Refine
Has Fulltext
- yes (130) (remove)
Year of publication
- 2019 (130) (remove)
Document Type
- Doctoral Thesis (130) (remove)
Is part of the Bibliography
- yes (130)
Keywords
- Klimawandel (4)
- Spektroskopie (4)
- climate change (4)
- Biodiversität (3)
- machine learning (3)
- spectroscopy (3)
- Anden (2)
- Andes (2)
- Himalaya (2)
- Klima (2)
Institute
- Institut für Biochemie und Biologie (22)
- Institut für Geowissenschaften (18)
- Institut für Physik und Astronomie (17)
- Institut für Chemie (15)
- Hasso-Plattner-Institut für Digital Engineering GmbH (10)
- Extern (9)
- Department Linguistik (8)
- Institut für Umweltwissenschaften und Geographie (6)
- Sozialwissenschaften (5)
- Institut für Mathematik (4)
Additive Manufacturing (AM) in terms of laser powder-bed fusion (L-PBF) offers new prospects regarding the design of parts and enables therefore the production of lattice structures. These lattice structures shall be implemented in various industrial applications (e.g. gas turbines) for reasons of material savings or cooling channels. However, internal defects, residual stress, and structural deviations from the nominal geometry are unavoidable.
In this work, the structural integrity of lattice structures manufactured by means of L-PBF was non-destructively investigated on a multiscale approach.
A workflow for quantitative 3D powder analysis in terms of particle size, particle shape, particle porosity, inter-particle distance and packing density was established. Synchrotron computed tomography (CT) was used to correlate the packing density with the particle size and particle shape. It was also observed that at least about 50% of the powder porosity was released during production of the struts.
Struts are the component of lattice structures and were investigated by means of laboratory CT. The focus was on the influence of the build angle on part porosity and surface quality. The surface topography analysis was advanced by the quantitative characterisation of re-entrant surface features. This characterisation was compared with conventional surface parameters showing their complementary information, but also the need for AM specific surface parameters.
The mechanical behaviour of the lattice structure was investigated with in-situ CT under compression and successive digital volume correlation (DVC). The deformation was found to be knot-dominated, and therefore the lattice folds unit cell layer wise.
The residual stress was determined experimentally for the first time in such lattice structures. Neutron diffraction was used for the non-destructive 3D stress investigation. The principal stress directions and values were determined in dependence of the number of measured directions. While a significant uni-axial stress state was found in the strut, a more hydrostatic stress state was found in the knot. In both cases, strut and knot, seven directions were at least needed to find reliable principal stress directions.
For many years, psycholinguistic evidence has been predominantly based on findings from native speakers of Indo-European languages, primarily English, thus providing a rather limited perspective into the human language system. In recent years a growing body of experimental research has been devoted to broadening this picture, testing a wide range of speakers and languages, aiming to understanding the factors that lead to variability in linguistic performance. The present dissertation investigates sources of variability within the morphological domain, examining how and to what extent morphological processes and representations are shaped by specific properties of languages and speakers. Firstly, the present work focuses on a less explored language, Hebrew, to investigate how the unique non-concatenative morphological structure of Hebrew, namely a non-linear combination of consonantal roots and vowel patterns to form lexical entries (L-M-D + CiCeC = limed ‘teach’), affects morphological processes and representations in the Hebrew lexicon. Secondly, a less investigated population was tested: late learners of a second language. We directly compare native (L1) and non-native (L2) speakers, specifically highly proficient and immersed late learners of Hebrew. Throughout all publications, we have focused on a morphological phenomenon of inflectional classes (called binyanim; singular: binyan), comparing productive (class Piel, e.g., limed ‘teach’) and unproductive (class Paal, e.g., lamad ‘learn’) verbal inflectional classes. By using this test case, two psycholinguistic aspects of morphology were examined: (i) how morphological structure affects online recognition of complex words, using masked priming (Publications I and II) and cross-modal priming (Publication III) techniques, and (ii) what type of cues are used when extending morpho-phonological patterns to novel complex forms, a process referred to as morphological generalization, using an elicited production task (Publication IV).
The findings obtained in the four manuscripts, either published or under review, provide significant insights into the role of productivity in Hebrew morphological processing and generalization in L1 and L2 speakers. Firstly, the present L1 data revealed a close relationship between productivity of Hebrew verbal classes and recognition process, as revealed in both priming techniques. The consonantal root was accessed only in the productive class (Piel) but not the unproductive class (Paal). Another dissociation between the two classes was revealed in the cross-modal priming, yielding a semantic relatedness effect only for Paal but not Piel primes. These findings are taken to reflect that the Hebrew mental representations display a balance between stored undecomposable unstructured stems (Paal) and decomposed structured stems (Piel), in a similar manner to a typical dual-route architecture, showing that the Hebrew mental lexicon is less unique than previously claimed in psycholinguistic research. The results of the generalization study, however, indicate that there are still substantial differences between inflectional classes of Hebrew and other Indo-European classes, particularly in the type of information they rely on in generalization to novel forms. Hebrew binyan generalization relies more on cues of argument structure and less on phonological cues.
Secondly, clear L1/L2 differences were observed in the sensitivity to abstract morphological and morpho-syntactic information during complex word recognition and generalization. While L1 Hebrew speakers were sensitive to the binyan information during recognition, expressed by the contrast in root priming, L2 speakers showed similar root priming effects for both classes, but only when the primes were presented in an infinitive form. A root priming effect was not obtained for primes in a finite form. These patterns are interpreted as evidence for a reduced sensitivity of L2 speakers to morphological information, such as information about inflectional classes, and evidence for processing costs in recognition of forms carrying complex morpho-syntactic information. Reduced reliance on structural information cues was found in production of novel verbal forms, when the L2 group displayed a weaker effect of argument structure for Piel responses, in comparison to the L1 group. Given the L2 results, we suggest that morphological and morphosyntactic information remains challenging for late bilinguals, even at high proficiency levels.
This thesis investigates whether multilingual speakers’ use of grammatical constraints in an additional language (La) is affected by the native (L1) and non-native grammars (L2) of their linguistic repertoire.
Previous studies have used untimed measures of grammatical performance to show that L1 and L2 grammars affect the initial stages of La acquisition. This thesis extends this work by examining whether speakers at intermediate levels of La proficiency, who demonstrate mature untimed/offline knowledge of the target La constraints, are differentially affected by their L1 and L2 knowledge when they comprehend sentences under processing pressure. With this purpose, several groups of La German speakers were tested on word order and agreement phenomena using online/timed measures of grammatical knowledge. Participants had mirror distributions of their prior languages and they were either L1English/L2Spanish speakers or L1Spanish/L2English speakers. Crucially, in half of the phenomena the target La constraint aligned with English but not with Spanish, while in the other half it aligned with Spanish but not with English. Results show that the L1 grammar plays a major role in the use of La constraints under processing pressure, as participants displayed increased sensitivity to La constraints when they aligned with their L1, and reduced sensitivity when they did not. Further, in specific phenomena in which the L2 and La constraints aligned, increased L2 proficiency resulted in an enhanced sensitivity to the La constraint. These findings suggest that both native and non-native grammars affect how speakers use La grammatical constraints under processing pressure. However, L1 and L2 grammars differentially influence on participants’ performance: While L1 constraints seem to be reliably recruited to cope with the processing demands of real-time La use, proficiency in an L2 can enhance sensitivity to La constraints only in specific circumstances, namely when L2 and La constraints align.
Phytoplankton growth depends not only on the mean intensity but also on the dynamics of the light supply. The nonlinear light-dependency of growth is characterized by a small number of basic parameters: the compensation light intensity PARcompμ, where production and losses are balanced, the growth efficiency at sub-saturating light αµ, and the maximum growth rate at saturating light µmax. In surface mixed layers, phytoplankton may rapidly move between high light intensities and almost darkness. Because of the different frequency distribution of light and/or acclimation processes, the light-dependency of growth may differ between constant and fluctuating light. Very few studies measured growth under fluctuating light at a sufficient number of mean light intensities to estimate the parameters of the growth-irradiance relationship. Hence, the influence of light dynamics on µmax, αµ and PARcompμ are still largely unknown. By extension, accurate modelling predictions of phytoplankton development under fluctuating light exposure remain difficult to make. This PhD thesis does not intend to directly extrapolate few experimental results to aquatic systems – but rather improving the mechanistic understanding of the variation of the light-dependency of growth under light fluctuations and effects on phytoplankton development.
In Lake TaiHu and at the Three Gorges Reservoir (China), we incubated phytoplankton communities in bottles placed either at fixed depths or moved vertically through the water column to mimic vertical mixing. Phytoplankton at fixed depths received only the diurnal changes in light (defined as constant light regime), while phytoplankton received rapidly fluctuating light by superimposing the vertical light gradient on the natural sinusoidal diurnal sunlight. The vertically moved samples followed a circular movement with 20 min per revolution, replicating to some extent the full overturn of typical Langmuir cells. Growth, photosynthesis, oxygen production and respiration of communities (at Lake TaiHu) were
measured. To complete these investigations, a physiological experiment was performed in the laboratory on a toxic strain of Microcystis aeruginosa (FACBH 1322) incubated under 20 min period fluctuating light. Here, we measured electron transport rates and net oxygen production at a much higher time resolution (single minute timescale).
The present PhD thesis provides evidence for substantial effects of fluctuating light on the eco-physiology of phytoplankton. Both experiments performed under semi-natural conditions in Lake TaiHu and at the Three Gorges Reservoir gave similar results. The significant decline in community growth efficiencies αµ under fluctuating light was caused for a great share by different frequency distribution of light intensities that shortened the effective daylength for production. The remaining gap in community αµ was attributed to species-specific photoacclimation mechanisms and to light-dependent respiratory losses. In contrast, community maximal growth rates µmax were similar between incubations at constant and fluctuating light. At daily growth saturating light supply, differences in losses for biosynthesis between the two light regimes were observed. Phytoplankton experiencing constant light suffered photo-inhibition - leading to photosynthesis foregone and additional respiratory costs for photosystems repair. On the contrary, intermittent exposure to low and high light intensities prevented photo-inhibition of mixed algae but forced them to develop alternative light strategy. They better harvested and exploited surface irradiance by enhancing their photosynthesis. In the laboratory, we showed that Microcystis aeruginosa increased its oxygen consumption by dark respiration in the light few minutes only after exposure to increasing light intensities. More, we proved that within a simulated Langmuir cell, the net production at saturating light and the compensation light intensity for production at limiting light are positively related. These results are best explained by an accumulation of photosynthetic products at increasing irradiance and mobilization of these fresh resources by rapid enhancement of dark respiration for maintenance and biosynthesis at decreasing irradiance. At the daily timescale, we showed that the enhancement of photosynthesis at high irradiance for biosynthesis of species increased their maintenance respiratory costs at limiting light. Species-specific growth at saturating light µmax and compensation light intensity for growth PARcompμ of species incubated in Lake TaiHu were positively related. Because of this species-specific physiological tradeoff, species displayed different light affinities to limiting and saturating light - thereby exhibiting a gleaner-opportunist tradeoff. In Lake TaiHu, we showed that inter-specific differences in light acquisition traits (µmax and PARcompμ) allowed coexis¬tence of species on a gradient of constant
light while avoiding competitive exclusion. More interestingly we demonstrated for the first time that vertical mixing (inducing fluctuating light supply for phytoplankton) may alter or even reverse the light utilization strategies of species within couple of days. The intra-specific variation in traits under fluctuating light increased the niche space for acclimated species, precluding competitive exclusion.
Overall, this PhD thesis contributes to a better understanding of phytoplankton eco-physiology under fluctuating light supply. This work could enhance the quality of predictions of phytoplankton development under certain weather conditions or climate change scenarios.
L’extériorisation de toute communication est assujettie à un mode d’accès du locuteur aux informations véhiculées. Les constatations faites de nos données prouvent que tous les huit verbes étudiés traduisent des mécanismes d’acquisition des connaissances que nous avons appelés en emprunt à (Vogeleer, 1995 :92) « l’accès cognitif au savoir ». C’est cette valeur intrinsèque qui vaut à ces termes la dénomination de verbes médiatifs. En d’autres mots, ce sont des éléments qui explicitent des processus d’accès du locuteur au savoir. Une source du savoir qui peut être directe (la vue, le touché, l’ouïe, l’odorat…) ou indirecte (ouï-dire) et surtout inférée. Nous entendons par inférence un processus d’analyse et de mise en relation d’éléments (prémisses), lesquelles permettent de tirer une conclusion par déduction, induction ou par abduction. Et selon que lesdites prémisses tendent à être plus ou moins fiables, ces processus inférentiels impliqueront des valeurs épistémiques à des degrés divers.
Sur le plan rhétorico-syntaxique, nos analyses ont montré tous les verbes cognitifs (VC) de cette étude exigent l’occurrence d’autres constituants (actants) phrastiques qu’ils régissent. C’est grâce à cette valence verbale qu’ils gardent un pouvoir rectionnel dans les constructions asyndétiques. Ce sont donc les matrices des éléments sur lesquels ils se rapportent. Quant au cinétisme de ces verbes, il possède une fonction rhétorique et syntaxique. En effet, cet agencement particulier et souvent perturbant permet de traduire l’expression d’une figure de syntaxe à effet rhétorique : l’hyperbate. Une construction atypique qui, à travers les agencements anticonformistes, donne un sens de regressivité à l’énoncé et confère une saillance à des termes mis ce fait en exergue.
Accurate weather observations are the keystone to many quantitative applications, such as precipitation monitoring and nowcasting, hydrological modelling and forecasting, climate studies, as well as understanding precipitation-driven natural hazards (i.e. floods, landslides, debris flow). Weather radars have been an increasingly popular tool since the 1940s to provide high spatial and temporal resolution precipitation data at the mesoscale, bridging the gap between synoptic and point scale observations. Yet, many institutions still struggle to tap the potential of the large archives of reflectivity, as there is still much to understand about factors that contribute to measurement errors, one of which is calibration. Calibration represents a substantial source of uncertainty in quantitative precipitation estimation (QPE). A miscalibration of a few dBZ can easily deteriorate the accuracy of precipitation estimates by an order of magnitude. Instances where rain cells carrying torrential rains are misidentified by the radar as moderate rain could mean the difference between a timely warning and a devastating flood.
Since 2012, the Philippine Atmospheric, Geophysical, and Astronomical Services Administration (PAGASA) has been expanding the country’s ground radar network. We had a first look into the dataset from one of the longest running radars (the Subic radar) after devastating week-long torrential rains and thunderstorms in August 2012 caused by the annual southwestmonsoon and enhanced by the north-passing Typhoon Haikui. The analysis of the rainfall spatial distribution revealed the added value of radar-based QPE in comparison to interpolated rain gauge observations. However, when compared with local gauge measurements, severe miscalibration of the Subic radar was found. As a consequence, the radar-based QPE would have underestimated the rainfall amount by up to 60% if they had not been adjusted by rain gauge observations—a technique that is not only affected by other uncertainties, but which is also not feasible in other regions of the country with very sparse rain gauge coverage.
Relative calibration techniques, or the assessment of bias from the reflectivity of two radars, has been steadily gaining popularity. Previous studies have demonstrated that reflectivity observations from the Tropical Rainfall Measuring Mission (TRMM) and its successor, the Global Precipitation Measurement (GPM), are accurate enough to serve as a calibration reference for ground radars over low-to-mid-latitudes (± 35 deg for TRMM; ± 65 deg for GPM). Comparing spaceborne radars (SR) and ground radars (GR) requires cautious consideration of differences in measurement geometry and instrument specifications, as well as temporal coincidence. For this purpose, we implement a 3-D volume matching method developed by Schwaller and Morris (2011) and extended by Warren et al. (2018) to 5 years worth of observations from the Subic radar. In this method, only the volumetric intersections of the SR and GR beams are considered.
Calibration bias affects reflectivity observations homogeneously across the entire radar domain. Yet, other sources of systematic measurement errors are highly heterogeneous in space, and can either enhance or balance the bias introduced by miscalibration. In order to account for such heterogeneous errors, and thus isolate the calibration bias, we assign a quality index to each matching SR–GR volume, and thus compute the GR calibration bias as a qualityweighted average of reflectivity differences in any sample of matching SR–GR volumes. We exemplify the idea of quality-weighted averaging by using beam blockage fraction (BBF) as a quality variable. Quality-weighted averaging is able to increase the consistency of SR and GR observations by decreasing the standard deviation of the SR–GR differences, and thus increasing the precision of the bias estimates.
To extend this framework further, the SR–GR quality-weighted bias estimation is applied to the neighboring Tagaytay radar, but this time focusing on path-integrated attenuation (PIA) as the source of uncertainty. Tagaytay is a C-band radar operating at a lower wavelength and is therefore more affected by attenuation. Applying the same method used for the Subic radar, a time series of calibration bias is also established for the Tagaytay radar.
Tagaytay radar sits at a higher altitude than the Subic radar and is surrounded by a gentler terrain, so beam blockage is negligible, especially in the overlapping region. Conversely, Subic radar is largely affected by beam blockage in the overlapping region, but being an SBand radar, attenuation is considered negligible. This coincidentally independent uncertainty contributions of each radar in the region of overlap provides an ideal environment to experiment with the different scenarios of quality filtering when comparing reflectivities from the two ground radars. The standard deviation of the GR–GR differences already decreases if we consider either BBF or PIA to compute the quality index and thus the weights. However, combining them multiplicatively resulted in the largest decrease in standard deviation, suggesting that taking both factors into account increases the consistency between the matched samples.
The overlap between the two radars and the instances of the SR passing over the two radars at the same time allows for verification of the SR–GR quality-weighted bias estimation method. In this regard, the consistency between the two ground radars is analyzed before and after bias correction is applied. For cases when all three radars are coincident during a significant rainfall event, the correction of GR reflectivities with calibration bias estimates from SR overpasses dramatically improves the consistency between the two ground radars which have shown incoherent observations before correction. We also show that for cases where adequate SR coverage is unavailable, interpolating the calibration biases using a moving average can be used to correct the GR observations for any point in time to some extent. By using the interpolated biases to correct GR observations, we demonstrate that bias correction reduces the absolute value of the mean difference in most cases, and therefore improves the consistency between the two ground radars.
This thesis demonstrates that in general, taking into account systematic sources of uncertainty that are heterogeneous in space (e.g. BBF) and time (e.g. PIA) allows for a more consistent estimation of calibration bias, a homogeneous quantity. The bias still exhibits an unexpected variability in time, which hints that there are still other sources of errors that remain unexplored. Nevertheless, the increase in consistency between SR and GR as well as between the two ground radars, suggests that considering BBF and PIA in a weighted-averaging approach is a step in the right direction.
Despite the ample room for improvement, the approach that combines volume matching between radars (either SR–GR or GR–GR) and quality-weighted comparison is readily available for application or further scrutiny. As a step towards reproducibility and transparency in atmospheric science, the 3D matching procedure and the analysis workflows as well as sample data are made available in public repositories. Open-source software such as Python and wradlib are used for all radar data processing in this thesis. This approach towards open science provides both research institutions and weather services with a valuable tool that can be applied to radar calibration, from monitoring to a posteriori correction of archived data.
Hepcidin-25 (Hep-25) plays a crucial role in the control of iron homeostasis. Since the dysfunction of the hepcidin pathway leads to multiple diseases as a result of iron imbalance, hepcidin represents a potential target for the diagnosis and treatment of disorders of iron metabolism. Despite intense research in the last decade targeted at developing a selective immunoassay for iron disorder diagnosis and treatment and better understanding the ferroportin-hepcidin interaction, questions remain. The key to resolving these underlying questions is acquiring exact knowledge of the 3D structure of native Hep-25. Since it was determined that the N-terminus, which is responsible for the bioactivity of Hep-25, contains a small Cu(II)-binding site known as the ATCUN motif, it was assumed that the Hep-25-Cu(II) complex is the native, bioactive form of the hepcidin. This structure has thus far not been elucidated in detail. Owing to the lack of structural information on metal-bound Hep-25, little is known about its possible biological role in iron metabolism. Therefore, this work is focused on structurally characterizing the metal-bound Hep-25 by NMR spectroscopy and molecular dynamics simulations. For the present work, a protocol was developed to prepare and purify properly folded Hep-25 in high quantities. In order to overcome the low solubility of Hep-25 at neutral pH, we introduced the C-terminal DEDEDE solubility tag. The metal binding was investigated through a series of NMR spectroscopic experiments to identify the most affected amino acids that mediate metal coordination. Based on the obtained NMR data, a structural calculation was performed in order to generate a model structure of the Hep-25-Ni(II) complex. The DEDEDE tag was excluded from the structural calculation due to a lack of NMR restraints. The dynamic nature and fast exchange of some of the amide protons with solvent reduced the overall number of NMR restraints needed for a high-quality structure. The NMR data revealed that the 20 Cterminal Hep-25 amino acids experienced no significant conformational changes, compared to published results, as a result of a pH change from pH 3 to pH 7 and metal binding. A 3D model of the Hep-25-Ni(II) complex was constructed from NMR data recorded for the hexapeptideNi(II) complex and Hep-25-DEDEDE-Ni(II) complex in combination with the fixed conformation of 19 C-terminal amino acids. The NMR data of the Hep-25-DEDEDE-Ni(II) complex indicates that the ATCUN motif moves independently from the rest of the structure. The 3D model structure of the metal-bound Hep-25 allows for future works to elucidate hepcidin’s interaction with its receptor ferroportin and should serve as a starting point for the development of antibodies with improved selectivity.
The Government will create a motivated, merit-based, performance-driven, and professional civil service that is resistant to temptations of corruption and which provides efficient, effective and transparent public services that do not force customers to pay bribes.
— (GoIRA, 2006, p. 106)
We were in a black hole! We had an empty glass and had nothing from our side to fill it with! Thus, we accepted anything anybody offered; that is how our glass was filled; that is how we reformed our civil service.
— (Former Advisor to IARCSC, personal communication, August 2015)
How and under what conditions were the post-Taleban Civil Service Reforms of Afghanistan initiated? What were the main components of the reforms? What were their objectives and to which extent were they achieved? Who were the leading domestic and foreign actors involved in the process? Finally, what specific factors influenced the success and failure Afghanistan’s Civil Service Reforms since 2002? Guided by such fundamental questions, this research studies the wicked process of reforming the Afghan civil service in an environment where a variety of contextual, programmatic, and external factors affected the design and implementation of reforms that were entirely funded and technically assisted by the international community.
Focusing on the core components of reforms—recruitment, remuneration, and appraisal of civil servants—the qualitative study provides a detailed picture of the pre-reform civil service and its major human resources developments in the past. Following discussions on the content and purposes of the main reform programs, it will then analyze the extent of changes in policies and practices by examining the outputs and effects of these reforms.
Moreover, the study defines the specific factors that led the reforms toward a situation where most of the intended objectives remain unachieved. Doing so, it explores and explains how an overwhelming influence of international actors with conflicting interests, large-scale corruption, political interference, networks of patronage, institutionalized nepotism, culturally accepted cronyism and widespread ethnic favoritism created a very complex environment and prevented the reforms from transforming Afghanistan’s patrimonial civil service into a professional civil service, which is driven by performance and merit.
A central insight from psychological studies on human eye movements is that eye movement patterns are highly individually characteristic. They can, therefore, be used as a biometric feature, that is, subjects can be identified based on their eye movements. This thesis introduces new machine learning methods to identify subjects based on their eye movements while viewing arbitrary content. The thesis focuses on probabilistic modeling of the problem, which has yielded the best results in the most recent literature. The thesis studies the problem in three phases by proposing a purely probabilistic, probabilistic deep learning, and probabilistic deep metric learning approach. In the first phase, the thesis studies models that rely on psychological concepts about eye movements. Recent literature illustrates that individual-specific distributions of gaze patterns can be used to accurately identify individuals. In these studies, models were based on a simple parametric family of distributions. Such simple parametric models can be robustly estimated from sparse data, but have limited flexibility to capture the differences between individuals. Therefore, this thesis proposes a semiparametric model of gaze patterns that is flexible yet robust for individual identification. These patterns can be understood as domain knowledge derived from psychological literature. Fixations and saccades are examples of simple gaze patterns. The proposed semiparametric densities are drawn under a Gaussian process prior centered at a simple parametric distribution. Thus, the model will stay close to the parametric class of densities if little data is available, but it can also deviate from this class if enough data is available, increasing the flexibility of the model. The proposed method is evaluated on a large-scale dataset, showing significant improvements over the state-of-the-art. Later, the thesis replaces the model based on gaze patterns derived from psychological concepts with a deep neural network that can learn more informative and complex patterns from raw eye movement data. As previous work has shown that the distribution of these patterns across a sequence is informative, a novel statistical aggregation layer called the quantile layer is introduced. It explicitly fits the distribution of deep patterns learned directly from the raw eye movement data. The proposed deep learning approach is end-to-end learnable, such that the deep model learns to extract informative, short local patterns while the quantile layer learns to approximate the distributions of these patterns. Quantile layers are a generic approach that can converge to standard pooling layers or have a more detailed description of the features being pooled, depending on the problem. The proposed model is evaluated in a large-scale study using the eye movements of subjects viewing arbitrary visual input. The model improves upon the standard pooling layers and other statistical aggregation layers proposed in the literature. It also improves upon the state-of-the-art eye movement biometrics by a wide margin. Finally, for the model to identify any subject — not just the set of subjects it is trained on — a metric learning approach is developed. Metric learning learns a distance function over instances. The metric learning model maps the instances into a metric space, where sequences of the same individual are close, and sequences of different individuals are further apart. This thesis introduces a deep metric learning approach with distributional embeddings. The approach represents sequences as a set of continuous distributions in a metric space; to achieve this, a new loss function based on Wasserstein distances is introduced. The proposed method is evaluated on multiple domains besides eye movement biometrics. This approach outperforms the state of the art in deep metric learning in several domains while also outperforming the state of the art in eye movement biometrics.
Emotions are a central element of human experience. They occur with high frequency in everyday life and play an important role in decision making. However, currently there is no consensus among researchers on what constitutes an emotion and on how emotions should be investigated. This dissertation identifies three problems of current emotion research: the problem of ground truth, the problem of incomplete constructs and the problem of optimal representation. I argue for a focus on the detailed measurement of emotion manifestations with computer-aided methods to solve these problems. This approach is demonstrated in three research projects, which describe the development of methods specific to these problems as well as their application to concrete research questions.
The problem of ground truth describes the practice to presuppose a certain structure of emotions as the a priori ground truth. This determines the range of emotion descriptions and sets a standard for the correct assignment of these descriptions. The first project illustrates how this problem can be circumvented with a multidimensional emotion perception paradigm which stands in contrast to the emotion recognition paradigm typically employed in emotion research. This paradigm allows to calculate an objective difficulty measure and to collect subjective difficulty ratings for the perception of emotional stimuli. Moreover, it enables the use of an arbitrary number of emotion stimuli categories as compared to the commonly used six basic emotion categories. Accordingly, we collected data from 441 participants using dynamic facial expression stimuli from 40 emotion categories. Our findings suggest an increase in emotion perception difficulty with increasing actor age and provide evidence to suggest that young adults, the elderly and men underestimate their emotion perception difficulty. While these effects were predicted from the literature, we also found unexpected and novel results. In particular, the increased difficulty on the objective difficulty measure for female actors and observers stood in contrast to reported findings. Exploratory analyses revealed low relevance of person-specific variables for the prediction of emotion perception difficulty, but highlighted the importance of a general pleasure dimension for the ease of emotion perception.
The second project targets the problem of incomplete constructs which relates to vaguely defined psychological constructs on emotion with insufficient ties to tangible manifestations. The project exemplifies how a modern data collection method such as face tracking data can be used to sharpen these constructs on the example of arousal, a long-standing but fuzzy construct in emotion research. It describes how measures of distance, speed and magnitude of acceleration can be computed from face tracking data and investigates their intercorrelations. We find moderate to strong correlations among all measures of static information on one hand and all measures of dynamic information on the other. The project then investigates how self-rated arousal is tied to these measures in 401 neurotypical individuals and 19 individuals with autism. Distance to the neutral face was predictive of arousal ratings in both groups. Lower mean arousal ratings were found for the autistic group, but no difference in correlation of the measures and arousal ratings could be found between groups. Results were replicated in a high autistic traits group consisting of 41 participants. The findings suggest a qualitatively similar perception of arousal for individuals with and without autism. No correlations between valence ratings and any of the measures could be found which emphasizes the specificity of our tested measures for the construct of arousal.
The problem of optimal representation refers to the search for the best representation of emotions and the assumption that there is a one-fits-all solution. In the third project we introduce partial least squares analysis as a general method to find an optimal representation to relate two high-dimensional data sets to each other. The project demonstrates its applicability to emotion research on the question of emotion perception differences between men and women. The method was used with emotion rating data from 441 participants and face tracking data computed on 306 videos. We found quantitative as well as qualitative differences in the perception of emotional facial expressions between these groups. We showed that women’s emotional perception systematically captured more of the variance in facial expressions. Additionally, we could show that significant differences exist in the way that women and men perceive some facial expressions which could be visualized as concrete facial expression sequences. These expressions suggest differing perceptions of masked and ambiguous facial expressions between the sexes. In order to facilitate use of the developed method by the research community, a package for the statistical environment R was written. Furthermore, to call attention to the method and its usefulness for emotion research, a website was designed that allows users to explore a model of emotion ratings and facial expression data in an interactive fashion.