Aktionenlernen mit Selbstorganisierenden Karten und Reinforcement Learning

Die vorliegende Arbeit beschäftigt sich mit der Entwicklung eines Funktionsapproximators und dessen Verwendung in Verfahren zum Lernen von diskreten und kontinuierlichen Aktionen: 1. Ein allgemeiner Funktionsapproximator – Locally Weighted Interpolating Growing Neural Gas (LWIGNG) – wird auf Basis eines Wachsenden Neuralen Gases (GNG) entwickelt. Die topologische Nachbarschaft in der Neuronenstruktur wird verwendet, um zwischen benachbarten Neuronen zu interpolieren und durch lokale Gewichtung die Approximation zu berechnen. Die Leistungsfähigkeit des Ansatzes, insbesondere in Hinsicht auf sich verändernde Zielfunktionen und sich verändernde Eingabeverteilungen, wird in verschiedenen Experimenten unter Beweis gestellt. 2. Zum Lernen diskreter Aktionen wird das LWIGNG-Verfahren mit Q-Learning zur Q-LWIGNG-Methode verbunden. Dafür muss der zugrunde liegende GNG-Algorithmus abgeändert werden, da die Eingabedaten beim Aktionenlernen eine bestimmte Reihenfolge haben. Q-LWIGNG erzielt sehr gute Ergebnisse beim Stabbalance- und beim Mountain-Car-Problem und gute Ergebnisse beim Acrobot-Problem. 3. Zum Lernen kontinuierlicher Aktionen wird ein REINFORCE-Algorithmus mit LWIGNG zur ReinforceGNG-Methode verbunden. Dabei wird eine Actor-Critic-Architektur eingesetzt, um aus zeitverzögerten Belohnungen zu lernen. LWIGNG approximiert sowohl die Zustands-Wertefunktion als auch die Politik, die in Form von situationsabhängigen Parametern einer Normalverteilung repräsentiert wird. ReinforceGNG wird erfolgreich zum Lernen von Bewegungen für einen simulierten 2-rädrigen Roboter eingesetzt, der einen rollenden Ball unter bestimmten Bedingungen abfangen soll.

Mixed finite-element methods for elliptic convection-dominated problems arising in semiconductor physics

In this work we develop and analyze an adaptive numerical scheme for simulating a class of macroscopic semiconductor models. At first the numerical modelling of semiconductors is reviewed in order to classify the Energy-Transport models for semiconductors that are later simulated in 2D. In this class of models the flow of charged particles, that are negatively charged electrons and so-called holes, which are quasi-particles of positive charge, as well as their energy distributions are described by a coupled system of nonlinear partial differential equations. A considerable difficulty in simulating these convection-dominated equations is posed by the nonlinear coupling as well as due to the fact that the local phenomena such as "hot electron effects" are only partially assessable through the given data. The primary variables that are used in the simulations are the particle density and the particle energy density. The user of these simulations is mostly interested in the current flow through parts of the domain boundary - the contacts. The numerical method considered here utilizes mixed finite-elements as trial functions for the discrete solution. The continuous discretization of the normal fluxes is the most important property of this discretization from the users perspective. It will be proven that under certain assumptions on the triangulation the particle density remains positive in the iterative solution algorithm. Connected to this result an a priori error estimate for the discrete solution of linear convection-diffusion equations is derived. The local charge transport phenomena will be resolved by an adaptive algorithm, which is based on a posteriori error estimators. At that stage a comparison of different estimations is performed. Additionally a method to effectively estimate the error in local quantities derived from the solution, so-called "functional outputs", is developed by transferring the dual weighted residual method to mixed finite elements. For a model problem we present how this method can deliver promising results even when standard error estimator fail completely to reduce the error in an iterative mesh refinement process.

Phylogeny and historical biogeography of the Australian Camphorosmeae (Chenopodiaceae)

Diese Studie befasst sich mit der Phylogenie und Biogeographie der australischen Camphorosmeae, die ein wichtiges Element der Flora arider Gebiete Australiens sind. Die molekularen Phylogenien wurden mit Hilfe Bayes’scher Statistik und „maximum likelihood”berechnet. Um das Alter der Gruppe und interner Linien abzuschätzen, wurden die Methoden „Nonparametric rate smoothing” und “penalized likelihood” benutzt. Morphologische Merkmale wurden nach Kriterien der Parsimonie auf den molekularen Baum aufgetragen. „Brooks parsimony analysis”, „cladistic analysis of distributions and endemism”, „dispersal-vicariance analysis”,„ancestral area analysis” und „weighted ancestral area analysis” wurden angewandt, um Abfolge und Richtungen der Ausbreitung der Gruppe in Australien zu analysieren.Von sieben getesteten Markern hatten nur die nukleären ETS und ITS genügend Variation für die phylogenetische Analyse der Camphorosmeae. Die plastidären Marker trnL-trnF spacer,trnP-psaJ spacer, rpS16 intron, rpL16 intron und trnS-trnG spacer zeigten kein ausreichendes phylogenetisches Signal. Die gefundenen phylogenetischen Hypothesen widersprechen der jetzigen Taxonomie der Gruppe. Neobassia, Threlkeldia, Osteocarpum und Enchylaena sollten den Gattungen Sclerolaena bzw. Maireana zugeordnet werden. Die kladistische Analyse der Fruchtanhängsel unterstützt die taxonomischen Ergebnisse der auf DNA basierenden Phylogenie. Allerdings hat die Behaarung, die bei anderen Gruppen der Chenopodiaceae als wichtiges taxonomisches Merkmal herangezogen wird, die Phylogenie nicht unterstützt. Vorfahren der heutigen Camphorosmeen sind im Miozän, vor ca. 8-14 Millionen Jahren, durch Fernausbreitung vermutlich aus Asien in Australien eingewandert. Anfängliche Diversifizierung fand während des späten Miozäns bis in das frühe Pliozän vor ca. 4-7 Millionen Jahren statt. Am Ende des Pliozäns existierten schon 45% - 72% der Abstammungslinien der jetzigen Camphorosmeen. Dies weist auf eine schnelle Ausbreitung hin. Das Alter stimmt mit dem Einsetzen der Aridisierung Australiens überein, und deutet darauf hin, dass die Ausbreitung der ariden Gebiete eine große Rolle bei der Diversifizierung der Gruppe spielte. Die Vorfahren der australischen Camphorosmeae scheinen die Südküste Australiens zuerst besiedeln zu haben. Dies geschah vor dem Einsetzen der Aridisierung des Kontinents. Die anschließende Ausbreitung erfolgte in verschiedene Richtungen und folgte der fortschreitenden Austrocknung im späten Tertiär und im ganzen Quartär. Durch ihre Anpassung an Trockenheit ist der Erfolg der Camphorosmeae in den ariden Gebieten zu erklären.Die Abwesenheit von klaren phylogenetischen und artspezifischen Signalen zwischen Arten der australischen Camphorosmeae ist auf das junge Alter und die schnelle Diversifizierung der Gruppe zurückzuführen, welche die Häufung von Mutationen und eine starke morphologische Differenzierung nicht zugelassen haben.

Investigation of the nuclear matter density distributions of the exotic 12 Be, 14 Be and 8 B nuclei by elastic proton scattering in inverse kinematics

The proton-nucleus elastic scattering at intermediate energies is a well-established method for the investigation of the nuclear matter distribution in stable nuclei and was recently applied also for the investigation of radioactive nuclei using the method of inverse kinematics. In the current experiment, the differential cross sections for proton elastic scattering on the isotopes $^{7,9,10,11,12,14}$Be and $^8$B were measured. The experiment was performed using the fragment separator at GSI, Darmstadt to produce the radioactive beams. The main part of the experimental setup was the time projection ionization chamber IKAR which was simultaneously used as hydrogen target and a detector for the recoil protons. Auxiliary detectors for projectile tracking and isotope identification were also installed. As results from the experiment, the absolute differential cross sections d$sigma$/d$t$ as a function of the four momentum transfer $t$ were obtained. In this work the differential cross sections for elastic p-$^{12}$Be, p-$^{14}$Be and p-$^{8}$B scattering at low $t$ ($t leq$~0.05~(GeV/c)$^2$) are presented. The measured cross sections were analyzed within the Glauber multiple-scattering theory using different density parameterizations, and the nuclear matter density distributions and radii of the investigated isotopes were determined. The analysis of the differential cross section for the isotope $^{14}$Be shows that a good description of the experimental data is obtained when density distributions consisting of separate core and halo components are used. The determined {it rms} matter radius is $3.11 pm 0.04 pm 0.13$~fm. In the case of the $^{12}$Be nucleus the results showed an extended matter distribution as well. For this nucleus a matter radius of $2.82 pm 0.03 pm 0.12$~fm was determined. An interesting result is that the free $^{12}$Be nucleus behaves differently from the core of $^{14}$Be and is much more extended than it. The data were also compared with theoretical densities calculated within the FMD and the few-body models. In the case of $^{14}$Be, the calculated cross sections describe the experimental data well while, in the case of $^{12}$Be there are discrepancies in the region of high momentum transfer. Preliminary experimental results for the isotope $^8$B are also presented. An extended matter distribution was obtained (though much more compact as compared to the neutron halos). A proton halo structure was observed for the first time with the proton elastic scattering method. The deduced matter radius is $2.60pm 0.02pm 0.26$~fm. The data were compared with microscopic calculations in the frame of the FMD model and reasonable agreement was observed. The results obtained in the present analysis are in most cases consistent with the previous experimental studies of the same isotopes with different experimental methods (total interaction and reaction cross section measurements, momentum distribution measurements). For future investigation of the structure of exotic nuclei a universal detector system EXL is being developed. It will be installed at the NESR at the future FAIR facility where higher intensity beams of radioactive ions are expected. The usage of storage ring techniques provides high luminosity and low background experimental conditions. Results from the feasibility studies of the EXL detector setup, performed at the present ESR storage ring, are presented.

Investigation of the dipole response of nickel isotopes in the presence of a high-frequency electromagnetic field

The electric dipole response of neutron-rich nickel isotopes has been investigated using the LAND setup at GSI in Darmstadt (Germany). Relativistic secondary beams of 56−57Ni and 67−72Ni at approximately 500 AMeV have been generated using projectile fragmentation of stable ions on a 4 g/cm2 Be target and subsequent separation in the magnetic dipole fields of the FRagment Separator (FRS). After reaching the LAND setup in Cave C, the radioactive ions were excited electromagnetically in the electric field of a Pb target. The decay products have been measured in inverse kinematics using various detectors. Neutron-rich 67−69Ni isotopes decay by the emission of neutrons, which are detected in the LAND detector. The present analysis concentrates on the (gamma,n) and (gamma,2n) channels in these nuclei, since the proton and three-neutron thresholds are unlikely to be reached considering the virtual photon spectrum for nickel ions at 500 AMeV. A measurement of the stable 58Ni isotope is used as a benchmark to check the accuracy of the present results with previously published data. The measured (gamma,n) and (gamma,np) channels are compared with an inclusive photoneutron measurement by Fultz and coworkers, which are consistent within the respective errors. The measured excitation energy distributions of 67−69Ni contain a large portion of the Giant Dipole Resonance (GDR) strength predicted by the Thomas-Reiche-Kuhn energy-weighted sum rule, as well as a significant amount of low-lying E1 strength, that cannot be attributed to the GDR alone. The GDR distribution parameters are calculated using well-established semi-empirical systematic models, providing the peak energies and widths. The GDR strength is extracted from the chi-square minimization of the model GDR to the measured data of the (gamma,2n) channel, thereby excluding any influence of eventual low-lying strength. The subtraction of the obtained GDR distribution from the total measured E1 strength provides the low-lying E1 strength distribution, which is attributed to the Pygmy Dipole Resonance (PDR). The extraction of the peak energy, width and strength is performed using a Gaussian function. The minimization of trial Gaussian distributions to the data does not converge towards a sharp minimum. Therefore, the results are presented by a chi-square distribution as a function of all three Gaussian parameters. Various predictions of PDR distributions exist, as well as a recent measurement of the 68Ni pygmy dipole-resonance obtained by virtual photon scattering, to which the present pygmy dipole-resonance distribution is also compared.