La visualisation interactive est un outil essentiel pour l'exploration, la compréhension et l'analyse de données. L'exploration interactive efficace de jeux de données grands ou complexes présente cependant deux difficultés fondamentales. La première est visuelle et concerne les limitations de la perception et cognition humaine, ainsi que celles des écrans. La seconde est computationnelle et concerne les limitations de capacité mémoire ou de traitement des machines standards. Dans cette thèse, nous nous intéressons aux techniques de passage à l'échelle relativement à ces deux difficultés, pour plusieurs contextes d'application.Pour le passage à l'échelle visuelle, nous présentons une approche versatile de mise en évidence de sous-ensembles d'éléments par déformation spatiale appliquée aux vues multiples et une représentation abstraite et multi-/échelle de coordonnées parallèles. Sur les vues multiples, la déformation spatiale vise à remédier à la diminution de l'efficacité de la surbrillance lorsque les éléments graphiques sont de taille réduite. Sur les coordonnées parallèles, l'abstraction multi-échelle consiste à simplifier la représentation tout en permettant d'accéder interactivement au détail des données, en les pré-agrégeant à plusieurs niveaux de détail.Pour le passage à l'échelle computationnelle, nous étudions des approches de pré-calcul et de calcul à la volée sur des infrastructures distribuées permettant l'exploration de jeux de données de plus d'un milliard d'éléments en temps interactif. Nous présentons un système pour l'exploration de données multi-dimensionnelles dont les interactions et l'abstraction respectent un budget en nombre d'éléments graphiques qui, en retour, fournit une borne théorique sur les latences d'interactions dues au transfert réseau entre client et serveur. Avec le même objectif, nous comparons des stratégies de réduction de données géométrique pour la reconstruction de cartes de densité d'ensembles de points. / Interactive visualization is helpful for exploring, understanding, and analyzing data. However, increasingly large and complex data challenges the efficiency of visualization systems, both visually and computationally. The visual challenge stems from human perceptual and cognitive limitations as well as screen space limitations while the computational challenge stems from the processing and memory limitations of standard computers.In this thesis, we present techniques addressing the two scalability issues for several interactive visualization applications.To address visual scalability requirements, we present a versatile spatial-distortion approach for linked emphasis on multiple views and an abstract and multi-scale representation based on parallel coordinates. Spatial distortion aims at alleviating the weakened emphasis effect of highlighting when applied to small-sized visual elements. Multiscale abstraction simplifies the representation while providing detail on demand by pre-aggregating data at several levels of detail.To address computational scalability requirements and scale data processing to billions of items in interactive times, we use pre-computation and real-time computation on a remote distributed infrastructure. We present a system for multi-/dimensional data exploration in which the interactions and abstract representation comply with a visual item budget and in return provides a guarantee on network-related interaction latencies. With the same goal, we compared several geometric reduction strategies for the reconstruction of density maps of large-scale point sets.

Broadband spectral content is required to support ultrashort pulses. However this broadband content is subject to dispersion and hence the pulse duration of corresponding ultrashort pulses may be stretched accordingly. I used a commercially-available adaptive ultrashort pulse shaper featuring multiphoton intrapulse interference phase scan technology to characterise and compensate for the dispersion of the optical system in situ and conducted experimental and theoretical studies in various inter-linked topics relating to the light-matter interaction. Firstly, I examined the role of broadband ultrashort pulses in novel light-matter interacting systems involving optically co-trapped particle systems in which inter-particle light scattering occurs between optically-bound particles. Secondly, I delivered dispersion-compensated broadband ultrashort pulses in a dispersive microscope system to investigate the role of pulse duration in a biological light-matter interaction involving laser-induced cell membrane permeabilisation through linear and nonlinear optical absorption. Finally, I examined some of the propagation characteristics of broadband ultrashort pulse propagation using a computer-controlled spatial light modulator. The propagation characteristics of ultrashort pulses is of paramount importance for defining the light-matter interaction in systems. The ability to control ultrashort pulse propagation by using adaptive dispersion compensation enables chirp-free ultrashort pulses to be used in experiments requiring the shortest possible pulses for a specified spectral bandwidth. Ultrashort pulsed beams may be configured to provide high peak intensities over long propagation lengths, for example, using novel beam shapes such as Bessel-type beams, which has applications in biological light-matter interactions including phototransfection based on laser-induced cell membrane permeabilisation. The need for precise positioning of the beam focus on the cell membrane becomes less strenuous by virtue of the spatial properties of the Bessel beam. Dispersion compensation can be used to control the temporal properties of ultrashort pulses thus permitting, for example, a high peak intensity to be maintained along the length of a Bessel beam, thereby reducing the pulse energy required to permeabilise the cell membrane and potentially reduce damage therein.