussed how experimental constraints from the measurements in the Solar System restrict the form of f(R) theories. Moreover, it is shown that models, which do not differ from general relativity at the weak field scale of the Solar System, can produce very different predictions for dense stars like neutron stars. Due to the nature of f(R) models, the role of independent connection of the spacetime is emphasized throughout the thesis.

Quantum Space-Time and Noncommutative Gauge Field Theories

Arguments arising from quantum mechanics and gravitation theory as well as from string theory, indicate that the description of space-time as a continuous manifold is not adequate at very short distances. An important candidate for the description of space-time at such scales is provided by noncommutative space-time where the coordinates are promoted to noncommuting operators. Thus, the study of quantum field theory in noncommutative space-time provides an interesting interface where ordinary field theoretic tools can be used to study the properties of quantum spacetime. The three original publications in this thesis encompass various aspects in the still developing area of noncommutative quantum field theory, ranging from fundamental concepts to model building. One of the key features of noncommutative space-time is the apparent loss of Lorentz invariance that has been addressed in different ways in the literature. One recently developed approach is to eliminate the Lorentz violating effects by integrating over the parameter of noncommutativity. Fundamental properties of such theories are investigated in this thesis. Another issue addressed is model building, which is difficult in the noncommutative setting due to severe restrictions on the possible gauge symmetries imposed by the noncommutativity of the space-time. Possible ways to relieve these restrictions are investigated and applied and a noncommutative version of the Minimal Supersymmetric Standard Model is presented. While putting the results obtained in the three original publications into their proper context, the introductory part of this thesis aims to provide an overview of the present situation in the field.

Local numerical modelling of magnetoconvection and turbulence - implications for mean-field theories

During the last decades mean-field models, in which large-scale magnetic fields and differential rotation arise due to the interaction of rotation and small-scale turbulence, have been enormously successful in reproducing many of the observed features of the Sun. In the meantime, new observational techniques, most prominently helioseismology, have yielded invaluable information about the interior of the Sun. This new information, however, imposes strict conditions on mean-field models. Moreover, most of the present mean-field models depend on knowledge of the small-scale turbulent effects that give rise to the large-scale phenomena. In many mean-field models these effects are prescribed in ad hoc fashion due to the lack of this knowledge. With large enough computers it would be possible to solve the MHD equations numerically under stellar conditions. However, the problem is too large by several orders of magnitude for the present day and any foreseeable computers. In our view, a combination of mean-field modelling and local 3D calculations is a more fruitful approach. The large-scale structures are well described by global mean-field models, provided that the small-scale turbulent effects are adequately parameterized. The latter can be achieved by performing local calculations which allow a much higher spatial resolution than what can be achieved in direct global calculations. In the present dissertation three aspects of mean-field theories and models of stars are studied. Firstly, the basic assumptions of different mean-field theories are tested with calculations of isotropic turbulence and hydrodynamic, as well as magnetohydrodynamic, convection. Secondly, even if the mean-field theory is unable to give the required transport coefficients from first principles, it is in some cases possible to compute these coefficients from 3D numerical models in a parameter range that can be considered to describe the main physical effects in an adequately realistic manner. In the present study, the Reynolds stresses and turbulent heat transport, responsible for the generation of differential rotation, were determined along the mixing length relations describing convection in stellar structure models. Furthermore, the alpha-effect and magnetic pumping due to turbulent convection in the rapid rotation regime were studied. The third area of the present study is to apply the local results in mean-field models, which task we start to undertake by applying the results concerning the alpha-effect and turbulent pumping in mean-field models describing the solar dynamo.

Simulations of dimensionally reduced effective theories of high temperature QCD

Quantum chromodynamics (QCD) is the theory describing interaction between quarks and gluons. At low temperatures, quarks are confined forming hadrons, e.g. protons and neutrons. However, at extremely high temperatures the hadrons break apart and the matter transforms into plasma of individual quarks and gluons. In this theses the quark gluon plasma (QGP) phase of QCD is studied using lattice techniques in the framework of dimensionally reduced effective theories EQCD and MQCD. Two quantities are in particular interest: the pressure (or grand potential) and the quark number susceptibility. At high temperatures the pressure admits a generalised coupling constant expansion, where some coefficients are non-perturbative. We determine the first such contribution of order g^6 by performing lattice simulations in MQCD. This requires high precision lattice calculations, which we perform with different number of colors N_c to obtain N_c-dependence on the coefficient. The quark number susceptibility is studied by performing lattice simulations in EQCD. We measure both flavor singlet (diagonal) and non-singlet (off-diagonal) quark number susceptibilities. The finite chemical potential results are optained using analytic continuation. The diagonal susceptibility approaches the perturbative result above 20T_c$, but below that temperature we observe significant deviations. The results agree well with 4d lattice data down to temperatures 2T_c.

Diagnostic Tests Based on Quantile Residuals for Nonlinear Time Series Models

This thesis studies quantile residuals and uses different methodologies to develop test statistics that are applicable in evaluating linear and nonlinear time series models based on continuous distributions. Models based on mixtures of distributions are of special interest because it turns out that for those models traditional residuals, often referred to as Pearson's residuals, are not appropriate. As such models have become more and more popular in practice, especially with financial time series data there is a need for reliable diagnostic tools that can be used to evaluate them. The aim of the thesis is to show how such diagnostic tools can be obtained and used in model evaluation. The quantile residuals considered here are defined in such a way that, when the model is correctly specified and its parameters are consistently estimated, they are approximately independent with standard normal distribution. All the tests derived in the thesis are pure significance type tests and are theoretically sound in that they properly take the uncertainty caused by parameter estimation into account. -- In Chapter 2 a general framework based on the likelihood function and smooth functions of univariate quantile residuals is derived that can be used to obtain misspecification tests for various purposes. Three easy-to-use tests aimed at detecting non-normality, autocorrelation, and conditional heteroscedasticity in quantile residuals are formulated. It also turns out that these tests can be interpreted as Lagrange Multiplier or score tests so that they are asymptotically optimal against local alternatives. Chapter 3 extends the concept of quantile residuals to multivariate models. The framework of Chapter 2 is generalized and tests aimed at detecting non-normality, serial correlation, and conditional heteroscedasticity in multivariate quantile residuals are derived based on it. Score test interpretations are obtained for the serial correlation and conditional heteroscedasticity tests and in a rather restricted special case for the normality test. In Chapter 4 the tests are constructed using the empirical distribution function of quantile residuals. So-called Khmaladze s martingale transformation is applied in order to eliminate the uncertainty caused by parameter estimation. Various test statistics are considered so that critical bounds for histogram type plots as well as Quantile-Quantile and Probability-Probability type plots of quantile residuals are obtained. Chapters 2, 3, and 4 contain simulations and empirical examples which illustrate the finite sample size and power properties of the derived tests and also how the tests and related graphical tools based on residuals are applied in practice.

Rethinking Media Pluralism : A Critique of Theories and Policy Discourses

This study examines different ways in which the concept of media pluralism has been theorized and used in contemporary media policy debates. Access to a broad range of different political views and cultural expressions is often regarded as a self-evident value in both theoretical and political debates on media and democracy. Opinions on the meaning and nature of media pluralism as a theoretical, political or empirical concept, however, are many, and it can easily be adjusted to different political purposes. The study aims to analyse the ambiguities surrounding the concept of media pluralism in two ways: by deconstructing its normative roots from the perspective of democratic theory, and by examining its different uses, definitions and underlying rationalities in current European media policy debates. The first part of the study examines the values and assumptions behind the notion of media pluralism in the context of different theories of democracy and the public sphere. The second part then analyses and assesses the deployment of the concept in contemporary European policy debates on media ownership and public service media. Finally, the study critically evaluates various attempts to create empirical indicators for measuring media pluralism and discusses their normative implications and underlying rationalities. The analysis of contemporary policy debates indicates that the notion of media pluralism has been too readily reduced to an empty catchphrase or conflated with consumer choice and market competition. In this narrow technocratic logic, pluralism is often unreflectively associated with quantitative data in a way that leaves unexamined key questions about social and political values, democracy, and citizenship. The basic argument advanced in the study is that media pluralism needs to be rescued from its depoliticized uses and re-imagined more broadly as a normative value that refers to the distribution of communicative power in the public sphere. Instead of something that could simply be measured through the number of media outlets available, the study argues that media pluralism should be understood in terms of its ability to challenge inequalities in communicative power and create a more democratic public sphere.

An Empirical Comparison of Linear and Nonlinear Volatility Models for Nordic Stock Returns

This paper examines how volatility in financial markets can preferable be modeled. The examination investigates how good the models for the volatility, both linear and nonlinear, are in absorbing skewness and kurtosis. The examination is done on the Nordic stock markets, including Finland, Sweden, Norway and Denmark. Different linear and nonlinear models are applied, and the results indicates that a linear model can almost always be used for modeling the series under investigation, even though nonlinear models performs slightly better in some cases. These results indicate that the markets under study are exposed to asymmetric patterns only to a certain degree. Negative shocks generally have a more prominent effect on the markets, but these effects are not really strong. However, in terms of absorbing skewness and kurtosis, nonlinear models outperform linear ones.