The goal of this thesis is to explore the debate between Bas van Fraassens constructive empiricism and scientific realism. For this purpose I discuss the existence of observable and unobservable entities, the observation/theoretical dichotomy, inference to the best explanation, the no miracles argument, pessimistic induction, and epistemic risk. I strive to show that, contrary to the view of constructive empiricism, there is no clear demarcation line between observable and unobservable entities, and that not only naked eye observation but also the instrument-based observation plays an important role in acquiring knowledge. I agree with scientific realists that there is no highest point to the human power of observation; it is open-ended for further development. Moreover, naked eye observations are not themselves beyond doubt, as sometimes even naked eye observations deceive us. In that context, theoretical explanations help us to understand the real situation. As such, there is no reason to give more credit to naked eye observations than to instrument-mediated, theory-informed observations. <p>Scientific realists are confident in their knowledge of unobservables, and reject the epistemic significance of the observable/unobservable distinction. To justify their knowledge of unobservables, they use inferences to the best explanation. Such inferences play an important role in choosing the best theory amongst a group of theories. For their part, constructive empiricists use what is called the bad lot argument to refute these inferences. I try to show that such bad lot arguments fail to succeed at undermining inferences to the best explanation. Following scientific realists, I assert that nothing is miraculous in the domain of science, and that we can be assured of the approximate truth of successful scientific theories. It is true that many contemporary scientific theories contradict previously successful scientific theories, but that does not compel us to be pessimistic about such contemporary theories. Instead of pessimism, we can have an optimistic attitude about the progress of science. Considering the different arguments of constructive empiricism and scientific realism, this thesis gives more credit to scientific realism than to constructive empiricism.

The goal of this thesis is to explore the debate between Bas van Fraassens constructive empiricism and scientific realism. For this purpose I discuss the existence of observable and unobservable entities, the observation/theoretical dichotomy, inference to the best explanation, the no miracles argument, pessimistic induction, and epistemic risk. I strive to show that, contrary to the view of constructive empiricism, there is no clear demarcation line between observable and unobservable entities, and that not only naked eye observation but also the instrument-based observation plays an important role in acquiring knowledge. I agree with scientific realists that there is no highest point to the human power of observation; it is open-ended for further development. Moreover, naked eye observations are not themselves beyond doubt, as sometimes even naked eye observations deceive us. In that context, theoretical explanations help us to understand the real situation. As such, there is no reason to give more credit to naked eye observations than to instrument-mediated, theory-informed observations. <p>Scientific realists are confident in their knowledge of unobservables, and reject the epistemic significance of the observable/unobservable distinction. To justify their knowledge of unobservables, they use inferences to the best explanation. Such inferences play an important role in choosing the best theory amongst a group of theories. For their part, constructive empiricists use what is called the bad lot argument to refute these inferences. I try to show that such bad lot arguments fail to succeed at undermining inferences to the best explanation. Following scientific realists, I assert that nothing is miraculous in the domain of science, and that we can be assured of the approximate truth of successful scientific theories. It is true that many contemporary scientific theories contradict previously successful scientific theories, but that does not compel us to be pessimistic about such contemporary theories. Instead of pessimism, we can have an optimistic attitude about the progress of science. Considering the different arguments of constructive empiricism and scientific realism, this thesis gives more credit to scientific realism than to constructive empiricism.

A structural approach for modelling a statistical problem permits to introduce a contextual theory based in previous knowledge. This approach makes the parameters completely meaningful; but, in the intermediate steps, some unobservable characteristics are introduced because of their contextual meaning. When the model is completely specified, the marginalisation into the observed variables is operated in order to obtain a tatistical model. The variables can be discrete or continuous both at the level of unobserved and at the level of observed or manifest variables. We are sometimes faced, especially in behavioural sciences, with ordinal variables; this is the case of the so-called Likert scales. Therefore, an ordinal variable could be nterpreted as a discrete version of a latent concept (the discretization model). The normality of the latent variables simplifies the study of this model into the analysis of the structure of the covariance matrix of the "ideally" measured variables, but only a sub-parameter of these matrix can be identified and consistently estimated (i.e. the matrix of polychoric correlations). Consequently, two questions rise here: Is the normality of the latent variables testable? If not, what is the aspect of this hypothesis which could be testable?. In the discretization model, we observe a loss of information with related to the information contained in the latent variables. In order to treat this situation we introduce the concept of partial observability through a (non bijective) measurable function of the latent variable. We explore this definition and verify that other models can be adjusted to this concept. The definition of partial observability permits us to distinguish between two cases depending on whether the involved function is or not depending on a Euclidean parameter. Once the partial observability is introduced, we expose a set of conditions for building a specification test at the level of latent variables. The test is built using the encompassing principle in a Bayesian framework. More precisely, the problem treated in this thesis is: How to test, in a Bayesian framework, the multivariate normality of a latent vector when only a discretized version of that vector is observed. More generally, the problem can be extended to (or re-paraphrased in): How to test, in Bayesian framework, a parametric specification on latent variables against a nonparametric alternative when only a partial observation of these latent variables is available.

Most reinforcement learning methods operate on propositional representations of the world state. Such representations are often intractably large and generalize poorly. Using a deictic representation is believed to be a viable alternative: they promise generalization while allowing the use of existing reinforcement-learning methods. Yet, there are few experiments on learning with deictic representations reported in the literature. In this paper we explore the effectiveness of two forms of deictic representation and a naive propositional representation in a simple blocks-world domain. We find, empirically, that the deictic representations actually worsen performance. We conclude with a discussion of possible causes of these results and strategies for more effective learning in domains with objects.

Precise navigation of autonomous underwater vehicles (AUV) is one of the most important challenges in the realization of distributed and cooperative algorithms for marine applications. We investigate an underwater navigation technology that enables an AUV to compute its trajectory in the presence of unknown currents in real time and simultaneously estimate the currents, using range or distance measurements from a single known location. This approach is potentially useful for small AUVs which have severe volume and power constraints. The main contribution of this work is observability analysis of the proposed navigation system using novel approaches towards uniform observability of linear time-varying (LTV) systems. We utilize the notion of limiting systems in order to address uniform observability of LTV systems. Uniform observability of an LTV system can be studied by assessing finite time observability of its limiting systems. A new definition of uniform observability over a finite interval is introduced in order to address existence of an observer whose estimation error is bounded by an exponentially decaying function on the finite interval. We also show that for a class of LTV systems, uniform observability of a lower dimensional subsystem derived from an LTV system is sufficient for uniform observability of the LTV system. / Ph. D.

<p>Starting from Maxwell’s equations, we derive a sensor model for three-axis magnetometerssuitable for localization and tracking applications. The model dependson the relative position between the sensor and the target, orientation of the targetand its magnetic signature. Both point targets and extended target modelsare provided. The models are validated on data taken from various road vehicles.The suitability of magnetometers for tracking is analyzed in terms of local observabilityand Cramér Rao lower bound as a function of the sensor positions in atwo sensor scenario. Also the signal to noise ratio is computed to determine theeffective range of the magnetometer. Results from field test data indicate excellenttracking of position and velocity of the target, as well as identification of themagnetic target model suitable for target classification.</p><p> </p>

Starting from Maxwell’s equations, we derive a sensor model for three-axis magnetometerssuitable for localization and tracking applications. The model dependson the relative position between the sensor and the target, orientation of the targetand its magnetic signature. Both point targets and extended target modelsare provided. The models are validated on data taken from various road vehicles.The suitability of magnetometers for tracking is analyzed in terms of local observabilityand Cramér Rao lower bound as a function of the sensor positions in atwo sensor scenario. Also the signal to noise ratio is computed to determine theeffective range of the magnetometer. Results from field test data indicate excellenttracking of position and velocity of the target, as well as identification of themagnetic target model suitable for target classification.

Cette thèse est consacrée à la contrôlabilité et à l’observabilité de l’équation d’évolution non autonome. Dans la première partie, nous donnons un aperçu de la théorie du contrôle ainsi que quelques résultats classiques sur le contrôle des systèmes autonomes et non autonomes. En fait, nous rappellerons les techniques de la théorie des semi-groupes, théorie de l’évolution familiale, théorie de la dualité et de l’opérateur. Dans la deuxième partie, nous sommes intéressés à étudier le problème de contrôle pour les systèmes EDP définis sur des domaines dépendant du temps. Nous développons de nouvelles techniques pour obtenir les résultats sur l’observabilité exacte des équations de l’onde et de Schrödinger 1D, puis par dualité nous établissons la contrôlabilité exacte du système adjoint. Le dernier résultat est une généralisation des tests de Hautus pour l’observabilité du système d’évolution non autonome. Notre méthode peut s’appliquer aux équations de Schrödinger et à l’équation d’onde avec des potentiels dépendant du temps. / This thesis is devoted to the controllability and observability of nonautonomous evolution equation. In the first part, we give an overview on control theory as well as some classical results on control of both autonomous and nonautonomous systems. In fact, we will recall the technique in semigroup theory, evolution familys theory, duality theory and operator theory. In the second part, we are interested to investigate the control problem for PDEs systems defined on time-dependent domains. We develope some new techniques to obtain the results on exact observability for one dimensional wave and Schrödinger equations, then by duality we establish exact controllability of adjoint system. The last result is a generalization of Hautus tests for observability of non- autonomous evolution system.Our method can apply for Schrodinger equations with time dependent potentials and to a damped wave-equation with time-dependent damping.

Orientador: Adalberto Spezamiglio / Banca: Weber Flávio Pereira / Banca: Waldemar Donizete Bastos / Resumo: Neste trabalho estudamos problemas clássicos de controle para sistemas de equações diferenciais ordinárias lineares. O ponto de partida é a equivalência entre uma equação diferencial ordinária linear de ordem n e um sistema de n equações de primeira ordem. Problemas de controlabilidade completa, observabilidade e estabi- lização de equilíbrio são considerados. / Abstract: We consider here classic problems on control for a system of ordinary diferential equations. The starting point is the equivalence between a linear equation of nth order of ordinary diferential equation and a system of n equations of first order. Problems on complete controllability, observability and stabilization of equilibrium are considered. / Mestre

Semiconductor supply chain is increasingly getting exposed to Reverse Engineering (RE) of Intellectual Property (IP). Camouflaging of gates in integrated circuits are typically employed to hide the gate functionality to prevent reverse engineering. The functionalities of these gates cannot be found by De-layering as they don’t leave any layout clues. Adversaries perform reverse engineering by replacing the camouflaged gate with the known gate and by developing custom software to determine test patterns. These test patterns are used to analyze the outputs and to conclude the functionality of the camouflaged gate. In this thesis, we show that reverse engineering of camouflaged design can be performed by exploiting the test features of commercial/publicly available Automatic Test Pattern Generation (ATPG) tools. We also propose controllability/observability and Hamming Distance sensitivity based metric to select target gates for camouflaging. Simulations on ISCAS85 benchmarks shows that the proposed techniques can increase the reverse engineering effort significantly by camouflaging small fraction of gates.