Full-system timing simulation is essential for computer system design, spanning platforms from mobile devices to servers. However, simulating even a few seconds of execution remains impractical due to timing simulators' extremely slow speed. Instead, designers often resort to higher-level models or significantly shorten measurement windows, which often result in inconclusive design decisions. Time-based sampling has been shown to reduce measurement requirements in timing simulation by several orders of magnitude while providing bounded error with statistical confidence. When applying time-based sampling to full-system timing simulation, a key challenge is accurate and rapid reconstruction of architectural and microarchitectural state between measurement intervals, particularly for server workloads exhibiting highly variable performance-whether within a single tier, across multiple tiers, or in consolidated environments. In this paper, we present Chronos, an end-to-end checkpointed sampling framework for full-system timing simulation. Chronos leverages a linear regression model for timekeeping in functional simulation to reduce bias in reconstructed checkpoint state and employs a parallel functional simulator to regenerate system state at speeds that scale with the number of host cores.

The present invention discloses compositions, such as pharmaceutical compositions, as well as methods, for use in the treatment and/or prevention of a disease or condition linked to oxidative stress, lipid peroxidation and ferroptosis. the pharmaceutical composition of the invention comprises an agent modulating the expression and/or activity of i) a krab-containing zinc finger protein znf354a (znf354a), ii) an mrna encoding the znf354a, and/or iii) the znf354a gene.

Microsecond time-resolved cryo-electron microscopy (cryo-em) is an emerging technique enabling observation of protein dynamics with microsecond temporal and near-atomic spatial resolution. a cryo-sample is flash melted by laser beam to provide a brief time window in which protein dynamics are initiated. as they unfold, the laser is switched off and the cryo-sample revitrifies, which arrests the proteins in their transient configurations. the temporal observation window is extended from tens to hundreds of microseconds by sealing the cryo-sample between two thin membranes created through vapor deposition of silicon dioxide onto the cryo-sample. the membranes allow imaging with near-atomic spatial resolution and eliminate the problem of preferred orientation. the membrane deposition technology may be used to deposit compounds other than silicon dioxide that induce protein conformational dynamics in the cryo-sample upon flash melting. capabilities of microsecond time-resolved cryo-em are significantly expanded and new avenues for in situ liquid cell cryo-em are opened.

The spatial dispersion of iron oxide pigments plays a critical role in determining the uniformity and functional performance of alkyd resin-based coatings. However, conventional two-dimensional (2D) characterization methods are often insufficient for resolving the three-dimensional (3D) aggregation state, particle morphology, and spatial arrangement of pigments within cured coating films. In this study, ptychographic X-ray computed tomography (PXCT) was employed to investigate the 3D microstructure of iron oxide alkyd coatings prepared under different mixing conditions. Quantitative analyses were conducted to characterize particle & cluster fractions, particle & cluster morphology, and inter-particle distance distribution. Compared with the 3 min mixing, the 1 h mixing treatment increased the volumetric fraction of dispersed particles from 51.09% to 76.04%, and reduced the cluster fraction from 48.91% to 23.96%. In addition, the mean inter-particle distance decreased from 14.02 to 12.69 μm. These results indicate that stronger and longer mixing promotes cluster disaggregation and leads to a more homogeneous distribution of pigment particles. This work provides a quantitative 3D approach for revealing pigment dispersion in alkyd coatings, and helps clarify its influence on coating structure uniformity and performance.

This article presents a comprehensive, multi-anatomical raw radio frequency (RF) ultrafast ultrasound dataset designed to facilitate advancements in beamforming, image reconstruction, and deep learning-based medical imaging. The data were acquired using a GE 11L-D linear array probe (192 elements, 7.3 MHz center frequency) interfaced with a Verasonics Vantage 256 system. The dataset comprises 20,431 total acquisitions, providing a diverse set of RF signals across varying tissue types. This includes 20,121 in vivo acquisitions collected from 16 healthy volunteers across six distinct anatomical regions—abdomen, carotids, breasts, back, upper limbs, and lower limbs—as well as 310 in vitro acquisitions from a CIRS 054GS phantom. Each acquisition consists of 105 plane waves (PWs) with a steering angle spacing of 0.32°, which enables coherent plane wave compounding (CPWC) for high-quality image reconstruction. All data collection was conducted under strict ethical safety guidelines approved by the Cantonal Commission on Ethics in Human Research (CER-VD, Switzerland). By providing a large-scale, heterogeneous dataset of raw signals, this dataset addresses the critical scarcity of open-access raw ultrasound data. It serves as a robust benchmark for evaluating traditional beamforming algorithms, validating novel image reconstruction techniques, and training data-intensive deep learning models. This resource is intended to foster reproducibility and accelerate research in medical ultrasound imaging.

In traditional psychophysics, stimuli are presented in isolation to decompose complex visual processes, with all other variables tightly controlled. This approach is rooted in classical feedforward models of vision, in which perception arises from the hierarchical integration of features and is largely determined within the first few hundred milliseconds. Such frameworks have long been used to explain visual crowding, a phenomenon in which the perception of a target deteriorates in the presence of nearby elements (flankers). Crowding has typically been interpreted as a consequence of structural limitations of the visual system, whereby limited computational resources in peripheral vision lead to compulsory integration of flankers and target features. Within this view, perception depends primarily on local stimulus properties, and temporal factors play a minimal role.

In this thesis, I demonstrate that this framework is insufficient to account for the richness and flexibility of visual perception. In the first chapter, I show that both spatial and temporal context critically shape visual crowding: target visibility depends on whether it is grouped with or segmented away from surrounding elements. When segmentation is achieved, performance improves, indicating that grouping is a fundamental perceptual strategy. Crucially, this process requires time, as either longer stimulus durations or facilitating segmentation via previewing the flankers are necessary to achieve segmentation. To account for these effects, models of vision must incorporate recurrent processing and explicit segmentation stages.

In the second chapter, I investigate how the visual system exploits preview information, showing that even previews that differ markedly from the flankers can facilitate segmentation. In the third chapter, I examine the neural dynamics underlying these effects and demonstrate that previewing flankers induces stable and long-lasting neural representations that influence the processing of subsequent stimuli. Finally, in the fourth chapter, I extend these findings to non-retinotopic interactions, showing that masks presented at distant spatial locations and long temporal delays can still impair perception, revealing that perceptual representations remain labile over extended periods.

Taken together, these results indicate that vision cannot be fully explained by strictly feedforward and retinotopic mechanisms, but instead relies on long-range, temporally extended interactions that dynamically shape perceptual outcomes.

Materials characterization is essential for linking internal structure to material properties and function. Small-angle scattering tensor tomography (SASTT) is particularly powerful for studying anisotropic and hierarchical materials because it combines nanoscale scattering sensitivity with tomographic reconstruction. Measurements acquired at different sample positions, rotations, and tilts are used to reconstruct a three-dimensional reciprocal-space map (3D RSM) for each voxel. The 3D RSM describes the local scattering response as a function of direction and scattering angle, providing information about the size, shape, spatial distribution, and orientation of scattering structures.

Despite this rich information content, established SASTT analysis often reduces orientational information to a few descriptors, such as the main scattering direction and degree of alignment. While useful for comparing dominant features across a sample, these descriptors may obscure multidirectional or complex orientation distributions. Interpretation is further complicated because the measured scattering response depends not only on scatterer orientation, but also on scatterer size and shape.

This thesis develops methods for extracting more complete orientational information from reconstructed 3D RSMs. The main approach estimates the orientation distribution function (ODF), which describes the probability of scatterers being oriented along different directions in three-dimensional space. A mathematical framework based on the spherical-harmonic convolution theorem was formulated to separate orientation-dependent information from shape- and size-dependent intensity contributions. Positivity constraints were included to ensure physically meaningful ODFs. Because this method requires prior knowledge of the scatterer form factor, an iterative blind-deconvolution approach was also investigated to estimate orientation and shape contributions directly from the data. The methods were implemented in a software package and evaluated using simulations and experimental measurements.

The methods were validated on two experimental datasets and complemented by material-specific analysis of the 3D RSMs. Gold nanorods embedded in a polymer matrix were studied as a plasmonic material for potential hydrogen-sensing applications. The recovered orientation distributions revealed how fabrication influences nanorod alignment and spatial dispersion. Animal tendons were examined to investigate hierarchical collagen organization across tendon types and species. The 3D RSMs provided information about scattering symmetry, orientational spread, characteristic length scales, and structural organization. Differences in meridional distance and orientation parameters, together with observed three-dimensional microstructural variations, suggest the presence of collagen crimp.

Overall, this thesis expands the analysis of SASTT reconstructions by enabling access to a larger fraction of the structural and orientational information contained in the 3D RSM. The results show that ODF extraction provide a more complete and interpretable description of complex anisotropic materials.

The present thesis sought to elucidate the possible underlying mechanisms to a striking phenomenon of sensing someone you cannot otherwise perceive, but you can nevertheless recognize varying degrees of its identity traits. I explored its phenomenology in the clinical context, probing iPH utility as a potential marker of Parkinson's disease progression and aimed to design quantification methods usable in clinical contexts, such as the novel HEIDI scale (Hallucination-evoked IDentity Index), proposed here. Exploring the differences in patients clinical state and neural network variability, the Limbic, Somatomotor and Default Mode Network regions might indicate self-related processing involvement, as well as potential activation of person recognition pathways. Turning to induced PH, I found a perceptual bias for the partner's identity reflecting findings from spontaneous presences commonly described as those of close others. Additionally, such iPH extends beyond the sensing of personally familiar individuals and can also be perceived as supernatural agents. Taken together, I present the possibility of iPH being derived from fundamental social perceptual and cognitive biases such as hyper-detection of social agents, to specific facilitation and advantageous processing of highly familiar conspecifics (like one's kin or in-group members). While these general social mechanisms are thought to confer evolutionary benefit for survival, they can also result in aberrant activations of those same over-represented identities following a disturbance in self-monitoring of somatomotor signals. While further work will have to elucidate the neural mechanisms involved in more detail and arbitrate between the proposed theoretical accounts for iPH emergence, a key idea has been illustrated: presences of identified Others are significant and impactful experiences across many different contexts and thus merit rigorous scientific efforts in its exploration. Taking the full breadth of their varieties into account, future work should be informed from research on fundamentally social mechanisms of our brain, as its likely primary mover.

Terahertz technology, spanning frequencies from 100 GHz to 10 THz, has enabled a wide range of applications in quantum sensing, near-field imaging, molecular spectroscopy, ranging, material characterization, and wireless communications. The current established platforms for terahertz generation and detection are photoconductive antennas, electro-optic crystals, and, more recently, spintronic devices. Continued advances in theoretical modeling, the discovery of novel terahertz materials, device design and fabrication, experimental setups, and optical and electronic technologies have steadily expanded the capabilities of terahertz systems. Despite this progress, terahertz technology remains largely reliant on bulky discrete components, in stark contrast to optics and electronics, where integrated solutions have become pervasive. While integrated implementations based on III-V materials have been demonstrated, they face fundamental challenges in fabrication scalability, optical power handling, and compatibility with mature photonic integrated platforms such as thin-film lithium niobate, silicon nitride, and silicon. In this thesis, we address these challenges by combining photonic integrated waveguides with terahertz transmission lines on thin-film lithium niobate. By simultaneously confining optical and terahertz modes, enabling phase-matched interactions, and integrating antennas on chip, we realize photonic-integrated terahertz emitters, detectors, cavities, and metasurfaces operating over four octaves, from 200 GHz to 3.5 THz. Our results establish a foundation for highly functional, chip-scale terahertz systems and open new opportunities in telecommunications, spectroscopy, quantum electrodynamics, and quantum information processing.

From reconstruction and generation to simulation and design, how machines model three-dimensional content is being steadily transformed by deep learning. Implicit Neural Representations (INRs), which describe a shape as the level set of a learned continuous field, are resolution-independent, differentiable end-to-end, and compact enough to act as learned shape parametrizations. Engineering design, however, makes sharp demands: shapes are not only observed but created, analyzed, and optimized; they form multi-part assemblies governed by explicit design parameters. Standard INRs, while expressive, fall short on these fronts, and this thesis develops structured implicit neural representations to address this gap.

We first tackle the lack of geometric regularity in pure neural representations: they fit free-form surfaces accurately but miss features prevalent in manufactured objects, e.g., cylinders or planar faces, and expose no parametric handles for direct edits. We propose a hybrid representation that unifies three primitive types: a fully implicit one for free-form parts, an analytic one with explicit parameters (radius, height, pose), and a geometry-assisted neural mixing the two. A disentangled decoder separates latent from explicit parameters, and the full shape is their union. This reconstructs complex and exactly regular parts together, supports parametric edits without re-optimization, and recovers geometric parameters from images and sketches.

Engineering objects are also inherently part-based, yet most implicit representations treat them as monolithic; part-aware alternatives model parts in isolation, output only a holistic shape, or rely on unsupervised discovery without semantic correspondence. We propose a supervised composite representation where each part has its own latent vector and pose, processed by an auto-decoder that alternates single-part and cross-part layers to share information across components. A region-based supervision scheme derives per-part signals from the global shape, without requiring watertight per-part meshes. With a single frozen decoder, we attain the best reconstruction and generation among part-based baselines, and uniquely support constrained optimization, e.g., drag reduction on a car body with fixed wheels.