Search

In this work, we present LesionLocator, a framework for zero-shot longitudinal lesion tracking and segmentation in 3D medical imaging, establishing the first end-to-end model capable of 4D tracking with dense spatial prompts. Our model leverages an extensive dataset of 23,262 annotated medical scans, as well as synthesized longitudinal data across diverse lesion types. The diversity and scale of our dataset significantly enhances model generalizability to real-world medical imaging challenges and addresses key limitations in longitudinal data availability. LesionLocator outperforms all existing promptable models in lesion segmentation by nearly 10 dice points, reaching human-level performance, and achieves state-of-the-art results in lesion tracking, with superior lesion retrieval and segmentation accuracy. LesionLocator not only sets a new benchmark in universal promptable lesion segmentation and automated longitudinal lesion tracking but also provides the first open-access solution of its kind, releasing our synthetic 4D dataset and model to the community, empowering future advancements in medical imaging. Code is available at: www.github.com/MIC-DKFZ/LesionLocator, Comment: Accepted at CVPR 2025

Data quantity and quality play a vital role in determining the performance of Large Language Models (LLMs). High-quality data, in particular, can significantly boost the LLM's ability to generalize on a wide range of downstream tasks. Large pre-training datasets for leading LLMs remain inaccessible to the public, whereas many open datasets are small in size (less than 5 trillion tokens), limiting their suitability for training large models. In this paper, we introduce GneissWeb, a large dataset yielding around 10 trillion tokens that caters to the data quality and quantity requirements of training LLMs. Our GneissWeb recipe that produced the dataset consists of sharded exact sub-string deduplication and a judiciously constructed ensemble of quality filters. GneissWeb achieves a favorable trade-off between data quality and quantity, producing models that outperform models trained on state-of-the-art open large datasets (5+ trillion tokens). We show that models trained using GneissWeb dataset outperform those trained on FineWeb-V1.1.0 by 2.73 percentage points in terms of average score computed on a set of 11 commonly used benchmarks (both zero-shot and few-shot) for pre-training dataset evaluation. When the evaluation set is extended to 20 benchmarks (both zero-shot and few-shot), models trained using GneissWeb still achieve a 1.75 percentage points advantage over those trained on FineWeb-V1.1.0.

Aspect sentiment quadruple prediction (ASQP) facilitates a detailed understanding of opinions expressed in a text by identifying the opinion term, aspect term, aspect category and sentiment polarity for each opinion. However, annotating a full set of training examples to fine-tune models for ASQP is a resource-intensive process. In this study, we explore the capabilities of large language models (LLMs) for zero- and few-shot learning on the ASQP task across five diverse datasets. We report F1 scores slightly below those obtained with state-of-the-art fine-tuned models but exceeding previously reported zero- and few-shot performance. In the 40-shot setting on the Rest16 restaurant domain dataset, LLMs achieved an F1 score of 52.46, compared to 60.39 by the best-performing fine-tuned method MVP. Additionally, we report the performance of LLMs in target aspect sentiment detection (TASD), where the F1 scores were also close to fine-tuned models, achieving 66.03 on Rest16 in the 40-shot setting, compared to 72.76 with MVP. While human annotators remain essential for achieving optimal performance, LLMs can reduce the need for extensive manual annotation in ASQP tasks.

The notion of almost LUR (ALUR) point is introduced in the paper [P. Bandyopadhyay et al., Some generalizations of locally uniform rotundity, J. Math. Anal. Appl., 252, 906-916 (2000)], where one says that the point $x$ of the unit sphere $S_{X}$ of a Banach space is an almost LUR (ALUR) point of $B_{X}$ if for any sequences $\{x_{n}\}\subseteq B_{X}$ and $\{x_{m}^{\ast}\}\subseteq B_{X^{\ast}}$, the condition $\lim_{m}\lim_{n}x_{m}^{\ast}\left(\frac{x_{n}+x}{2}\right)=1$ implies $\lim_{n}\left\Vert x_{n}-x\right\Vert =0$, without mentioning what is meant by $\lim_{m}\lim_{n}\gamma_{m,n}=\gamma$ for $\gamma$, $\gamma_{m,n}\in\mathbb{R}$; $X$ is ALUR if $X$ is almost LUR at any $x\in S_{X}$. Of course, the natural definition for this iterated limit would be that for each $m$ sufficiently large there exists $\mu_{m}:=\lim_{n\rightarrow\infty}\gamma_{m,n}\in\mathbb{R}$ and $\gamma =\lim_{m\rightarrow\infty}\mu_{m}$. However, as seen in some proofs where $\lim_{m}\lim_{n}$ appears, this interpretation is not confirmed. In this paper we examine several works in which almost LUR is mentioned and, especially, the proofs of those results in which the above definition of "almost LUR" point (or space) is invoked. Moreover, we analyze similar problems related to the notion CWALUR which extend ALUR. Furthermore, we mention several gaps in the proofs of some results. Finally, we propose the change of $\lim_{m}\lim_{n}$ by $\lim_{m}\liminf_{n}$ in the definitions of several types of ALUR points; moreover, we provide the complete proofs of two results from the literature in which $\lim_{m}\lim_{n}$ were used effectively, using $\lim_{m}\liminf_{n}$ instead., Comment: The paper has 16 pages an a bibliography containing 21 references. In the present version we corrected some inaccuracies and added a subnote on page 10

Cluster scaling relations are key ingredients in cluster abundance-based cosmological studies. In optical cluster cosmology, weak gravitational lensing has proven to be a powerful tool to constrain the cluster mass-richness relation. This work is conducted as part of the Dark Energy Science Collaboration (DESC), which aims to analyze the Legacy Survey of Space and Time (LSST) of Vera C. Rubin Observatory, starting in 2026. Weak lensing-inferred cluster properties, such as mass, suffer from several sources of bias. In this paper, we aim to test the impact of modeling choices and observational systematics in cluster lensing on the inference of the mass-richness relation. We constrain the mass-richness relation of 3,600 clusters detected by the redMaPPer algorithm in the cosmoDC2 extra-galactic mock catalog (covering $440$ deg$^2$) of the LSST DESC DC2 simulation, using number count measurements and stacked weak lensing profiles in several intervals of richness ($20 \leq \lambda \leq 200$) and redshift ($0.2 \leq z \leq 1$). By modeling the mean of the scaling relation as $\langle \ln \lambda|M_{\rm 200c}, z\rangle = \ln\lambda_0 + \mu_z\log[(1+z)/(1+0.5)] + \mu_m[\log_{10}(M_{\rm 200c}) - 14.3]$, our baseline constraints are $\ln\lambda_0 = 3.37\pm 0.03$, $\mu_z = 0.08\pm 0.07$ and $\mu_m = 2.18 \pm 0.07$. We have found that, for a LSST-like source galaxy density, our constraints are robust to a change in concentration-mass relation and dark matter density profile modeling choices, when source redshifts and shapes are perfectly known. We have found that photometric redshift uncertainties can introduce bias at the $1\sigma$ level, which can be mitigated by an overall correcting factor, fitted jointly with scaling parameters. We find that including positive shear-richness covariance in the fit shifts the results by up to 0.5$\sigma$., Comment: 26 pages, 15 figures, 3 tables, submitted to A&A (abstract shortened for arXiv)

Ultrafast electron microscopy aims for imaging transient phenomena occurring on nanoscale. One of its goals is to visualize localized optical and plasmonic modes generated by coherent excitation in the vicinity of various types of nanostructures. Such imaging capability was enabled by photon-induced near-field optical microscopy, which is based on spectral filtering of electrons inelastically scattered due to the stimulated interaction with the near-field. Here we report on the development of ultrafast 4D scanning transmission electron microscopy, which allows us to image the transverse components of the optical near-field while avoiding the need of electron spectral filtering. We demonstrate that this method is capable of imaging optical near-fields of a tungsten nanotip and ponderomotive potential of an optical standing wave with a spatial resolution of 21 nm., Comment: v1: preprint; licence: CC BY 4.0

The encapsulation of hydrogen within fullerene/fullerane cages offers a promising avenue for studying high pressure hydrogen dynamics. Through ab initio molecular dynamics simulations, we investigate the behavior of a system consisting of hydrogen atoms enclosed in a \ch{C20H20} dodecahedrane. Our findings reveal significant structural and dynamical changes as the cage undergoes compression, corresponding to radial symmetric vibration. We analyze geometric, energetic, and thermodynamic parameters, highlighting correlations and observing behavior analogous to high pressure phases of hydrogen. Notably, our study bridges the gap between theory and experiment by proposing a novel approach to achieving high pressures and temperatures experimentally. These results not only contribute to the understanding of hydrogen behavior under extreme conditions but also hold implications for the quest to attain metallic hydrogen - a milestone in materials science with potential applications in various fields., Comment: 19 pages, 4 figures

The Maximum Independent Set (MIS) problem is a fundamental combinatorial optimization task that can be naturally mapped onto the Ising Hamiltonian of neutral atom quantum processors. Given its connection to NP-hard problems and real-world applications, there has been significant experimental interest in exploring quantum advantage for MIS. Pioneering experiments on King's Lattice graphs suggested a quadratic speed-up over simulated annealing, but recent benchmarks using state-of-the-art methods found no clear advantage, likely due to the structured nature of the tested instances. In this work, we generate hard instances of unit-disk graphs by leveraging complexity theory results and varying key hardness parameters such as density and treewidth. For a fixed graph size, we show that increasing these parameters can lead to prohibitive classical runtime increases of several orders of magnitude. We then compare classical and quantum approaches on small instances and find that, at this scale, quantum solutions are slower than classical ones for finding exact solutions. Based on extended classical benchmarks at larger problem sizes, we estimate that scaling up to a thousand atoms with a 1 kHz repetition rate is a necessary step toward demonstrating a computational advantage with quantum methods., Comment: 11 pages, 5 figures

This paper reports on the results from a pilot study investigating the impact of automatic speech recognition (ASR) technology on interpreting quality in remote healthcare interpreting settings. Employing a within-subjects experiment design with four randomised conditions, this study utilises scripted medical consultations to simulate dialogue interpreting tasks. It involves four trainee interpreters with a language combination of Chinese and English. It also gathers participants' experience and perceptions of ASR support through cued retrospective reports and semi-structured interviews. Preliminary data suggest that the availability of ASR, specifically the access to full ASR transcripts and to ChatGPT-generated summaries based on ASR, effectively improved interpreting quality. Varying types of ASR output had different impacts on the distribution of interpreting error types. Participants reported similar interactive experiences with the technology, expressing their preference for full ASR transcripts. This pilot study shows encouraging results of applying ASR to dialogue-based healthcare interpreting and offers insights into the optimal ways to present ASR output to enhance interpreter experience and performance. However, it should be emphasised that the main purpose of this study was to validate the methodology and that further research with a larger sample size is necessary to confirm these findings., Comment: to appear in the Proceedings of Translation and the Computer (TC46)

While conventional Transformers generally operate on sequence data, they can be used in conjunction with structure models, typically SE(3)-invariant or equivariant graph neural networks (GNNs), for 3D applications such as protein structure modelling. These hybrids typically involve either (1) preprocessing/tokenizing structural features as input for Transformers or (2) taking Transformer embeddings and processing them within a structural representation. However, there is evidence that Transformers can learn to process structural information on their own, such as the AlphaFold3 structural diffusion model. In this work we show that Transformers can function independently as structure models when passed linear embeddings of coordinates. We first provide a theoretical explanation for how Transformers can learn to filter attention as a 3D Gaussian with learned variance. We then validate this theory using both simulated 3D points and in the context of masked token prediction for proteins. Finally, we show that pre-training protein Transformer encoders with structure improves performance on a downstream task, yielding better performance than custom structural models. Together, this work provides a basis for using standard Transformers as hybrid structure-language models.

Nucleus segmentation is an important analysis task in digital pathology. However, methods for automatic segmentation often struggle with new data from a different distribution, requiring users to manually annotate nuclei and retrain data-specific models. Vision foundation models (VFMs), such as the Segment Anything Model (SAM), offer a more robust alternative for automatic and interactive segmentation. Despite their success in natural images, a foundation model for nucleus segmentation in histopathology is still missing. Initial efforts to adapt SAM have shown some success, but did not yet introduce a comprehensive model for diverse segmentation tasks. To close this gap, we introduce PathoSAM, a VFM for nucleus segmentation, based on training SAM on a diverse dataset. Our extensive experiments show that it is the new state-of-the-art model for automatic and interactive nucleus instance segmentation in histopathology. We also demonstrate how it can be adapted for other segmentation tasks, including semantic nucleus segmentation. For this task, we show that it yields results better than popular methods, while not yet beating the state-of-the-art, CellViT. Our models are open-source and compatible with popular tools for data annotation. We also provide scripts for whole-slide image segmentation. Our code and models are publicly available at https://github.com/computational-cell-analytics/patho-sam.

Segmentation is an important analysis task for biomedical images, enabling the study of individual organelles, cells or organs. Deep learning has massively improved segmentation methods, but challenges remain in generalization to new conditions, requiring costly data annotation. Vision foundation models, such as Segment Anything Model (SAM), address this issue through broad segmentation capabilities. However, these models still require finetuning on annotated data, although with less annotations, to achieve optimal results for new conditions. As a downside, they require more computational resources. This makes parameter-efficient finetuning (PEFT) relevant for their application. We contribute the first comprehensive study of PEFT for SAM applied to biomedical segmentation by evaluating 9 PEFT methods on diverse datasets. We also provide an implementation of QLoRA for vision transformers and a new approach for resource-efficient finetuning of SAM. Our code is publicly available at https://github.com/computational-cell-analytics/peft-sam.

We show on complete metric spaces a polynomial tail decay for stationary measures of contracting on average generating measures. Our result applies for example to self-similar or self-affine measures., Comment: 9 pages

Large language models (LLMs) are vulnerable to universal jailbreaks-prompting strategies that systematically bypass model safeguards and enable users to carry out harmful processes that require many model interactions, like manufacturing illegal substances at scale. To defend against these attacks, we introduce Constitutional Classifiers: safeguards trained on synthetic data, generated by prompting LLMs with natural language rules (i.e., a constitution) specifying permitted and restricted content. In over 3,000 estimated hours of red teaming, no red teamer found a universal jailbreak that could extract information from an early classifier-guarded LLM at a similar level of detail to an unguarded model across most target queries. On automated evaluations, enhanced classifiers demonstrated robust defense against held-out domain-specific jailbreaks. These classifiers also maintain deployment viability, with an absolute 0.38% increase in production-traffic refusals and a 23.7% inference overhead. Our work demonstrates that defending against universal jailbreaks while maintaining practical deployment viability is tractable.

We generalise Hochman's theorem on the dimension of self-similar measures to contracting on average measures and show that a weaker condition than exponential separation on all scales is sufficient. Our proof uses a technique we call the variance summation method, avoiding the use of inverse theorems for entropy., Comment: 26 pages. arXiv admin note: text overlap with arXiv:2409.18936

In this paper we provide insight into the classes of strongly subadditive/superadditive functions by highlighting numerous new examples and new results., Comment: 17 pages

We prove a conjectured relation between the energy-momentum and the displacement norm of superconformal defects. The proof completes earlier results, and shows that supersymmetry identifies two natural notions of brane tension in Anti-de Sitter gravity. As a byproduct we show that a modification of the energy-momentum tensor that removes the stress of static superconformal defects, ensures also that the radiation these emit obeys the Null Energy Condition. This sheds new light on the radiation-reaction problem for moving charges., Comment: 33 pages; typos corrected and minor changes

In this paper, we tackle the challenge of instance segmentation for foreign objects in chest radiographs, commonly seen in postoperative follow-ups with stents, pacemakers, or ingested objects in children. The diversity of foreign objects complicates dense annotation, as shown in insufficient existing datasets. To address this, we propose the simple generation of synthetic data through (1) insertion of arbitrary shapes (lines, polygons, ellipses) with varying contrasts and opacities, and (2) cut-paste augmentations from a small set of semi-automatically extracted labels. These insertions are guided by anatomy labels to ensure realistic placements, such as stents appearing only in relevant vessels. Our approach enables networks to segment complex structures with minimal manually labeled data. Notably, it achieves performance comparable to fully supervised models while using 93\% fewer manual annotations.

Medical image segmentation is an important analysis task in clinical practice and research. Deep learning has massively advanced the field, but current approaches are mostly based on models trained for a specific task. Training such models or adapting them to a new condition is costly due to the need for (manually) labeled data. The emergence of vision foundation models, especially Segment Anything, offers a path to universal segmentation for medical images, overcoming these issues. Here, we study how to improve Segment Anything for medical images by comparing different finetuning strategies on a large and diverse dataset. We evaluate the finetuned models on a wide range of interactive and (automatic) semantic segmentation tasks. We find that the performance can be clearly improved for interactive segmentation. However, semantic segmentation does not benefit from pretraining on medical images. Our best model, MedicoSAM, is publicly available at https://github.com/computational-cell-analytics/medico-sam. We show that it is compatible with existing tools for data annotation and believe that it will be of great practical value.

Tokenization is a fundamental step in natural language processing, breaking text into units that computational models can process. While learned subword tokenizers have become the de-facto standard, they present challenges such as large vocabularies, limited adaptability to new domains or languages, and sensitivity to spelling errors and variations. To overcome these limitations, we investigate a hierarchical architecture for autoregressive language modelling that combines character-level and word-level processing. It employs a lightweight character-level encoder to convert character sequences into word embeddings, which are then processed by a word-level backbone model and decoded back into characters via a compact character-level decoder. This method retains the sequence compression benefits of word-level tokenization without relying on a rigid, predefined vocabulary. We demonstrate, at scales up to 7 billion parameters, that hierarchical transformers match the downstream task performance of subword-tokenizer-based models while exhibiting significantly greater robustness to input perturbations. Additionally, during continued pretraining on an out-of-domain language, our model trains almost twice as fast, achieves superior performance on the target language, and retains more of its previously learned knowledge. Hierarchical transformers pave the way for NLP systems that are more robust, flexible, and generalizable across languages and domains.

In this work, we present the implementation of all-optical method for directly measuring electron pulse duration in an ultrafast scanning electron microscope. Our approach is based on the interaction of electrons with the ponderomotive potential of an optical standing wave and provides a precise in situ technique to characterize femtosecond electron pulses across a wide range of electron energies (1-30 keV). By using single-photon photoemission of electrons by ultraviolet femtosecond laser pulses from a Schottky emitter we achieve electron pulse durations ranging from 0.5 ps at 30 keV to 2.7 ps at 5.5 keV under optimal conditions where Coulomb interactions are negligible. Additionally, we demonstrate that reducing the photon energy of the femtosecond pulses used for photoemission from 4.8 eV (257.5 nm) to 2.4 eV (515 nm) decreases the initial energy spread of emitted electrons, leading to significantly shorter pulse durations, particularly at lower electron energies., Comment: v1: preprint; licence: CC BY 4.0; all data available upon source download

We propose and theoretically demonstrate a technique that allows one to compensate for chromatic aberrations of traditional electron lenses in ultrafast electron microscopes. The technique is based on space- and time-dependent phase modulation of a pulsed electron beam using interaction with a shaped pulsed ponderomotive lens. The energy-selective focal distance is reached by combining the electron temporal chirp with the time-dependent size of the effective potential, with which the electrons interact. As a result, chromatic aberration can be reduced by up to a factor of seven. This approach paves the way for advanced transverse and longitudinal wavefront shaping of electrons in free space., Comment: v1: AAM; licence: CC BY 4.0, Supplementary material is available upon download of source file

Using the Tannakian formalism, one can attach to a principally polarized abelian variety a reductive group, along with a representation. We show that this group and the representation characterize Jacobians in genus up to $5$. More generally, our results hold on the bielliptic Prym locus in all genera. This gives the first evidence towards a recent conjecture by Weissauer and Kr\"amer. The main tool in our proof is a criterion for detecting Jacobians relying on Chern-Mather classes.

We develop entropy and variance results for the product of independent identically distributed random variables on Lie groups. Our results apply to the study of stationary measures in various contexts., Comment: 18 pages. arXiv admin note: substantial text overlap with arXiv:2409.18936

We theoretically demonstrate that ponderomotive interactions near the electron cross-over can be used for aberration correction in ultrafast electron microscopes. Highly magnified electron shadow images from Si$_3$N$_4$ thin films are utilized to visualize the distortions induced by spherical aberrations. Our simulations of electron-light interactions indicate that spherical aberrations can be compensated resulting in an aberration free angle of \SI{8.1}{mrad}. For achieving the necessary light distribution, we use a gradient descent algorithm to optimize Zernike polynomials and shape the light beam into a modified Gaussian and Laguerre-Gaussian beam., Comment: v1: AAM; licence: CC BY 4.0

This paper investigates the reference-less evaluation of machine translation for low-resource language pairs, known as quality estimation (QE). Segment-level QE is a challenging cross-lingual language understanding task that provides a quality score (0-100) to the translated output. We comprehensively evaluate large language models (LLMs) in zero/few-shot scenarios and perform instruction fine-tuning using a novel prompt based on annotation guidelines. Our results indicate that prompt-based approaches are outperformed by the encoder-based fine-tuned QE models. Our error analysis reveals tokenization issues, along with errors due to transliteration and named entities, and argues for refinement in LLM pre-training for cross-lingual tasks. We release the data, and models trained publicly for further research.

This study presents an open-source toolkit to address critical challenges in preprocessing data for self-supervised learning (SSL) for 3D medical imaging, focusing on data privacy and computational efficiency. The toolkit comprises two main components: a segmentation network that delineates foreground regions to optimize data sampling and thus reduce training time, and a segmentation network that identifies anonymized regions, preventing erroneous supervision in reconstruction-based SSL methods. Experimental results demonstrate high robustness, with mean Dice scores exceeding 98.5 across all anonymization methods and surpassing 99.5 for foreground segmentation tasks, highlighting the efficacy of the toolkit in supporting SSL applications in 3D medical imaging for both CT and MRI images. The weights and code is available at https://github.com/MIC-DKFZ/Foreground-and-Anonymization-Area-Segmentation., Comment: 6 pages

Building trusted datasets is critical for transparent and responsible Medical AI (MAI) research, but creating even small, high-quality datasets can take years of effort from multidisciplinary teams. This process often delays AI benefits, as human-centric data creation and AI-centric model development are treated as separate, sequential steps. To overcome this, we propose ScaleMAI, an agent of AI-integrated data curation and annotation, allowing data quality and AI performance to improve in a self-reinforcing cycle and reducing development time from years to months. We adopt pancreatic tumor detection as an example. First, ScaleMAI progressively creates a dataset of 25,362 CT scans, including per-voxel annotations for benign/malignant tumors and 24 anatomical structures. Second, through progressive human-in-the-loop iterations, ScaleMAI provides Flagship AI Model that can approach the proficiency of expert annotators (30-year experience) in detecting pancreatic tumors. Flagship Model significantly outperforms models developed from smaller, fixed-quality datasets, with substantial gains in tumor detection (+14%), segmentation (+5%), and classification (72%) on three prestigious benchmarks. In summary, ScaleMAI transforms the speed, scale, and reliability of medical dataset creation, paving the way for a variety of impactful, data-driven applications.

The field of 3D medical vision self-supervised learning lacks consistency and standardization. While many methods have been developed it is impossible to identify the current state-of-the-art, due to i) varying and small pre-training datasets, ii) varying architectures, and iii) being evaluated on differing downstream datasets. In this paper we bring clarity to this field and lay the foundation for further method advancements: We a) publish the largest publicly available pre-training dataset comprising 114k 3D brain MRI volumes and b) benchmark existing SSL methods under common architectures and c) provide the code of our framework publicly to facilitate rapid adoption and reproduction. This pre-print \textit{only describes} the dataset contribution (a); Data, benchmark, and codebase will be made available shortly., Comment: Pre-Print for Challenge proposal; Dataset, Benchmark and Codebase will be made available shortly once Benchmarking concludes

This summary of the second Terrestrial Very-Long-Baseline Atom Interferometry (TVLBAI) Workshop provides a comprehensive overview of our meeting held in London in April 2024, building on the initial discussions during the inaugural workshop held at CERN in March 2023. Like the summary of the first workshop, this document records a critical milestone for the international atom interferometry community. It documents our concerted efforts to evaluate progress, address emerging challenges, and refine strategic directions for future large-scale atom interferometry projects. Our commitment to collaboration is manifested by the integration of diverse expertise and the coordination of international resources, all aimed at advancing the frontiers of atom interferometry physics and technology, as set out in a Memorandum of Understanding signed by over 50 institutions., Comment: Summary of the second Terrestrial Very-Long-Baseline Atom Interferometry Workshop held at Imperial College London: https://indico.cern.ch/event/1369392/

Digital pathology offers a groundbreaking opportunity to transform clinical practice in histopathological image analysis, yet faces a significant hurdle: the substantial file sizes of pathological Whole Slide Images (WSI). While current digital pathology solutions rely on lossy JPEG compression to address this issue, lossy compression can introduce color and texture disparities, potentially impacting clinical decision-making. While prior research addresses perceptual image quality and downstream performance independently of each other, we jointly evaluate compression schemes for perceptual and downstream task quality on four different datasets. In addition, we collect an initially uncompressed dataset for an unbiased perceptual evaluation of compression schemes. Our results show that deep learning models fine-tuned for perceptual quality outperform conventional compression schemes like JPEG-XL or WebP for further compression of WSI. However, they exhibit a significant bias towards the compression artifacts present in the training data and struggle to generalize across various compression schemes. We introduce a novel evaluation metric based on feature similarity between original files and compressed files that aligns very well with the actual downstream performance on the compressed WSI. Our metric allows for a general and standardized evaluation of lossy compression schemes and mitigates the requirement to independently assess different downstream tasks. Our study provides novel insights for the assessment of lossy compression schemes for WSI and encourages a unified evaluation of lossy compression schemes to accelerate the clinical uptake of digital pathology.

The digital transformation of production requires new methods of data integration and storage, as well as decision making and support systems that work vertically and horizontally throughout the development, production, and use cycle. In this paper, we propose Data-to-Knowledge (and Knowledge-to-Data) pipelines for production as a universal concept building on a network of Digital Shadows (a concept augmenting Digital Twins). We show a proof of concept that builds on and bridges existing infrastructure to 1) capture and semantically annotates trajectory data from multiple similar but independent robots in different organisations and use cases in a data lakehouse and 2) an independent process that dynamically queries matching data for training an inverse dynamic foundation model for robotic control. The article discusses the challenges and benefits of this approach and how Data-to-Knowledge pipelines contribute efficiency gains and industrial scalability in a World Wide Lab as a research outlook., Comment: 15 pages, 6 figures, submitted to CAiSE 2025

This paper compares the accuracy of the terms extracted using SketchEngine, TBXTools and ChatGPT. In addition, it evaluates the quality of the definitions produced by ChatGPT for these terms. The research is carried out on a comparable corpus of fashion magazines written in English and Russian collected from the web. A gold standard for the fashion terminology was also developed by identifying web pages that can be harvested automatically and contain definitions of terms from the fashion domain in English and Russian. This gold standard was used to evaluate the quality of the extracted terms and of the definitions produced. Our evaluation shows that TBXTools and SketchEngine, while capable of high recall, suffer from reduced precision as the number of terms increases, which affects their overall performance. Conversely, ChatGPT demonstrates superior performance, maintaining or improving precision as more terms are considered. Analysis of the definitions produced by ChatGPT for 60 commonly used terms in English and Russian shows that ChatGPT maintains a reasonable level of accuracy and fidelity across languages, but sometimes the definitions in both languages miss crucial specifics and include unnecessary deviations. Our research reveals that no single tool excels universally; each has strengths suited to particular aspects of terminology extraction and application., Comment: To appear in the Proceedings of Translating and the Computer 2024 (TC46)

Robust Simultaneous Localization and Mapping (SLAM) is a crucial enabler for autonomous navigation in natural, unstructured environments such as parks and gardens. However, these environments present unique challenges for SLAM due to frequent seasonal changes, varying light conditions, and dense vegetation. These factors often degrade the performance of visual SLAM algorithms originally developed for structured urban environments. To address this gap, we present ROVER, a comprehensive benchmark dataset tailored for evaluating visual SLAM algorithms under diverse environmental conditions and spatial configurations. We captured the dataset with a robotic platform equipped with monocular, stereo, and RGB-D cameras, as well as inertial sensors. It covers 39 recordings across five outdoor locations, collected through all seasons and various lighting scenarios, i.e., day, dusk, and night with and without external lighting. With this novel dataset, we evaluate several traditional and deep learning-based SLAM methods and study their performance in diverse challenging conditions. The results demonstrate that while stereo-inertial and RGB-D configurations generally perform better under favorable lighting and moderate vegetation, most SLAM systems perform poorly in low-light and high-vegetation scenarios, particularly during summer and autumn. Our analysis highlights the need for improved adaptability in visual SLAM algorithms for outdoor applications, as current systems struggle with dynamic environmental factors affecting scale, feature extraction, and trajectory consistency. This dataset provides a solid foundation for advancing visual SLAM research in real-world, natural environments, fostering the development of more resilient SLAM systems for long-term outdoor localization and mapping. The dataset and the code of the benchmark are available under https://iis-esslingen.github.io/rover., Comment: 17 pages, 7 figures, 11 tables

Wire media (WM) resonators have emerged as promising realization for plasma haloscopes -- devices designed to detect axions, a potential component of dark matter. Key factors influencing the detection probability include cavity volume, resonance quality factor, and form factor. While the form factor has been explored for resonant frequency tuning, its optimization for axion detection remains unexplored. In this work, we present a novel approach to significantly enhance the form factor of WM plasma haloscopes. By shifting the metal walls of the resonator by a quarter wavelength, we effectively convert an electric wall boundary condition into a magnetic wall one, allowing for an almost uniform mode. Theoretical analysis and numerical simulations confirm that this modification improves the electric field profile and boosts the form factor. We validate these findings through experimental results from two prototype resonators: one with a standard geometry and another with a quarter-wave air gap between the WM and the walls. Additionally, our method provides a simple way to control the field profile within WM cavities, which can be explored for further applications.

This paper presents our approach to scaling the nnU-Net framework for multi-structure segmentation on Cone Beam Computed Tomography (CBCT) images, specifically in the scope of the ToothFairy2 Challenge. We leveraged the nnU-Net ResEnc L model, introducing key modifications to patch size, network topology, and data augmentation strategies to address the unique challenges of dental CBCT imaging. Our method achieved a mean Dice coefficient of 0.9253 and HD95 of 18.472 on the test set, securing a mean rank of 4.6 and with it the first place in the ToothFairy2 challenge. The source code is publicly available, encouraging further research and development in the field., Comment: Fabian Isensee and Yannick Kirchhoff contributed equally

In the quantum optimisation paradigm, variational quantum algorithms face challenges with hardware-specific and instance-dependent parameter tuning, which can lead to computational inefficiencies. However, the promising potential of parameter transferability across problem instances with similar local structures has been demonstrated in the context of the Quantum Approximate Optimisation Algorithm. In this paper, we build on these advancements by extending the concept to annealing-based protocols, employing Bayesian optimisation to design robust quasi-adiabatic schedules. Our study reveals that, for Maximum Independent Set problems on graph families with shared geometries, optimal parameters naturally concentrate, enabling efficient transfer from smaller to larger instances. Experimental results on the Orion Alpha platform validate the effectiveness of our approach, scaling to problems with up to $100$ qubits. We apply this method to address a smart-charging optimisation problem on a real dataset. These findings highlight a scalable, resource-efficient path for hybrid optimisation strategies applicable in real-world scenarios., Comment: 22 pages, 12 figures

A novel phase field method is proposed to model the continuous transition of binary fluids exhibiting temperature sensitive miscibility gap, from immiscible state to miscible state via partially miscible states. The model is employed to investigate the isothermal single-mode Rayleigh-Taylor (RT) instability for binary fluids as the system temperature is varied. Assuming potential flow and utilizing Boussinesq approximation, we derived the dispersion relation for gravity-capillary waves and the RT instability. The study reveals the early-stage growth characteristics of the interfacial perturbation. Three zones with distinct qualitative behaviour for the growth rate are identified as a function of Atwood number and Weber Number. Subsequently, Boussinesq approximation is relaxed to obtain coupled Cahn-Hilliard-Navier-Stokes equations to perform numerical simulations. The results from the numerical simulations corroborate the findings from the dispersion relation at early-stages. Further investigation of the late-time dynamics for viscous fluid pair reveal the tortuous topology presumed by the interface. The emanation of secondary instability in form of Kelvin-Helmholtz rolls is observed. The formation of Kelvin-Helmholtz rolls is found to be dependent on the system temperature. Finally, we present the effect of the slow nature of diffusion process., Comment: 28 pages

The performance of storage hardware has improved vastly recently, leaving the traditional I/O stack incapable of exploiting these gains due to increasingly large relative overheads. Newer asynchronous I/O APIs, such as io_uring, have significantly improved performance by reducing such overheads, but exhibit limited adoption in practice. In this paper, we discuss the complexities that the usage of these contemporary I/O APIs introduces to applications, which we believe are mostly responsible for their low adoption rate. Finally, we share implications and trade offs made by architectures that may be used to integrate asynchronous I/O into DB applications.

We report the detection of H$_2$O maser emission in 4 out of 77 (5.2%) mid-IR red galaxies that meet the color criteria of $W1-W2 > 0.5$ and $W1-W4 > 7$ and are classified as Type-2 AGNs based on optical, near-IR, and mid-IR spectral energy distribution (SED) fitting. Here, $W1$, $W2$, and $W4$ represent the IR magnitudes at 3.4, 4.6, and 22 micron, respectively, as measured by the Wide-field Infrared Survey Explorer. Three of the four newly identified maser galaxies are classified as either Seyfert 2 or LINER systems, but none are disk maser systems. Our analysis indicates that AGN identifications based solely on SED fitting are unreliable, resulting in an unexpectedly low detection rate. By restricting our sample to optically classified Type 2 AGNs that satisfy the mid-IR color criteria, we achieve a maser detection rate of ~13-18%, aligning with previous predictions for mid-IR red sources. These selection criteria are the most effective to date for facilitating new maser detections, particularly in light of the recent identification of additional Type 2 AGNs identified from ongoing galaxy and AGN surveys., Comment: 11 pages, 3 tables, 3 figures

In this work, we consider instantaneous transitions of an infinitely extended uniaxial dielectric into a wire medium (WM) of continuous infinitely long conducting wires. Due to the strong spatial dispersion in the WM the known (Morgenthaler's) theory of temporal discontinuities is not applicable. We solve this problem analytically in time domain. We show that a transverse electromagnetic (TM) plane wave transforms into four waves: a pair of TM waves and a pair of transverse electromagnetic waves. This way, the power flow splits into two different directions, with one of them along the wires. Such a transition can possibly be achieved by nanoplasma discharges in the gaps of the split wires, initiated by an external voltage source applied to the wire and transforming the split wires forming the uniaxial dielectric into continuous ones., Comment: A conference paper submitted to EuCAP 2025 Conference

In this paper, we prove the existence and the uniqueness of a weak and mild solution of the following nonlinear parabolic problem involving the porous $p$-fractional Laplacian: \begin{equation*} \begin{cases} \partial_t u+(-\Delta)^s_p(|u|^{m-1}u)=h(t,x,|u|^{m-1}u) & \text{in} \; (0,T)\times \Omega,\\ u=0 & \text{in} \; (0,T) \times \mathbb{R}^d\backslash \Omega, \\ u(0,\cdot)=u_0 & \text{in} \; \Omega . \end{cases}\ \end{equation*} We also study further the the homogeneous case $h(u)=|u|^{q-1}u$ with $q>0$. In particular we investigate global time existence, uniqueness, global behaviour of weak solutions and stabilization.

Quint and Shubik (1997) conjectured that a non-degenerate n-by-n game has at most 2^n-1 Nash equilibria in mixed strategies. The conjecture is true for n at most 4 but false for n=6 or larger. We answer it positively for the remaining case n=5, which had been open since 1999. The problem can be translated to a combinatorial question about the vertices of a pair of simple n-polytopes with 2n facets. We introduce a novel obstruction based on the index of an equilibrium, which states that equilibrium vertices belong to two equal-sized disjoint stable sets of the graph of the polytope. This bound is verified directly using the known classification of the 159,375 combinatorial types of dual neighborly polytopes in dimension 5 with 10 facets. Non-neighborly polytopes are analyzed with additional combinatorial techniques where the bound is used for their disjoint facets., Comment: Only lemmas and theorems with separate counters, as per journal requirement. Minor changes otherwise

In the context of satellite aerodynamics in the Very-Low-Earth-Orbit (VLEO) regime, accurate modeling of gas-surface interactions (GSI) is crucial for determining aerodynamic forces and torques. Common models such as Sentman's assume that gas particles are reflected diffusely from a surface, which leads to the incorporation of energy accommodation into the model. This technical note discusses the limitations of existing approaches for handling energy accommodation and provides a generalized treatment thereof that is valid for any molecular speed ratio. A new general expression for the temperature ratio of reflected to impinging particles is derived, which, when used in a GSI model, retains its validity even in hypothermal flows. Additionally, a simplified hyperthermal approximation is presented, proven to be an asymptote of the general expression, and shown to be an improvement upon existing approximations by comparison for a realistic VLEO scenario. The results contribute to a better understanding and modeling of GSI, potentially benefiting scientific investigations and operational applications in satellite aerodynamics.

In this work, we theoretically study temporal interfaces between media with strong spatial dispersion and dielectrics. In particular, we consider a temporal discontinuity that transforms a wire medium sample, a metamaterial with resonant spatial dispersion, into a uniaxial dielectric. We show that this transition results in a transformation of the deeply subwavelength spatial spectrum of TEM waves propagating in the wire medium at a certain frequency into a spectrum of plane waves at new frequencies that are all higher than the initial one. The waves at different frequencies propagate in different directions. Their complex amplitudes and propagation directions are uniquely related to the amplitudes of the spatial harmonics of the fields which existed before the transition. We explain how to implement this transition. The revealed effect may result in a promising method of subwavelength imaging.

Current interactive segmentation approaches, inspired by the success of META's Segment Anything model, have achieved notable advancements, however, they come with substantial limitations that hinder their practical application in 3D radiological scenarios. These include unrealistic human interaction requirements, such as slice-by-slice operations for 2D models on 3D data, a lack of iterative interactive refinement, and insufficient evaluation experiments. These shortcomings prevent accurate assessment of model performance and lead to inconsistent outcomes across studies. The RadioActive benchmark overcomes these challenges by offering a comprehensive and reproducible evaluation of interactive segmentation methods in realistic, clinically relevant scenarios. It includes diverse datasets, target structures, and interactive segmentation methods, and provides a flexible, extendable codebase that allows seamless integration of new models and prompting strategies. We also introduce advanced prompting techniques to enable 2D models on 3D data by reducing the needed number of interaction steps, enabling a fair comparison. We show that surprisingly the performance of slice-wise prompted approaches can match native 3D methods, despite the domain gap. Our findings challenge the current literature and highlight that models not specifically trained on medical data can outperform the current specialized medical methods. By open-sourcing RadioActive, we invite the research community to integrate their models and prompting techniques, ensuring continuous and transparent evaluation of interactive segmentation models in 3D medical imaging., Comment: Undergoing Peer-Review

Hyperspectral optoacoustic microscopy (OAM) enables obtaining images with label-free biomolecular contrast, offering excellent perspectives as a diagnostic tool to assess freshly excised and unprocessed tissues. However, time-consuming raster-scanning image formation currently limits the translation potential of OAM into the clinical setting-for instance, in intraoperative histopathological assessments-where micrographs of excised tissue need to be taken within a few minutes for fast clinical decision-making. Here, we present a non-data-driven computational framework tailored to enable fast OAM by sparse data acquisition and model-based image reconstruction, termed Bayesian raster-computed optoacoustic microscopy (BayROM). Unlike conventional machine learning, BayROM doesn't require training datasets, but instead, it employs 1) optomechanical system properties to define a forward model and 2) prior knowledge of the imaged samples to facilitate reconstructing images based on the sparsely acquired data. We show that BayROM enables acquiring micrographs ten times faster and with structural similarity (SSIM) indices greater than 0.93 compared to conventional raster scanning microscopy, thus facilitating the clinical translation of OAM for fast, label-free intraoperative histopathology.

How can we test AI performance? This question seems trivial, but it isn't. Standard benchmarks often have problems such as in-distribution and small-size test sets, oversimplified metrics, unfair comparisons, and short-term outcome pressure. As a consequence, good performance on standard benchmarks does not guarantee success in real-world scenarios. To address these problems, we present Touchstone, a large-scale collaborative segmentation benchmark of 9 types of abdominal organs. This benchmark is based on 5,195 training CT scans from 76 hospitals around the world and 5,903 testing CT scans from 11 additional hospitals. This diverse test set enhances the statistical significance of benchmark results and rigorously evaluates AI algorithms across various out-of-distribution scenarios. We invited 14 inventors of 19 AI algorithms to train their algorithms, while our team, as a third party, independently evaluated these algorithms on three test sets. In addition, we also evaluated pre-existing AI frameworks--which, differing from algorithms, are more flexible and can support different algorithms--including MONAI from NVIDIA, nnU-Net from DKFZ, and numerous other open-source frameworks. We are committed to expanding this benchmark to encourage more innovation of AI algorithms for the medical domain., Comment: Accepted to NeurIPS-2024

As LLMs become more widely deployed, there is increasing interest in directly optimizing for feedback from end users (e.g. thumbs up) in addition to feedback from paid annotators. However, training to maximize human feedback creates a perverse incentive structure for the AI to resort to manipulative or deceptive tactics to obtain positive feedback from users who are vulnerable to such strategies. We study this phenomenon by training LLMs with Reinforcement Learning with simulated user feedback in environments of practical LLM usage. In our settings, we find that: 1) Extreme forms of "feedback gaming" such as manipulation and deception are learned reliably; 2) Even if only 2% of users are vulnerable to manipulative strategies, LLMs learn to identify and target them while behaving appropriately with other users, making such behaviors harder to detect; 3) To mitigate this issue, it may seem promising to leverage continued safety training or LLM-as-judges during training to filter problematic outputs. Instead, we found that while such approaches help in some of our settings, they backfire in others, sometimes even leading to subtler manipulative behaviors. We hope our results can serve as a case study which highlights the risks of using gameable feedback sources -- such as user feedback -- as a target for RL., Comment: Accepted to ICLR 2025