Search

Datasets play a critical role in medical imaging research, yet issues such as label quality, shortcuts, and metadata are often overlooked. This lack of attention may harm the generalizability of algorithms and, consequently, negatively impact patient outcomes. While existing medical imaging literature reviews mostly focus on machine learning (ML) methods, with only a few focusing on datasets for specific applications, these reviews remain static -- they are published once and not updated thereafter. This fails to account for emerging evidence, such as biases, shortcuts, and additional annotations that other researchers may contribute after the dataset is published. We refer to these newly discovered findings of datasets as research artifacts. To address this gap, we propose a living review that continuously tracks public datasets and their associated research artifacts across multiple medical imaging applications. Our approach includes a framework for the living review to monitor data documentation artifacts, and an SQL database to visualize the citation relationships between research artifact and dataset. Lastly, we discuss key considerations for creating medical imaging datasets, review best practices for data annotation, discuss the significance of shortcuts and demographic diversity, and emphasize the importance of managing datasets throughout their entire lifecycle. Our demo is publicly available at http://130.226.140.142., Comment: Manuscript under review

Binospec is a wide-field optical (360 to 1000 nm) spectrograph commissioned at the MMT 6.5m telescope in 2017. In direct imaging mode Binospec addresses twin 8$^\prime$ (wide) by 15$^\prime$ (slit length) fields of view. We describe an optical fiber based integral field unit (IFU) that remaps a 12$^{\prime\prime}$ x 16$^{\prime\prime}$ contiguous region onto two pseudo slits, one in each Binospec channel. The IFU, commissioned in 2023, fits into the space of a standard slit mask frame and can be deployed as desired in a mixed program of slit masks, long slits, and IFU observations. The IFU fibers are illuminated by a hexagonal lenslet array with a 0.6$^{\prime\prime}$ pitch. A separate bundle of sky fibers consists of close-packed bare fibers arranged within an 11.8$^{\prime\prime}$ circular aperture. The 640 IFU fibers and 80 sky fibers have a core diameter of 150$\mu$m, corresponding to 0.90$^{\prime\prime}$. Three gratings are available, 270lpm with R$\sim$2000, 600lpm with R$\sim$5300, and 1000 lpm with R$\sim$6000., Comment: Accepted for publication in PASP, 12 pages, 11 figures, 1 table

Given a hypergraph $H=(V,E)$, define for every edge $e\in E$ a linear expression with arguments corresponding with the vertices. Next, let the polynomial $p_H$ be the product of such linear expressions for all edges. Our main goal was to find a relationship between the Alon-Tarsi number of $p_H$ and the edge density of $H$. We prove that $AT(p_H)=\lceil ed(H)\rceil+1$ if all the coefficients in $p_H$ are equal to $1$. Our main result is that, no matter what those coefficients are, they can be permuted within the edges so that for the resulting polynomial $p_H^\prime$, $AT(p_H^\prime)\leq 2\lceil ed(H)\rceil+1$ holds. We conjecture that, in fact, permuting the coefficients is not necessary. If this were true, then in particular a significant generalization of the famous 1-2-3 Conjecture would follow.

Single photons are resources needed for developing quantum networks QN. They distribute quantum information services across commercial optical fiber links and are key ingredient in developing quantum repeaters architectures. Currently, the most robust quantum light sources are Quantum Dots made of III-V materials. They emit highly indistinguishable photons on-demand and with high efficiency. Established devices work at near-infrared wavelengths (NIR) and further research is needed to develop devices working in telecommunication wavelengths O- and S-bands. In this contribution, we propose and model a broadband optical antenna working in O-band. It exhibits high extraction efficiencies with small Purcell enhancement around 2. We also examine far field emission from these structures, ensuring Gaussian mode profile is observed.

Scintillators, which convert high-energy radiation into detectable photons, play a crucial role in medical imaging and security applications. The enhancement of scintillator performance through nanophotonics and nanoplasmonics, specifically using the Purcell effect, has shown promise but has so far been limited to ultrathin scintillator films due to the localized nature of this effect. In this study, we present a method to extend nanoplasmonic scintillators to the bulk regime. By integrating 100-nm-size plasmonic spheroid and cuboid nanoparticles with perovskite scintillator nanocrystals, we enable nanoplasmonic scintillators to function effectively within bulk-scale devices. We experimentally demonstrate power and decay rate enhancements of up to (3.20 $\pm$ 0.20) and (4.20 $\pm$ 0.31) fold for plasmonic spheroid and cuboid nanoparticles, respectively, in a 5-mm thick CsPbBr$_3$ nanocrystal-polymer scintillator at RT. Theoretical modeling further predicts similar enhancements of up to (2.63 $\pm$ 0.79) and (5.62 $\pm$ 1.71) fold for the same nanoparticle shapes and dimensions. These findings provide a viable pathway for using nanoplasmonics to enhance bulk scintillator devices, advancing radiation detection technology., Comment: 42 pages with 14 figures, split between main text and supporting information. This is a full-length research article

Salt precipitation triggered by the evaporation of formation brine into injected supercritical CO2 can cause injectivity and containment issues in near-wellbore regions. Predicting the distribution of precipitated salts and their impact on near-wellbore properties remains challenging. This study investigates the influence of surface wettability on CO2-induced halite precipitation and growth within hydrophilic and hydrophobic microfluidic chips designed to mimic rock-structure porous geometries. A series of high-pressure brine-CO2 flow experiments, direct microscopic observations, and detailed image processing were conducted to explore how substrate wettability affects salt precipitation. The experiments show that wettability markedly controls residual brine relocation, film flow movement, solute consumption, and salt formation. Tracking halite precipitation dynamics revealed distinct crystal formation signatures: hydrophilic chips exhibited irregular, larger aggregation patches, while the hydrophobic network showed more numerous, smaller, and limited aggregations. Large individual crystals were observed in both chips, with a notable dominance in the hydrophobic one. Crystallization dynamics varied, with nucleation and growth occurring earlier, progressing faster, and forming bulkier aggregates in the hydrophilic chip. Despite these differences, the three identified temporal stages of brine evaporation and halite surface coverage were comparable. Spatial analysis along the chips indicated that crystal aggregation properties, such as size and distribution, were position-dependent, with hydrophilic chips exhibiting greater probabilistic variability. These observations underscore the impact of surface wettability on salt precipitation through brine accessibility and capillarity, with implications for mitigating and remediating salt issues in saline aquifers.

The Hartmann test is a method used to measure the wavefront error in a focal optical system, wherein a mask with a pattern of small holes is placed at the system's aperture stop. By taking an image at a defocused plane, the differences between the ideal and real positions of the reimaged holes (called the transverse ray aberrations) can be measured, which can then be used to estimate the wavefront error. However, the Hartmann test is usually used with an on-axis field. In this paper, we present a wavefront sensing method which generalizes the classical Hartmann test for off-axis field angles and arbitrary reference wavefronts. Our method involves taking images at two defocused planes, and then using the real reimaged hole positions on both planes to estimate the trajectories of rays from the system's exit pupil, at which the reference wavefront is situated. We then propagate the rays forward from the reference wavefront to one of the two defocused planes, in order to find the ideal reimaged hole positions, from which we can compute transverse ray aberrations. We derive and solve a pair of nonlinear partial differential equations relating transverse ray aberrations to wavefront error, using Zernike decomposition and nonlinear least squares. Our method has been verified on simulated data from the 7-lens f/2.25 red camera system of the GMT-Consortium Large Earth Finder (G-CLEF), a high resolution optical echelle spectrograph which will be a first light instrument for the Giant Magellan Telescope (GMT)., Comment: 18 pages, 23 figures, SPIE Astronomical Telescopes + Instrumentation 2024

Language models exhibit scaling laws, whereby increasing model and dataset size yields predictable decreases in negative log likelihood, unlocking a dazzling array of capabilities. This phenomenon spurs many companies to train ever larger models in pursuit of ever improved performance. Yet, these models are vulnerable to adversarial inputs such as ``jailbreaks'' and prompt injections that induce models to perform undesired behaviors, posing a growing risk as models become more capable. Prior work indicates that computer vision models become more robust with model and data scaling, raising the question: does language model robustness also improve with scale? We study this question empirically in the classification setting, finding that without explicit defense training, larger models tend to be modestly more robust on most tasks, though the effect is not reliable. Even with the advantage conferred by scale, undefended models remain easy to attack in absolute terms, and we thus turn our attention to explicitly training models for adversarial robustness, which we show to be a much more compute-efficient defense than scaling model size alone. In this setting, we also observe that adversarially trained larger models generalize faster and better to modified attacks not seen during training when compared with smaller models. Finally, we analyze the offense/defense balance of increasing compute, finding parity in some settings and an advantage for offense in others, suggesting that adversarial training alone is not sufficient to solve robustness, even at greater model scales., Comment: 36 pages; updated to include new results and analysis

Pre-fault tolerant quantum computers have already demonstrated the ability to estimate observable values accurately, at a scale beyond brute-force classical computation. This has been enabled by error mitigation techniques that often rely on a representative model on the device noise. However, learning and maintaining these models is complicated by fluctuations in the noise over unpredictable time scales, for instance, arising from resonant interactions between superconducting qubits and defect two-level systems (TLS). Such interactions affect the stability and uniformity of device performance as a whole, but also affect the noise model accuracy, leading to incorrect observable estimation. Here, we experimentally demonstrate that tuning of the qubit-TLS interactions helps reduce noise instabilities and consequently enables more reliable error-mitigation performance. These experiments provide a controlled platform for studying the performance of error mitigation in the presence of quasi-static noise. We anticipate that the capabilities introduced here will be crucial for the exploration of quantum applications on solid-state processors at non-trivial scales., Comment: 8 pages, 4 figures (13 pages, 8 figures for supplementary material), reference numbering has been fixed at v2

The stabilization of liquid biological products is a complex task that depends on the chemical composition of both the active ingredient and any excipients in solution. Frequently, a large number of unique excipients are required to stabilize biologics, though it is not well-known how these excipients interact with one another. To probe these excipient-excipient interactions, we performed molecular dynamics simulations of arginine -- a widely used excipient with unique properties -- in solution either alone or with equimolar lysine or glutamate. We studied the effects of these mixtures on a hydrophobic polymer model to isolate excipient mechanisms on hydrophobic interactions, relevant to both protein folding and biomolecular self-assembly. We observed that arginine is the most effective single excipient in stabilizing hydrophobic polymer collapse, and its effectiveness can be augmented by lysine or glutamate addition. We utilized a decomposition of the potential of mean force to identify that the key source of arginine-lysine and arginine-glutamate synergy on polymer collapse is a reduction in attractive polymer-excipient direct interactions. Further, we applied principles from network theory to characterize the local solvent network that embeds the hydrophobic polymer. Through this approach, we found that arginine enables a more highly connected and stable network than in pure water, lysine, or glutamate solutions. Importantly, these network properties are preserved when lysine or glutamate are added to arginine solutions. Overall, we highlight the importance of identifying key molecular consequences of co-excipient selection, aiding in the establishment of rational formulation design rules.

Adaptive networks can sense and adjust to dynamic environments to optimize their performance. Understanding their nanoscale responses to external stimuli is essential for applications in nanodevices and neuromorphic computing. However, it is challenging to image such responses on the nanoscale with crystallographic sensitivity. Here, the evolution of nanodomain networks in (PbTiO3)n/(SrTiO3)n superlattices was directly visualized in real space as the system adapts to ultrafast repetitive optical excitations that emulate controlled neural inputs. The adaptive response allows the system to explore a wealth of metastable states that were previously inaccessible. Their reconfiguration and competition were quantitatively measured by scanning x-ray nanodiffraction as a function of the number of applied pulses, in which crystallographic characteristics were quantitatively assessed by assorted diffraction patterns using unsupervised machine-learning methods. The corresponding domain boundaries and their connectivity were drastically altered by light, holding promise for light-programmable nanocircuits in analogy to neuroplasticity. Phase-field simulations elucidate that the reconfiguration of the domain networks is a result of the interplay between photocarriers and transient lattice temperature. The demonstrated optical control scheme and the uncovered nanoscopic insights open opportunities for remote control of adaptive nanoscale domain networks.

The NV$^{-}$ color center in diamond has been demonstrated as a powerful nanosensor for quantum metrology due to the sensitivity of its optical and spin properties to external electric, magnetic, and strain fields. In view of these applications, we use quantum embedding to derive a many-body description of strain and charge induced Stark effects on the NV$^{-}$ center. We quantify how strain longitudinal to the axis of NV$^{-}$ shifts the excited states in energy, while strain with a component transverse to the NV$^{-}$ axis splits the degeneracies of the $^{3}E$ and $^{1}E$ states. The largest effects are for the optically relevant $^{3}E$ manifold, which splits into $E_{x}$ and $E_{y}$ with transverse strain. From these responses we extract strain susceptibilities for the $E_{x/y}$ states within the quasi-linear regime. Additionally, we study the many-body dipole matrix elements of the NV$^{-}$ and find a permanent dipole 1.64 D at zero strain, which is somewhat smaller than that obtained from recent density functional theory calculations. We also determine the transition dipole between the $E_{x}$ and $E_{y}$ and how it evolves with strain.

Artificial Intelligence (AI) repeatedly match or outperform radiologists in lab experiments. However, real-world implementations of radiological AI-based systems are found to provide little to no clinical value. This paper explores how to design AI for clinical usefulness in different contexts. We conducted 19 design sessions and design interventions with 13 radiologists from 7 clinical sites in Denmark and Kenya, based on three iterations of a functional AI-based prototype. Ten sociotechnical dependencies were identified as crucial for the design of AI in radiology. We conceptualised four technical dimensions that must be configured to the intended clinical context of use: AI functionality, AI medical focus, AI decision threshold, and AI Explainability. We present four design recommendations on how to address dependencies pertaining to the medical knowledge, clinic type, user expertise level, patient context, and user situation that condition the configuration of these technical dimensions.

Arginine has been a mainstay in biological formulation development for decades. To date, the way arginine modulates protein stability has been widely studied and debated. Here, we employed a hydrophobic polymer to decouple hydrophobic effects from other interactions relevant to protein folding. While existing hypotheses for the effects of arginine can generally be categorized as either direct or indirect, our results indicate that direct and indirect mechanisms of arginine co-exist and oppose each other. At low concentrations, arginine was observed to stabilize hydrophobic polymer collapse via a sidechain-dominated direct mechanism, while at high concentrations, arginine stabilized polymer collapse via a backbone-dominated indirect mechanism. When adding partial charges to sites on the polymer, arginine destabilized polymer collapse. Further, we found arginine-induced destabilization of a model virus similar to direct-mechanism destabilization of the charged polymer, and concentration-dependent stabilization of a model protein similar to the indirect mechanism of hydrophobic polymer stabilization. These findings highlight the modular nature of the widely used additive arginine, with relevance in the design of stable biological formulations.

Quantum computers promise polynomial or exponential speed-up in solving certain problems compared to classical computers. However, in practical use, there are currently a number of fundamental technical challenges. One of them concerns the loading of data into quantum computers, since they cannot access common databases. In this vision paper, we develop a hybrid data management architecture in which databases can serve as data sources for quantum algorithms. To test the architecture, we perform experiments in which we assign data points stored in a database to clusters. For cluster assignment, a quantum algorithm processes this data by determining the distances between data points and cluster centroids., Comment: This preprint has not undergone any post-submission improvements or corrections. The Version of Record of this contribution is published in "Service-Oriented Computing - ICSOC 2023 Workshops" and is available online at https://doi.org/10.1007/978-981-97-0989-2_14

Many modern manufacturing companies have evolved from a single production site to a multi-factory production environment that must handle both geographically dispersed production orders and their multi-site production steps. The availability of a range of machines in different locations capable of performing the same operation and shipping times between factories have transformed planning systems from the classic Job Shop Scheduling Problem (JSSP) to Distributed Flexible Job Shop Scheduling Problem (DFJSP). As a result, the complexity of production planning has increased significantly. In our work, we use Quantum Annealing (QA) to solve the DFJSP. In addition to the assignment of production orders to production sites, the assignment of production steps to production sites also takes place. This requirement is based on a real use case of a wool textile manufacturer. To investigate the applicability of this method to large problem instances, problems ranging from 50 variables up to 250 variables, the largest problem that could be embedded into a D-Wave quantum annealer Quantum Processing Unit (QPU), are formulated and solved. Special attention is dedicated to the determination of the Lagrange parameters of the Quadratic Unconstrained Binary Optimization (QUBO) model and the QPU configuration parameters, as these factors can significantly impact solution quality. The obtained solutions are compared to solutions obtained by Simulated Annealing (SA), both in terms of solution quality and calculation time. The results demonstrate that QA has the potential to solve large problem instances specific to the industry., Comment: This work has been submitted to the IEEE for possible publication

Magnetic vortices are topological spin structures frequently found in ferromagnets, yet novel to antiferromagnets. By combining experiment and theory, we demonstrate that in a nanostructured antiferromagnetic-ferromagnetic NiO(111)-Fe(110) bilayer, a magnetic vortex is naturally stabilized by magnetostatic interactions in the ferromagnet and is imprinted onto the adjacent antiferromagnet via interface exchange coupling. We use micromagnetic simulations to construct a corresponding phase diagram of the stability of the imprinted antiferromagnetic vortex state. Our in depth analysis reveals that the interplay between interface exchange coupling and the antiferromagnet magnetic anisotropy plays a crucial role in locally reorienting the N\'eel vector out-of-plane in the prototypical in-plane antiferromagnet NiO and thereby stabilizing the vortices in the antiferromagnet.

Medical Imaging (MI) datasets are fundamental to artificial intelligence in healthcare. The accuracy, robustness, and fairness of diagnostic algorithms depend on the data (and its quality) used to train and evaluate the models. MI datasets used to be proprietary, but have become increasingly available to the public, including on community-contributed platforms (CCPs) like Kaggle or HuggingFace. While open data is important to enhance the redistribution of data's public value, we find that the current CCP governance model fails to uphold the quality needed and recommended practices for sharing, documenting, and evaluating datasets. In this paper, we conduct an analysis of publicly available machine learning datasets on CCPs, discussing datasets' context, and identifying limitations and gaps in the current CCP landscape. We highlight differences between MI and computer vision datasets, particularly in the potentially harmful downstream effects from poor adoption of recommended dataset management practices. We compare the analyzed datasets across several dimensions, including data sharing, data documentation, and maintenance. We find vague licenses, lack of persistent identifiers and storage, duplicates, and missing metadata, with differences between the platforms. Our research contributes to efforts in responsible data curation and AI algorithms for healthcare., Comment: NeurIPS 2024 Track on Datasets and Benchmarks. Please note that v1 has a different title

Fine-tuning is a widespread technique that allows practitioners to transfer pre-trained capabilities, as recently showcased by the successful applications of foundation models. However, fine-tuning reinforcement learning (RL) models remains a challenge. This work conceptualizes one specific cause of poor transfer, accentuated in the RL setting by the interplay between actions and observations: forgetting of pre-trained capabilities. Namely, a model deteriorates on the state subspace of the downstream task not visited in the initial phase of fine-tuning, on which the model behaved well due to pre-training. This way, we lose the anticipated transfer benefits. We identify conditions when this problem occurs, showing that it is common and, in many cases, catastrophic. Through a detailed empirical analysis of the challenging NetHack and Montezuma's Revenge environments, we show that standard knowledge retention techniques mitigate the problem and thus allow us to take full advantage of the pre-trained capabilities. In particular, in NetHack, we achieve a new state-of-the-art for neural models, improving the previous best score from $5$K to over $10$K points in the Human Monk scenario., Comment: ICML 2024 Spotlight

Language model attacks typically assume one of two extreme threat models: full white-box access to model weights, or black-box access limited to a text generation API. However, real-world APIs are often more flexible than just text generation: these APIs expose "gray-box" access leading to new threat vectors. To explore this, we red-team three new functionalities exposed in the GPT-4 APIs: fine-tuning, function calling and knowledge retrieval. We find that fine-tuning a model on as few as 15 harmful examples or 100 benign examples can remove core safeguards from GPT-4, enabling a range of harmful outputs. Furthermore, we find that GPT-4 Assistants readily divulge the function call schema and can be made to execute arbitrary function calls. Finally, we find that knowledge retrieval can be hijacked by injecting instructions into retrieval documents. These vulnerabilities highlight that any additions to the functionality exposed by an API can create new vulnerabilities., Comment: 10 pages, 1 figure, 4 tables

This paper shows how information about the network's community structure can be used to define node features with high predictive power for classification tasks. To do so, we define a family of community-aware node features and investigate their properties. Those features are designed to ensure that they can be efficiently computed even for large graphs. We show that community-aware node features contain information that cannot be completely recovered by classical node features or node embeddings (both classical and structural) and bring value in node classification tasks. This is verified for various classification tasks on synthetic and real-life networks., Comment: 21 pages, 3 figures, 7 tables

Solid-state devices made from correlated oxides such as perovskite nickelates are promising for neuromorphic computing by mimicking biological synaptic function. However, comprehending dopant action at the nanoscale poses a formidable challenge to understanding the elementary mechanisms involved. Here, we perform operando infrared nanoimaging of hydrogen-doped correlated perovskite, neodymium nickel oxide (H-NdNiO3) devices and reveal how an applied field perturbs dopant distribution at the nanoscale. This perturbation leads to stripe phases of varying conductivity perpendicular to the applied field, which define the macroscale electrical characteristics of the devices. Hyperspectral nano-FTIR imaging in conjunction with density functional theory calculations unveil a real-space map of multiple vibrational states of H-NNO associated with OH stretching modes and their dependence on the dopant concentration. Moreover, the localization of excess charges induces an out-of-plane lattice expansion in NNO which was confirmed by in-situ - x-ray diffraction and creates a strain that acts as a barrier against further diffusion. Our results and the techniques presented here hold great potential to the rapidly growing field of memristors and neuromorphic devices wherein nanoscale ion motion is fundamentally responsible for function., Comment: 30 pages, 5 figures in the main text and 5 figures in the Supplementary Material

One of the core goals of responsible AI development is ensuring high-quality training datasets. Many researchers have pointed to the importance of the annotation step in the creation of high-quality data, but less attention has been paid to the work that enables data annotation. We define this work as the design of ground truth schema and explore the challenges involved in the creation of datasets in the medical domain even before any annotations are made. Based on extensive work in three health-tech organisations, we describe five external and internal factors that condition medical dataset creation processes. Three external factors include regulatory constraints, the context of creation and use, and commercial and operational pressures. These factors condition medical data collection and shape the ground truth schema design. Two internal factors include epistemic differences and limits of labelling. These directly shape the design of the ground truth schema. Discussions of what constitutes high-quality data need to pay attention to the factors that shape and constrain what is possible to be created, to ensure responsible AI design.

This paper provides a practical approach to stochastic Lie systems, i.e. stochastic differential equations whose general solutions can be written as a function depending only on a generic family of particular solutions and some constants, so as to emphasise their applications. We correct the known stochastic Lie theorem characterising stochastic Lie systems, proving that, contrary to previous claims, it satisfies the Malliavin's principle. Meanwhile, we show that stochastic Lie systems admit new stochastic features in the Ito approach. New generalisations of stochastic Lie systems, like the so-called stochastic foliated Lie systems, are devised. Subsequently, we focus on stochastic (foliated) Lie systems that can be studied as Hamiltonian systems using different types of differential geometric structures. We study their stability properties and we devise the basics of an energy-momentum method. A stochastic Poisson coalgebra method is developed to derive superposition rules for Hamiltonian stochastic Lie systems. Applications of our results are found in coronavirus stochastic models, stochastic Lotka-Volterra systems, stochastic SIS models of different types, etc. Our results improve previous approaches by using stochastic differential equations instead of deterministic models designed to grasp some of their stochastic features., Comment: 24 pages

Class-incremental learning (CIL) is a particularly challenging variant of continual learning, where the goal is to learn to discriminate between all classes presented in an incremental fashion. Existing approaches often suffer from excessive forgetting and imbalance of the scores assigned to classes that have not been seen together during training. In this study, we introduce a novel approach, Prediction Error-based Classification (PEC), which differs from traditional discriminative and generative classification paradigms. PEC computes a class score by measuring the prediction error of a model trained to replicate the outputs of a frozen random neural network on data from that class. The method can be interpreted as approximating a classification rule based on Gaussian Process posterior variance. PEC offers several practical advantages, including sample efficiency, ease of tuning, and effectiveness even when data are presented one class at a time. Our empirical results show that PEC performs strongly in single-pass-through-data CIL, outperforming other rehearsal-free baselines in all cases and rehearsal-based methods with moderate replay buffer size in most cases across multiple benchmarks., Comment: ICLR 2024 camera ready

We consider applications of neural networks in nonlinear system identification and formulate a hypothesis that adjusting general network structure by incorporating frequency information or other known orthogonal transform, should result in an efficient neural network retaining its universal properties. We show that such a structure is a universal approximator and that using any orthogonal transform in a proposed way implies regularization during training by adjusting the learning rate of each parameter individually. We empirically show in particular, that such a structure, using the Fourier transform, outperforms equivalent models without orthogonality support.

Digital twins are virtual representations of physical objects or systems used for the purpose of analysis, most often via computer simulations, in many engineering and scientific disciplines. Recently, this approach has been introduced to computational medicine, within the concept of Digital Twin in Healthcare (DTH). Such research requires verification and validation of its models, as well as the corresponding sensitivity analysis and uncertainty quantification (VVUQ). From the computing perspective, VVUQ is a computationally intensive process, as it requires numerous runs with variations of input parameters. Researchers often use high-performance computing (HPC) solutions to run VVUQ studies where the number of parameter combinations can easily reach tens of thousands. However, there is a viable alternative to HPC for a substantial subset of computational models - serverless computing. In this paper we hypothesize that using the serverless computing model can be a practical and efficient approach to selected cases of running VVUQ calculations. We show this on the example of the EasyVVUQ library, which we extend by providing support for many serverless services. The resulting library - CloudVVUQ - is evaluated using two real-world applications from the computational medicine domain adapted for serverless execution. Our experiments demonstrate the scalability of the proposed approach., Comment: Accepted at CCGrid2023 conference

Diffusion models have achieved remarkable success in generating high-quality images thanks to their novel training procedures applied to unprecedented amounts of data. However, training a diffusion model from scratch is computationally expensive. This highlights the need to investigate the possibility of training these models iteratively, reusing computation while the data distribution changes. In this study, we take the first step in this direction and evaluate the continual learning (CL) properties of diffusion models. We begin by benchmarking the most common CL methods applied to Denoising Diffusion Probabilistic Models (DDPMs), where we note the strong performance of the experience replay with the reduced rehearsal coefficient. Furthermore, we provide insights into the dynamics of forgetting, which exhibit diverse behavior across diffusion timesteps. We also uncover certain pitfalls of using the bits-per-dimension metric for evaluating CL.

The present paper is devoted to possible generalizations of the classic Lagrange Mean Value Theorem. We consider a real-valued function of several variables that is only assumed to be continuous. The main concept is to replace the notion of the derivative by the so called bisequential tangent cone. We first prove Rolle and Lagrange type results and then we turn to comparing this cone with the Clarke subdifferential in the case of a Lipschitz function. We also investigate an approach using normal cones., Comment: 14 pages

Traditional 3D face models are based on mesh representations with texture. One of the most important models is FLAME (Faces Learned with an Articulated Model and Expressions), which produces meshes of human faces that are fully controllable. Unfortunately, such models have problems with capturing geometric and appearance details. In contrast to mesh representation, the neural radiance field (NeRF) produces extremely sharp renders. But implicit methods are hard to animate and do not generalize well to unseen expressions. It is not trivial to effectively control NeRF models to obtain face manipulation. The present paper proposes a novel approach, named NeRFlame, which combines the strengths of both NeRF and FLAME methods. Our method enables high-quality rendering capabilities of NeRF while also offering complete control over the visual appearance, similar to FLAME. Unlike conventional NeRF-based architectures that utilize neural networks to model RGB colors and volume density, NeRFlame employs FLAME mesh as an explicit density volume. As a result, color values are non-zero only in the proximity of the FLAME mesh. This FLAME backbone is then integrated into the NeRF architecture to predict RGB colors, allowing NeRFlame to explicitly model volume density and implicitly model RGB colors.

As quantum gates improve, it becomes increasingly difficult to characterize the remaining errors. Here we describe a class of coherent non-Markovian errors -- excitations due to an off-resonant drive -- that occur naturally in quantum devices that use time-dependent fields to generate gate operations. We show how these errors are mischaracterized using standard Quantum Computer Verification and Validation (QCVV) techniques that rely on Markovianity and are therefore often overlooked or assumed to be incoherent. We first demonstrate off-resonant errors within a simple toy model of Z-gates created by the AC Stark effect, then show how off-resonant errors manifest in all gates driven on a fixed-frequency transmon architecture, a prominent example being incidental cross-resonance interaction driven during single-qubit gates. Furthermore, the same methodology can access the errors caused by two-level systems (TLS), showing evidence of coherent, off-resonant interactions with subsystems that are not intentional qubits. While we explore these results and their impact on gate error for fixed-frequency devices, we note that off-resonant excitations potentially limit any architectures that use frequency selectivity., Comment: fixed typos, updated references, and improved explanations

In this manuscript we report on the scintillation properties and pulse shape discrimination (PSD) performance of new organic glass scintillator. Two cylindrical samples with dimensions of 2x2 inches were tested. Additionally, this two samples were used in stack configuration in order to measure the PSD characteristics of a sample with a size of 2x4 inches. The study covers the measurements of neutron/gamma discrimination capability, emission spectra, photoelectron yield and analysis of the light pulse shapes originating from events related to gamma-rays and fast neutrons. The results were compared to data recorded previously using an EJ-276 plastic scintillator, an EJ-309 liquid scintillator and a stilbene single crystal.

Understanding why certain individuals work well (or poorly) together as a team is a key research focus in the psychological and behavioral sciences and a fundamental problem for team-based organizations. Nevertheless, we have a limited ability to predict the social and work-related dynamics that will emerge from a given combination of team members. In this work, we model vocal turn-taking behavior within conversations as a parametric stochastic process on a network composed of the team members. More precisely, we model the dynamic of exchanging the `speaker token' among team members as a random walk in a graph that is driven by both individual level features and the conversation history. We fit our model to conversational turn-taking data extracted from audio recordings of multinational student teams during undergraduate engineering design internships. Through this real-world data we validate the explanatory power of our model and we unveil statistically significant differences in speaking behaviors between team members of different nationalities., Comment: 5 pages, 2 figures. To be published in the 2022 Conference Proceedings of the Asilomar Conference on Signals, Systems and Computers