aus dem Inhalt: „Offene Angaben in Umfragen bringen einen hohen Erkenntnisgewinn und tragen unvoreingenommene Meinungen zu Tage. Die Analyse dieser offenen Angaben erfordert allerdings einen hohen personellen Aufwand.”

aus dem Inhalt: „Spätestens seit Ausbruch der Corona-Pandemie sind Datenvisualisierungen und Infografiken in aller Munde oder besser gesagt «in aller Augen ». Kaum ein News-Portal, kaum eine Online-Ausgabe renommierter Zeitungen kommt ohne die fast schon obligatorische interaktive Datenvisualisierung über den Verlauf der Pandemie, die Entwicklung der Infektionszahlen oder einen Ländervergleich aus.”

aus dem Inhalt: „Many aspects of society, science, business, finance, journalism, and everyday human activity, become ever more quantified. As a result, our world is awash with data of increasing amount and complexity. Still, we must keep afloat with our innate human abilities and limitations. Visualization is one way to manage this information overload: well-designed representations replace difficult cognitive calculations with simpler perceptual interpretations.”

aus dem Inhalt: „Leonardo da Vinci is worldwide considered to be one of the greatest geniuses in human history. The famous frescoes and paintings that we can still admire today are only a tiny fraction of what constitutes the gigantic heritage of Leonardo da Vinci’s significance. Part of his heritage is an incredible amount of sketches that survived the total dismemberment thanks to vicissitudes that comprehend legacies, lootings, millionaire purchases, and thefts.”

aus dem Inhalt: „Visuelle Formen der Information und Kommunikation dominieren heute nahezu alle Lebensbereiche. Sie haben lange schon unsere Erwartungsebene erreicht – man möchte keine langen Bedienungsanleitungen lesen, sondern intuitiv über eine visuelle Führung das Gerät verstehen oder über eine Lebensmittelampel auf den ersten Blick sehen, wie «gesund» ein Produkt ist. Werden Konsumenten/-innen auf diesem Weg abgeholt, ist der erste Pluspunkt auf der Ebene der User-Experience gesammelt. Visualisierungen werden vielfach erwartet, die Fähigkeit sie zu dechiffrieren wächst – umgekehrt sinkt der Wille und teils die Fähigkeit, textbasierte Information verarbeiten zu können.”

aus dem Inhalt: „Sprache stellt unsere Verbindung zur Welt dar – dazu, wie wir die Welt verstehen und mit ihr interagieren. Digitalisierung hat dazu geführt, dass Konsumenten Tag für Tag und in unterschiedlichsten Kanälen digitale, textbasierte Sprachspuren kreieren und hinterlassen.”

aus dem Inhalt: „Datenvisualisierung wird immer wichtiger in der Kommunikation. Gerade in der Zeit der Corona-Pandemie spielt Datenvisualisierung eine zentrale Rolle, um die Lage und Dynamik zu kommunizieren. Wenn Daten erhoben und mit immer neuen Methoden analysiert werden, ist es wichtig, diese Daten addressatengerecht aufzubereiten.”

In the realm of ever increasing data volume and traffic the processing of data as a stream is key in order to build flexible and scalable data processing engines. On-line machine learning provides powerful algorithms for extracting predictive models from such data streams even if the modeled relation is time-variant in nature. The modeling of real valued data in on-line regression tasks is especially important as it connects to modeling and system identification tasks in engineering domains and bridges to other fields of machine learning like classification and reinforcement learning. Therefore, this thesis considers the problem of on-line regression on time variant data streams and introduces a new multi resolution perspective for tackling it. The proposed incremental learning system, called AS-MRA, comprises a new interpolation scheme for symmetric simplicial input segmentations, a layered approximation structure of sequential local refinement layers and a learning architecture for efficiently training the layer structure. A key concept for making these components work together in harmony is a differential parameter encoding between subsequent refinement layers which allows to decompose the target function into independent additional components represented as individual refinement layers. The whole AS-MRA approach is designed to form a smooth approximation while having its computational demands scaling linearly towards the input dimension and the overall expressiveness and therefore potential storage demands scaling exponentially towards input dimension. The AS-MRA provides no mandatory design parameters, but offers opportunities for the user to state tolerance parameters for the expected prediction performance which automatically and adaptively shape the resulting layer structure during the learning process. Other optional design parameters allow to restrict the resource consumption with respect to computational and memory demands. The effect of these parameters and the learning behavior of the AS-MRA as such are investigated with respect to various learning issues and compared to different related on-line learning approaches. The merits and contributions of the AS-MRA are experimentally shown and linked to general considerations about the relation between key concepts of the AS-MRA and fundamental results in machine learning.

In der Arbeit wird ein neuartiges Lernverfahren für Gauß-Mischverteilungen entwickelt. Es basiert auf der Technik der Markov-Chain Monte-Carlo Verfahren und ist in der Lage, in einem Zuge die Größe der Mischverteilung sowie deren Parameter zu bestimmen. Das Verfahren zeichnet sich sowohl durch eine gute Anpassung an die Trainingsdaten als auch durch eine gute Generalisierungsleistung aus. Ausgehend von einer Beschreibung der stochastischen Grundlagen und einer Analyse der Probleme, die beim Lernen von Gauß-Mischverteilungen auftreten, wird in der Abeit das neue Lernverfahren schrittweise entwickelt und seine Eigenschaften untersucht. Ein experimenteller Vergleich mit bekannten Lernverfahren für Gauß-Mischverteilungen weist die Eignung des neuen Verfahrens auch empirisch nach.

Physical processes rely on the transmission of energy and information across scales. In the last century, theoretical tools have been developed in the field of statistical physics to infer macroscopic properties starting from a microscopic description of the system. However, less attention has been devoted to the remodelling of microscopic degrees of freedom by macroscopic feedback. In recent years, ideas from non-equilibrium physics have been applied to characterise biological and artificial intelligence systems. These systems share in common their structure in discrete scales of organisation that perform specialised functions. To correctly regulate these functions, the accurate transmission of information across scales is crucial. In this thesis we study the role of macroscopic feedback in the remodelling of microscopic degrees of freedom in two paradigmatic examples, one taken from the field of biology, the self-organisation of specialisation and plasticity in a social wasp, and one from artificial intelligence, the remodelling of deep neural networks in a stochastic many-particle system. In the first part of this thesis we study how the primitively social wasp Polistes canadensis simultaneously achieves robust specialization and rapid plasticity. Combining a unique experimental strategy correlating time-resolved measurements across vastly different scales with a theoretical approach, we characterise the re-establishment of the social steady state after queen removal. We show that Polistes integrates antagonistic processes on multiple scales to distinguish between extrinsic and intrinsic perturbations and thereby achieve both robust specialisation and rapid plasticity. Furthermore, we show that the long-term stability of the social structure relies on the regulation of transcriptional noise by dynamic DNA methylation. In the second part of this thesis, we ask whether emergent collective interactions can be used to remodel deep neural networks. To this end, we study a paradigmatic stochastic manyparticle model where the dynamics are defined by the reaction rates of single particles, given by the output of distinct deep neural networks. The neural networks are in turn dynamically remodelled using deep reinforcement learning depending on the previous history of the system. In particular, we implement this model as a one dimensional stochastic lattice gas. Our results show the formation of two groups of particles that move in opposite directions, diffusively at early times and ballistically over longer time-scales, with the transition between these regimes corresponding to the time-scale of left/right symmetry breaking at the level of individual particles. Over a hierarchy of characteristic time-scales these particles develop emergent, increasingly complex interactions characterised by short-range repulsion and long-range attraction. As a result, the system asymptotically converges to a regime characterised by the presence of anti-ferromagnetic particle clusters. To conclude, we characterise the impact of memory effects and demographic disorder on the dynamics. Together, our results shed light on how non-equilibrium systems can employ macroscopic feedback to regulate the propagation of fluctuations across scales.