Search

Chronic rhinosinusitis (CRS) is a common and persistent sinus imflammation that affects 5 - 12\% of the general population. It significantly impacts quality of life and is often difficult to assess due to its subjective nature in clinical evaluation. We introduce PARASIDE, an automatic tool for segmenting air and soft tissue volumes of the structures of the sinus maxillaris, frontalis, sphenodalis and ethmoidalis in T1 MRI. By utilizing that segmentation, we can quantify feature relations that have been observed only manually and subjectively before. We performed an exemplary study and showed both volume and intensity relations between structures and radiology reports. While the soft tissue segmentation is good, the automated annotations of the air volumes are excellent. The average intensity over air structures are consistently below those of the soft tissues, close to perfect separability. Healthy subjects exhibit lower soft tissue volumes and lower intensities. Our developed system is the first automated whole nasal segmentation of 16 structures, and capable of calculating medical relevant features such as the Lund-Mackay score.

Machine learning-based classifiers are commonly evaluated by metrics like accuracy, but deeper analysis is required to understand their strengths and weaknesses. MLMC is a visual exploration tool that tackles the challenge of multi-label classifier comparison and evaluation. It offers a scalable alternative to confusion matrices which are commonly used for such tasks, but don't scale well with a large number of classes or labels. Additionally, MLMC allows users to view classifier performance from an instance perspective, a label perspective, and a classifier perspective. Our user study shows that the techniques implemented by MLMC allow for a powerful multi-label classifier evaluation while preserving user friendliness., Comment: 12 pages

Benchmarks are important tools for tracking the rapid advancements in large language model (LLM) capabilities. However, benchmarks are not keeping pace in difficulty: LLMs now achieve over 90\% accuracy on popular benchmarks like MMLU, limiting informed measurement of state-of-the-art LLM capabilities. In response, we introduce Humanity's Last Exam (HLE), a multi-modal benchmark at the frontier of human knowledge, designed to be the final closed-ended academic benchmark of its kind with broad subject coverage. HLE consists of 3,000 questions across dozens of subjects, including mathematics, humanities, and the natural sciences. HLE is developed globally by subject-matter experts and consists of multiple-choice and short-answer questions suitable for automated grading. Each question has a known solution that is unambiguous and easily verifiable, but cannot be quickly answered via internet retrieval. State-of-the-art LLMs demonstrate low accuracy and calibration on HLE, highlighting a significant gap between current LLM capabilities and the expert human frontier on closed-ended academic questions. To inform research and policymaking upon a clear understanding of model capabilities, we publicly release HLE at https://lastexam.ai., Comment: 25 pages, 6 figures

During the past decade, state-of-the-art planet-finder instruments like SPHERE@VLT, coupling coronagraphic devices and extreme adaptive optics systems, unveiled, thanks to large surveys, around 20 planetary mass companions at semi-major axis greater than 10 astronomical units. Direct imaging being the only detection technique to be able to probe this outer region of planetary systems, the SPHERE infrared survey for exoplanets (SHINE) was designed and conducted from 2015 to 2021 to study the demographics of such young gas giant planets around 400 young nearby solar-type stars. In this paper, we present the observing strategy, the data quality, and the point sources analysis of the full SHINE statistical sample as well as snapSHINE. Both surveys used the SPHERE@VLT instrument with the IRDIS dual band imager in conjunction with the integral field spectrograph IFS and the angular differential imaging observing technique. All SHINE data (650 datasets), corresponding to 400 stars, including the targets of the F150 survey, are processed in a uniform manner with an advanced post-processing algorithm called PACO ASDI. An emphasis is put on the classification and identification of the most promising candidate companions. Compared to the previous early analysis SHINE F150, the use of advanced post-processing techniques significantly improved by one or 2 magnitudes (x3-x6) the contrast detection limits, which will allow us to put even tighter constraints on the radial distribution of young gas giants. This increased sensitivity directly places SHINE as the largest and deepest direct imaging survey ever conducted. We detected and classified more than 3500 physical sources. One additional substellar companion has been confirmed during the second phase of the survey (HIP 74865 B), and several new promising candidate companions are awaiting second epoch confirmations., Comment: Accepted for publication in A&A. 19 pages, 19 figures. Abstract shortened to comply with ArxiV standards. Data ingestion at the CDS is ongoing

We report on an experimental study of spin and valley blockade in two-electron bilayer graphene (BLG) double quantum dots (DQDs) and explore the limits set by asymmetric orbitals and electronelectron interactions. The results obtained from magnetotransport measurements on two-electron BLG DQDs, where the resonant tunneling transport involves both orbital symmetric and antisymmetric two-particle states, show a rich level spectrum. We observe a magnetic field tunable spin and valley blockade, which is limited by the orbital splitting, the strength of the electron-electron interaction and the difference in the valley g-factors between the symmetric and antisymmetric twoparticle orbital states. Our conclusions are supported by simulations based on rate equations, which allow the identification of prominent interdot transitions associated with the transition from single to two-particle states observed in the experiment., Comment: 16 pages, 11 figures

Recent cosmological analyses measuring distances of Type Ia Supernovae (SNe Ia) and Baryon Acoustic Oscillations (BAO) have all given similar hints at time-evolving dark energy. To examine whether underestimated SN Ia systematics might be driving these results, Efstathiou (2024) compared overlapping SN events between Pantheon+ and DES-SN5YR (20% SNe are in common), and reported evidence for a $\sim$0.04 mag offset between the low and high-redshift distance measurements of this subsample of events. If these offsets are arbitrarily subtracted from the entire DES-SN5YR sample, the preference for evolving dark energy is reduced. In this paper, we reproduce this offset and show that it has two sources. First, 43% of the offset is due to DES-SN5YR improvements in the modelling of supernova intrinsic scatter and host galaxy properties. These are scientifically-motivated modelling updates implemented in DES-SN5YR and their associated uncertainties are captured within the DES-SN5YR systematic error budget. Even if the less accurate scatter model and host properties from Pantheon+ are used instead, the DES-SN5YR evidence for evolving dark energy is only reduced from 3.9$\sigma$ to 3.3$\sigma$. Second, 38% of the offset is due to a misleading comparison because different selection functions characterize the DES subsets included in Pantheon+ and DES-SN5YR and therefore individual SN distance measurements are expected to be different because of different bias corrections. In conclusion, we confirm the validity of the published DES-SN5YR results., Comment: 9 pages, 5 figures

In three-dimensional space an electron moving in the field of a magnetic monopole has no bound states. In this paper we explore the physics when the electron is restricted to a two-dimensional plane adjacent to a magnetic monopole. We find bound states in the classical version of the problem and quasi-bound states in the quantum one, in addition to a continuum of scattering states. We calculate the lifetimes of the quasi-bound states using several complementary approximate methods, which agree well in the cases where the lifetimes are relatively short. The threshold monopole magnetic charge required to realise a single quasi-bound state is approximately $18Q_D$, where $Q_D$ is the magnetic charge of a Dirac monopole. We examine the feasibility of achieving this magnetic charge in currently available monopole analogues: spin ice, artificial spin ice, and magnetic needles., Comment: 26 pages

Counterfactual examples are widely used in natural language processing (NLP) as valuable data to improve models, and in explainable artificial intelligence (XAI) to understand model behavior. The automated generation of counterfactual examples remains a challenging task even for large language models (LLMs), despite their impressive performance on many tasks. In this paper, we first introduce ZeroCF, a faithful approach for leveraging important words derived from feature attribution methods to generate counterfactual examples in a zero-shot setting. Second, we present a new framework, FitCF, which further verifies aforementioned counterfactuals by label flip verification and then inserts them as demonstrations for few-shot prompting, outperforming two state-of-the-art baselines. Through ablation studies, we identify the importance of each of FitCF's core components in improving the quality of counterfactuals, as assessed through flip rate, perplexity, and similarity measures. Furthermore, we show the effectiveness of LIME and Integrated Gradients as backbone attribution methods for FitCF and find that the number of demonstrations has the largest effect on performance. Finally, we reveal a strong correlation between the faithfulness of feature attribution scores and the quality of generated counterfactuals., Comment: In submission

In this article we prove results about finite soluble groups that act with fixity 2 or 3.

Recently, a direct connection between static structure factors and quantum ground states for two-spin interaction Hamiltonians was proven. This suggests the possibility of quantum state tomography from neutron scattering. Here, we investigate the associated fitness landscape numerically. We find a linear relationship between the mean square distances of the structure factors and the associated state overlaps, implying a well-behaved fitness landscape. Furthermore, we find evidence suggesting that the approach can be generalized to thermal equilibrium states. We also extend the arguments to the cases of applied magnetic fields and finite clusters., Comment: 9 pages, 3 figures, to be submitted to the Physical Review Letters

Being able to reconstruct training data from the parameters of a neural network is a major privacy concern. Previous works have shown that reconstructing training data, under certain circumstances, is possible. In this work, we analyse such reconstructions empirically and propose a new formulation of the reconstruction as a solution to a bilevel optimisation problem. We demonstrate that our formulation as well as previous approaches highly depend on the initialisation of the training images $x$ to reconstruct. In particular, we show that a random initialisation of $x$ can lead to reconstructions that resemble valid training samples while not being part of the actual training dataset. Thus, our experiments on affine and one-hidden layer networks suggest that when reconstructing natural images, yet an adversary cannot identify whether reconstructed images have indeed been part of the set of training samples.

As humanoid service robots are becoming more and more perceptible in public service settings for instance as a guide to welcome visitors or to explain a procedure to follow, it is desirable to improve the comprehensibility of complex issues for human customers and to adapt the level of difficulty of the information provided as well as the language used to individual requirements. This work examines a case study using a humanoid social robot Pepper performing support for customers in a public service environment offering advice and information. An application architecture is proposed that improves the intelligibility of the information received by providing the possibility to translate this information into easy language and/or into another spoken language.

Different hybrid quantum-classical algorithms have recently been developed as a near-term way to solve linear systems of equations on quantum devices. However, the focus has so far been mostly on the methods, rather than the problems that they need to tackle. In fact, these algorithms have been run on real hardware only for problems in quantum physics, such as Hamiltonians of a few qubits systems. These problems are particularly favorable for quantum hardware, since their matrices are the sum of just a few unitary terms and since only shallow quantum circuits are required to estimate the cost function. However, for many interesting problems in linear algebra, it appears far less trivial to find an efficient decomposition and to trade it off with the depth of the cost quantum circuits. A first simple yet interesting instance to consider are tridiagonal systems of equations. These arise, for instance, in the discretization of one-dimensional finite element analyses. This work presents a method to solve a class of tridiagonal systems of equations with the variational quantum linear solver (VQLS), a recently proposed variational hybrid algorithm for solving linear systems. In particular, we present a new decomposition for this class of matrices based on both Pauli strings and multi--qubit gates, resulting in less terms than those obtained by just using Pauli gates. Based on this decomposition, we discuss the tradeoff between the number of terms and the near-term implementability of the quantum circuits. Furthermore, we present the first simulated and real-hardware results obtained by solving tridiagonal linear systems with VQLS, using the decomposition proposed., Comment: 7 pages, 2 figures, 2 tables

The giant molecular cloud complex Sagittarius B2 (Sgr~B2) in the central molecular zone of our Galaxy hosts several high-mass star formation sites, with Sgr~B2(M) and Sgr~B2(N) being the main centers of activity. This analysis aims to comprehensively model each core spectrum, considering molecular lines, dust attenuation, and free-free emission interactions. We describe the molecular content analysis of each hot core and identify the chemical composition of detected sources. Using ALMA's high sensitivity, we aim to characterize the hot core population in Sgr~B2(M) and N, gaining a better understanding of the different evolutionary phases of star formation processes in this complex. We conducted an unbiased ALMA spectral line survey of 47 sources in band 6 (211-275 GHz). Chemical composition and column densities were derived using XCLASS, assuming local thermodynamic equilibrium. Quantitative descriptions for each molecule were determined, considering all emission and absorption features across the spectral range. Temperature and velocity distributions were analyzed, and derived abundances were compared with other spectral line surveys. We identified 65 isotopologs from 41 different molecules, ranging from light molecules to complex organic compounds, originating from various environments. Most sources in the Sgr~B2 complex were assigned different evolutionary phases of high-mass star formation. Sgr~B2(N) hot cores show more complex molecules such as CH$_3$OH, CH$_3$OCHO, and CH$_3$OCH$_3$, while M cores contain lighter molecules such as SO$_2$, SO, and NO. Some sulfur-bearing molecules are more abundant in N than in M. The derived molecular abundances can be used for comparison and to constrain astrochemical models. Inner sources in both regions were generally more developed than outer sources, with some exceptions.

Object pose estimation from a single view remains a challenging problem. In particular, partial observability, occlusions, and object symmetries eventually result in pose ambiguity. To account for this multimodality, this work proposes training a diffusion-based generative model for 6D object pose estimation. During inference, the trained generative model allows for sampling multiple particles, i.e., pose hypotheses. To distill this information into a single pose estimate, we propose two novel and effective pose selection strategies that do not require any additional training or computationally intensive operations. Moreover, while many existing methods for pose estimation primarily focus on the image domain and only incorporate depth information for final pose refinement, our model solely operates on point cloud data. The model thereby leverages recent advancements in point cloud processing and operates upon an SE(3)-equivariant latent space that forms the basis for the particle selection strategies and allows for improved inference times. Our thorough experimental results demonstrate the competitive performance of our approach on the Linemod dataset and showcase the effectiveness of our design choices. Code is available at https://github.com/zitronian/6DPoseDiffusion .

We present qlbm, a Python software package designed to facilitate the development, simulation, and analysis of Quantum Lattice Boltzmann Methods (QBMs). qlbm is a modular framework that introduces a quantum component abstraction hierarchy tailored to the implementation of novel QBMs. The framework interfaces with state-of-the-art quantum software infrastructure to enable efficient simulation and validation pipelines, and leverages novel execution and pre-processing techniques that significantly reduce the computational resources required to develop quantum circuits. We demonstrate the versatility of the software by showcasing multiple QBMs in 2D and 3D with complex boundary conditions, integrated within automated benchmarking utilities. Accompanying the source code are extensive test suites, thorough online documentation resources, analysis tools, visualization methods, and demos that aim to increase the accessibility of QBMs while encouraging reproducibility and collaboration. The source code of qlbm is publicly available under a permissive MPL 2.0 license at \url{https://github.com/QCFD-Lab/qlbm}., Comment: 31 pages, 19 figures

Compact binary mergers detectable in gravitational waves can be accompanied by a kilonova, an electromagnetic transient powered by radioactive decay of newly synthesised r-process elements. A few kilonova candidates have been observed during short gamma-ray burst follow-up, and one found associated with a gravitational wave detection, GW170817. However, robust kilonova candidates are yet to be found in un-triggered, wide-field optical surveys, that is, a search not requiring an initial gravitational wave or gamma-ray burst trigger. Here we present the first observing run for the Kilonova and Transients Program (KNTraP) using the Dark Energy Camera. The first KNTraP run ran for 11 nights, covering 31 fields at a nightly cadence in two filters. The program can detect transients beyond the LIGO/Virgo/KAGRA horizon, be agnostic to the merger orientation, avoid the Sun and/or Galactic plane, and produces high cadence multi-wavelength light curves. The data were processed nightly in real-time for rapid identification of transient candidates, allowing for follow-up of interesting candidates before they faded away. Three fast-rising candidates were identified in real-time, however none had the characteristics of the kilonova AT2017gfo associated with GW170817 or with the expected evolution for kilonovae from our fade-rate models. After the run, the data were reprocessed, then subjected to stringent filtering and model fitting to search for kilonovae offline. Multiple KNTraP runs (3+) are expected to detect kilonovae via this optical-only search method. No kilonovae were detected in this first KNTraP run using our selection criteria, constraining the KN rate to $R < 1.8\times10^{5}$ Gpc$^{-3}$ yr$^{-1}$., Comment: 17 pages, 6 figures, 3 tables, submitted to MNRAS

GRB$\,$220831A is a gamma-ray burst (GRB) with a duration and spectral peak energy that places it at the interface between the distribution of long-soft and short-hard GRBs. In this paper, we present the multi-wavelength follow-up campaign to GRB$\,$220831A and its optical, near-infrared, X-ray and radio counterparts. Our deep optical and near-infrared observations do not reveal an underlying host galaxy, and establish that GRB$\,$220831A is observationally hostless to depth, $m_i\gtrsim26.6$ AB mag. Based on the Amati relation and the non-detection of an accompanying supernova, we find that this GRB is most likely to have originated from a collapsar at $z>2$, but it could also possibly be a compact object merger at $z<0.4$ with a large separation distance from its host galaxy. Regardless of its origin, we show that its optical and near-infrared counterpart departs from the evolution expected from a forward shock dominated synchrotron afterglow, exhibiting a steep post-break temporal powerlaw index of $-3.83^{+0.62}_{-0.79}$, too steep to be the jet-break. By analysing a range of models, we find that the observed steep departure from forward shock closure relations is likely due to an internal process producing either a flare or a plateau., Comment: 17 pages, 14 figures, 7 tables, accepted for publication in MNRAS

We explicitly construct nondegenerate braided $\mathbb{Z}_2$-crossed tensor categories of the form $\operatorname{Vect}_\Gamma\oplus\operatorname{Vect}_{\Gamma/2\Gamma}$. They are $\mathbb{Z}_2$-crossed extensions, in the sense of arXiv:0909.3140, of the braided tensor category $\operatorname{Vect}_\Gamma$ with $\mathbb{Z}_2$-action given by $-\mathrm{id}$ on the finite, abelian group $\Gamma$. Thus, we obtain generalisations of the Tambara-Yamagami categories, where now the abelian group $\Gamma$ may have even order and the nontrivial sector $\operatorname{Vect}_{\Gamma/2\Gamma}$ more than one simple object. The idea for this construction comes from a physically motivated approach in arXiv:2409.16357 to construct $\mathbb{Z}_2$-crossed extensions of $\operatorname{Vect}_\Gamma$ for any $\Gamma$ from an infinite Tambara-Yamagami category $\operatorname{Vect}_{\mathbb{R}^d}\oplus\operatorname{Vect}$, which itself is not fully rigorously defined, and then using condensation from $\operatorname{Vect}_{\mathbb{R}^d}$ to $\operatorname{Vect}_\Gamma$, which we prove commutes with crossed extensions. The $\mathbb{Z}_2$-equivariantisation of $\operatorname{Vect}_\Gamma\oplus\operatorname{Vect}_{\Gamma/2\Gamma}$ yields new modular tensor categories, which correspond to the orbifold of an arbitrary lattice vertex operator algebra under a lift of $-\mathrm{id}$, as discussed in arXiv:2409.16357., Comment: 47 pages, LaTeX; the construction in this paper was originally part of a longer paper arXiv:2409.16357, from which it is split off to improve readability

Lifted probabilistic inference exploits symmetries in a probabilistic model to allow for tractable probabilistic inference with respect to domain sizes of logical variables. We found that the current state-of-the-art algorithm to construct a lifted representation in form of a parametric factor graph misses symmetries between factors that are exchangeable but scaled differently, thereby leading to a less compact representation. In this paper, we propose a generalisation of the advanced colour passing (ACP) algorithm, which is the state of the art to construct a parametric factor graph. Our proposed algorithm allows for potentials of factors to be scaled arbitrarily and efficiently detects more symmetries than the original ACP algorithm. By detecting strictly more symmetries than ACP, our algorithm significantly reduces online query times for probabilistic inference when the resulting model is applied, which we also confirm in our experiments., Comment: Accepted to the Proceedings of the 3rd Learning on Graphs Conference (LoG 2024)

XAI research often focuses on settings where people learn about and assess algorithmic systems individually. However, as more public AI systems are deployed, it becomes essential for XAI to facilitate collective understanding and deliberation. We conducted a task-based interview study involving 8 focus groups and 12 individual interviews to explore how explanations can support AI novices in understanding and forming opinions about AI systems. Participants received a collection of explanations organized into four information categories to solve tasks and decide about a system's deployment. These explanations improved or calibrated participants' self-reported understanding and decision confidence and facilitated group discussions. Participants valued both technical and contextual information and the self-directed and modular explanation structure. Our contributions include an explanation approach that facilitates both individual and collaborative interaction and explanation design recommendations, including active and controllable exploration, different levels of information detail and breadth, and adaptations to the needs of decision subjects., Comment: 24 pages main text, 7 figures, 4 tables, supplementary material included

In the last decade, crowdsourcing has become a popular method for conducting quantitative empirical studies in human-machine interaction. The remote work on a given task in crowdworking settings suits the character of typical speech/language-based interactive systems for instance with regard to argumentative conversations and information retrieval. Thus, crowdworking promises a valuable opportunity to study and evaluate the usability and user experience of real humans in interactions with such interactive systems. In contrast to physical attendance in laboratory studies, crowdsourcing studies offer much more flexible and easier access to large numbers of heterogeneous participants with a specific background, e.g., native speakers or domain expertise. On the other hand, the experimental and environmental conditions as well as the participant's compliance and reliability (at least better monitoring of the latter) are much better controllable in a laboratory. This paper seeks to present a (self-)critical examination of crowdsourcing-based studies in the context of complex (spoken) dialogue systems. It describes and discusses observed issues in crowdsourcing studies involving complex tasks and suggests solutions to improve and ensure the quality of the study results. Thereby, our work contributes to a better understanding and what needs to be considered when designing and evaluating studies with crowdworkers for complex dialogue systems., Comment: 14 pages, accepted to 13th International Workshop on Spoken Dialogue Systems (IWSDS), Los Angeles, USA, February 21--24, 2023

We, the QAIMS lab lab at the Aerospace Faculty of TU Delft, participated as finalists in the Airbus/BMW Quantum Computing Challenge 2024. Stacking sequence retrieval, a complex combinatorial task within a bi-level optimization framework, is crucial for designing laminated composites that meet aerospace requirements for weight, strength, and stiffness. This document presents the scientifically relevant sections of our submission, which builds on our prior research on applying quantum computation to this challenging design problem. For the competition, we expanded our previous work in several significant ways. First, we incorporated a full set of manufacturing constraints into our algorithmic framework, including those previously established theoretically but not yet demonstrated, thereby aligning our approach more closely with real-world manufacturing demands. We implemented the F-VQE algorithm, which enhances the probability shaping of optimal solutions, improving on simpler variational quantum algorithms. Our approach also demonstrates flexibility by accommodating diverse objectives as well as finer ply-angle increments alongside the previously demonstrated conventional ply angles. Scalability was tested using the DMRG algorithm, which, despite limitations in entanglement representation, enabled simulations with up to 200 plies. Results were directly compared to conventional stacking sequence retrieval algorithms with DMRG showing high competitiveness. Given DMRG's limited entanglement capabilities, it serves as a conservative baseline, suggesting potential for even greater performance on fully realized quantum systems. This document serves to make our competition results publicly available as we prepare a formal publication on these findings and their implications for aerospace materials design optimization., Comment: Scientifically relevant sections of a submission to 2024 Airbus/BMW Quantum Computing Challenge. 26 pages, 10 figures, 3 tables

This study explored a concept for using Augmented Reality (AR) glasses to customize augmented table setup and lighting in a restaurant. The aim was to provide insights into AR usage in restaurants and contribute to existing research by introducing an extendable and versatile concept for scholars and restaurateurs. A controlled laboratory study, using a within-subjects design, was conducted to investigate the effects of a customizable augmented table setup and lighting on user experience (UX), perceived waiting time, psychological ownership, and social acceptability. A simulated restaurant environment was created using a 360-degree image in Virtual Reality (VR). The study implemented default and customizable table setup and lighting. Results from a paired samples t-test showed a statistically significant effect of table setup and lighting on the pragmatic quality of UX, hedonic quality of UX, overall UX, valence, dominance, psychological ownership, and affect. Furthermore, table setup had a significant effect on arousal and perceived waiting time. Moreover, table setup significantly affected AR interaction, isolation, and safety acceptability, while lighting only affected AR interaction acceptability. Findings suggest that these investigated variables are worth considering for AR applications in a restaurant, especially when offering customizable augmented table setup and lighting., Comment: This paper was accepted as short paper at IEEE ISM 2024, and will be presented on the 13th of December 2024

Lifting uses a representative of indistinguishable individuals to exploit symmetries in probabilistic relational models, denoted as parametric factor graphs, to speed up inference while maintaining exact answers. In this paper, we show how lifting can be applied to causal inference in partially directed graphs, i.e., graphs that contain both directed and undirected edges to represent causal relationships between random variables. We present partially directed parametric causal factor graphs (PPCFGs) as a generalisation of previously introduced parametric causal factor graphs, which require a fully directed graph. We further show how causal inference can be performed on a lifted level in PPCFGs, thereby extending the applicability of lifted causal inference to a broader range of models requiring less prior knowledge about causal relationships., Comment: Accepted to the Proceedings of the 16th International Conference on Scalable Uncertainty Management (SUM 2024)

We study the existence of periodic colorings and orientations in locally finite graphs. A coloring or orientation of a graph $G$ is periodic if the resulting colored or oriented graph is quasi-transitive, meaning that $V(G)$ has finitely many orbits under the action of the group of automorphisms of $G$ preserving the coloring or the orientation. When such a periodic coloring or orientation of $G$ exists, $G$ itself must be quasi-transitive and it is natural to investigate when quasi-transitive graphs have such periodic colorings or orientations. We provide examples of Cayley graphs with no periodic orientation or non-trivial coloring, and examples of quasi-transitive graphs of treewidth 2 without periodic orientation or proper coloring. On the other hand we show that every quasi-transitive graph $G$ of bounded pathwidth has a periodic proper coloring with $\chi(G)$ colors and a periodic orientation. We relate these problems with techniques and questions from symbolic dynamics and distributed computing and conclude with a number of open problems., Comment: 23 pages, 3 figures. v2: small corrections in the final section

The ability to read, interpret, and critique data visualizations has mainly been assessed using data visualization tasks like value retrieval. Although evidence on different facets of Visual Data Literacy (VDL) is well established in visualization research and includes numeracy, graph familiarity, or aesthetic elements, they have not been sufficiently considered in ability assessments. Here, VDL is considered a multidimensional ability whose facets can be partially self-assessed. We introduce an assessment in which VDL is deconstructed as a process of understanding, in reference to frameworks from the learning sciences. MAVIL, Multidimensional Assessment of Visual Data Literacy, is composed of six ability dimensions: General Impression/Abstract Thinking, Graph Elements/Familiarity, Aesthetic Perception, Visualization Criticism, Data Reading Tasks and Numeracy/Topic Knowledge. MAVIL was designed for general audiences and implemented in a survey (n=438), representative of Austria's age groups (18-74 years) and gender split. The survey mirrors the population's VDL and shows the perception of two climate data visualizations, a line and bar chart. We found that $48\%$ of respondents make mistakes with the simple charts, while $5\%$ believe that they cannot summarize the visualization content. About a quarter have deficits in comprehending simple data units, and $19-20\%$ are unfamiliar with each displayed chart type., Comment: 9 pages, 8 figures, 1 table

Recent advancements in large language models (LLMs) have led to their increased application across various tasks, with reinforcement learning from human feedback (RLHF) being a crucial part of their training to align responses with user intentions. In the RLHF process, a reward model is trained using responses preferences determined by human labelers or AI systems, which then refines the LLM through reinforcement learning. This work introduces weak supervision as a strategy to extend RLHF datasets and enhance reward model performance. Weak supervision employs noisy or imprecise data labeling, reducing reliance on expensive manually labeled data. By analyzing RLHF datasets to identify heuristics that correlate with response preference, we wrote simple labeling functions and then calibrated a label model to weakly annotate unlabeled data. Our evaluation show that while weak supervision significantly benefits smaller datasets by improving reward model performance, its effectiveness decreases with larger, originally labeled datasets. Additionally, using an LLM to generate and then weakly label responses offers a promising method for extending preference data.

GPT-4o is an autoregressive omni model that accepts as input any combination of text, audio, image, and video, and generates any combination of text, audio, and image outputs. It's trained end-to-end across text, vision, and audio, meaning all inputs and outputs are processed by the same neural network. GPT-4o can respond to audio inputs in as little as 232 milliseconds, with an average of 320 milliseconds, which is similar to human response time in conversation. It matches GPT-4 Turbo performance on text in English and code, with significant improvement on text in non-English languages, while also being much faster and 50\% cheaper in the API. GPT-4o is especially better at vision and audio understanding compared to existing models. In line with our commitment to building AI safely and consistent with our voluntary commitments to the White House, we are sharing the GPT-4o System Card, which includes our Preparedness Framework evaluations. In this System Card, we provide a detailed look at GPT-4o's capabilities, limitations, and safety evaluations across multiple categories, focusing on speech-to-speech while also evaluating text and image capabilities, and measures we've implemented to ensure the model is safe and aligned. We also include third-party assessments on dangerous capabilities, as well as discussion of potential societal impacts of GPT-4o's text and vision capabilities.

The even spin components of the strata of Abelian differentials are difficult to handle from a birational geometry perspective due to the fact that their spin line bundles have more sections than expected. Nevertheless, in this paper, we prove that for large genus, the minimal even spin components are of general type. This result complements the previous work by the second and third authors, together with Costantini, on the Kodaira dimension of general strata and the minimal odd spin components of Abelian differentials. Our main technical tool is the computation and estimation of a series of effective divisor classes on the even spin components.

This work addresses the problem of \textit{shape completion}, i.e., the task of restoring incomplete shapes by predicting their missing parts. While previous works have often predicted the fractured and restored shape in one step, we approach the task by separately predicting the fractured and newly restored parts, but ensuring these predictions are interconnected. We use a decoder network motivated by related work on the prediction of signed distance functions (DeepSDF). In particular, our representation allows us to consider test-time-training, i.e., finetuning network parameters to match the given incomplete shape more accurately during inference. While previous works often have difficulties with artifacts around the fracture boundary, we demonstrate that our overfitting to the fractured parts leads to significant improvements in the restoration of eight different shape categories of the ShapeNet data set in terms of their chamfer distances.

Understanding the neurobiology of opioid use disorder (OUD) using resting-state functional magnetic resonance imaging (rs-fMRI) may help inform treatment strategies to improve patient outcomes. Recent literature suggests temporal characteristics of rs-fMRI blood oxygenation level-dependent (BOLD) signals may offer complementary information to functional connectivity analysis. However, existing studies of OUD analyze BOLD signals using measures computed across all time points. This study, for the first time in the literature, employs data-driven machine learning (ML) modeling of rs-fMRI BOLD features representing multiple time points to identify region(s) of interest that differentiate OUD subjects from healthy controls (HC). Following the triple network model, we obtain rs-fMRI BOLD features from the default mode network (DMN), salience network (SN), and executive control network (ECN) for 31 OUD and 45 HC subjects. Then, we use the Boruta ML algorithm to identify statistically significant BOLD features that differentiate OUD from HC, identifying the DMN as the most salient functional network for OUD. Furthermore, we conduct brain activity mapping, showing heightened neural activity within the DMN for OUD. We perform 5-fold cross-validation classification (OUD vs. HC) experiments to study the discriminative power of functional network features with and without fusing demographic features. The DMN shows the most discriminative power, achieving mean AUC and F1 scores of 80.91% and 73.97%, respectively, when fusing BOLD and demographic features. Follow-up Boruta analysis using BOLD features extracted from the medial prefrontal cortex, posterior cingulate cortex, and left and right temporoparietal junctions reveals significant features for all four functional hubs within the DMN., Comment: 20 pages, 7 figures, 8 tables

When can two strongly rational vertex operator algebras or 1+1d rational conformal field theories (RCFTs) be related by topological manipulations? For vertex operator algebras, the term "topological manipulations" refers to operations like passing to a conformal extension or restricting to a conformal subalgebra; for RCFTs, topological manipulations include operations like gauging (or orbifolding) a finite subpart of a generalized global symmetry or interpolating to a new theory via a topological line interface of finite quantum dimension. Inspired by results in the theory of even lattices and tensor categories, we say that two strongly rational vertex operator algebras are Witt equivalent if their central charges agree and if their modular tensor categories are Witt equivalent. Two RCFTs are said to be Witt equivalent if their central charges agree and if their associated 2+1d topological field theories can be separated by a topological surface. We argue that Witt equivalence is necessary for two theories to be related by topological manipulations. We conjecture that it is also sufficient, and give proofs in various special cases. We relate this circle of ideas to the problem of classifying RCFTs, and to lore concerning deformation classes of quantum field theories. We use the notion of Witt equivalence to argue, assuming the conjectural classification of unitary, $c=1$ RCFTs, that all of the finite symmetries of the $SU(2)_1$ Wess-Zumino-Witten model are invertible. We also sketch a "quantum Galois theory" for chiral CFTs, which generalizes prior mathematical literature by incorporating non-invertible symmetries; we illustrate this non-invertible Galois theory in the context of the monster CFT, for which we produce two Fibonacci lines. Finally, we discuss $p$-neighborhood of vertex operator algebras, which is a special topological manipulation related to $\mathbb{Z}_p$-orbifolding., Comment: 77 pages, 17 figures, LaTeX, new section on p-neighborhood of vertex operator algebras added, new Fibonacci symmetry of the monster CFT constructed, references updated

Wrinkling is the phenomenon of out-of-plane deformation patterns in thin walled structures, as a result of a local compressive (internal) loads in combination with a large membrane stiffness and a small but non-zero bending stiffness. Numerical modelling typically involves thin shell formulations. As the mesh resolution depends on the wrinkle wave lengths, the analysis can become computationally expensive for shorter ones. Implicitly modeling the wrinkles using a modified kinematic or constitutive relationship based on a taut, slack or wrinkled state derived from a so-called tension field, a simplification is introduced in order to reduce computational efforts. However, this model was restricted to linear elastic material models in previous works. Aiming to develop an implicit isogeometric wrinkling model for large strain and hyperelastic material applications, a modified deformation gradient has been assumed, which can be used for any strain energy density formulation. The model is an extension of a previously published model for linear elastic material behaviour and is generalized to other types of discretisation as well. The extension for hyperelastic materials requires the derivative of the material tensor, which can be computed numerically or derived analytically. The presented model relies on a combination of dynamic relaxation and a Newton-Raphson solver, because of divergence in early Newton-Raphson iterations as a result of a changing tension field, which is not included in the stress tensor variation. Using four benchmarks, the model performance is evaluated. Convergence with the expected order for Newton-Raphson iterations has been observed, provided a fixed tension field. The model accurately approximates the mean surface of a wrinkled membrane with a reduced number of degrees of freedom in comparison to a shell solution.

In this work, we introduce a methodology for alignment designed to enhance the ability of large language models (LLMs) to articulate their reasoning (self-explanation) even in the absence of annotated rationale explanations. Our alignment methodology comprises three key components: explanation quality assessment, self-instruction dataset generation, and model alignment. Additionally, we present a novel technique called Alignment with Anchor Preference Pairs, which improves the selection of preference pairs by categorizing model outputs into three groups: consistently correct, consistently incorrect, and variable. By applying tailored strategies to each category, we enhance the effectiveness of Direct Preference Optimization (DPO). Our experimental results demonstrate that this approach significantly improves explanation quality while maintaining accuracy compared to other fine-tuning strategies.

Deep learning has made significant strides in medical imaging, leveraging the use of large datasets to improve diagnostics and prognostics. However, large datasets often come with inherent errors through subject selection and acquisition. In this paper, we investigate the use of Diffusion Autoencoder (DAE) embeddings for uncovering and understanding data characteristics and biases, including biases for protected variables like sex and data abnormalities indicative of unwanted protocol variations. We use sagittal T2-weighted magnetic resonance (MR) images of the neck, chest, and lumbar region from 11186 German National Cohort (NAKO) participants. We compare DAE embeddings with existing generative models like StyleGAN and Variational Autoencoder. Evaluations on a large-scale dataset consisting of sagittal T2-weighted MR images of three spine regions show that DAE embeddings effectively separate protected variables such as sex and age. Furthermore, we used t-SNE visualization to identify unwanted variations in imaging protocols, revealing differences in head positioning. Our embedding can identify samples where a sex predictor will have issues learning the correct sex. Our findings highlight the potential of using advanced embedding techniques like DAEs to detect data quality issues and biases in medical imaging datasets. Identifying such hidden relations can enhance the reliability and fairness of deep learning models in healthcare applications, ultimately improving patient care and outcomes., Comment: This paper was accepted in the "Workshop on Interpretability of Machine Intelligence in Medical Image Computing" (iMIMIC) at MICCAI 2024

Understanding protein motion within the cell is crucial for predicting reaction rates and macromolecular transport in the cytoplasm. A key question is how crowded environments affect protein dynamics through hydrodynamic and direct interactions at molecular length scales. Using megahertz X-ray Photon Correlation Spectroscopy (MHz-XPCS) at the European X-ray Free Electron Laser (EuXFEL), we investigate ferritin diffusion at microsecond time scales. Our results reveal anomalous diffusion, indicated by the non-exponential decay of the intensity autocorrelation function $g_2(q,t)$ at high concentrations. This behavior is consistent with the presence of cage-trapping in between the short- and long-time protein diffusion regimes. Modeling with the $\delta\gamma$-theory of hydrodynamically interacting colloidal spheres successfully reproduces the experimental data by including a scaling factor linked to the protein direct interactions. These findings offer new insights into the complex molecular motion in crowded protein solutions, with potential applications for optimizing ferritin-based drug delivery, where protein diffusion is the rate-limiting step.

The growing use of Machine Learning and Artificial Intelligence (AI), particularly Large Language Models (LLMs) like OpenAI's GPT series, leads to disruptive changes across organizations. At the same time, there is a growing concern about how organizations handle personal data. Thus, privacy policies are essential for transparency in data processing practices, enabling users to assess privacy risks. However, these policies are often long and complex. This might lead to user confusion and consent fatigue, where users accept data practices against their interests, and abusive or unfair practices might go unnoticed. LLMss can be used to assess privacy policies for users automatically. In this interdisciplinary work, we explore the challenges of this approach in three pillars, namely technical feasibility, ethical implications, and legal compatibility of using LLMs to assess privacy policies. Our findings aim to identify potential for future research, and to foster a discussion on the use of LLM technologies for enabling users to fulfil their important role as decision-makers in a constantly developing AI-driven digital economy., Comment: Published at AISyS 2024

Gravitational lensing magnification of Type Ia supernovae (SNe Ia) allows information to be obtained about the distribution of matter on small scales. In this paper, we derive limits on the fraction $\alpha$ of the total matter density in compact objects (which comprise stars, stellar remnants, small stellar groupings and primordial black holes) of mass $M > 0.03 M_{\odot}$ over cosmological distances. Using 1,532 SNe Ia from the Dark Energy Survey Year 5 sample (DES-SN5YR) combined with a Bayesian prior for the absolute magnitude $M$, we obtain $\alpha < 0.12$ at the 95\% confidence level after marginalisation over cosmological parameters, lensing due to large-scale structure, and intrinsic non-Gaussianity. Similar results are obtained using priors from the cosmic microwave background, baryon acoustic oscillations and galaxy weak lensing, indicating our results do not depend on the background cosmology. We argue our constraints are likely to be conservative (in the sense of the values we quote being higher than the truth), but discuss scenarios in which they could be weakened by systematics of the order of $\Delta \alpha \sim 0.04$, Comment: Accepted by MNRAS

Classically, the splitting principle says how to pull back a vector bundle in such a way that it splits into line bundles and the pullback map induces an injection on $K$-theory. Here we categorify the splitting principle and generalize it to the context of 2-rigs. A 2-rig is a kind of categorified "ring without negatives", such as a category of vector bundles with $\oplus$ as addition and $\otimes$ as multiplication. Technically, we define a 2-rig to be a Cauchy complete $k$-linear symmetric monoidal category where $k$ has characteristic zero. We conjecture that for any suitably finite-dimensional object $r$ of a 2-rig $\mathsf{R}$, there is a 2-rig map $E \colon \mathsf{R} \to \mathsf{R'}$ such that $E(r)$ splits as a direct sum of finitely many "subline objects" and $E$ has various good properties: it is faithful, conservative, essentially injective, and the induced map of Grothendieck rings $K(E) \colon K(\mathsf{R}) \to K(\mathsf{R'})$ is injective. We prove this conjecture for the free 2-rig on one object, namely the category of Schur functors, whose Grothendieck ring is the free $\lambda$-ring on one generator, also known as the ring of symmetric functions. We use this task as an excuse to develop the representation theory of affine categories - that is, categories enriched in affine schemes - using the theory of 2-rigs., Comment: 47 pages

Nowadays, Mixed Reality (MR) headsets are a game-changer for knowledge work. Unlike stationary monitors, MR headsets allow users to work with large virtual displays anywhere they wear the headset, whether in a professional office, a public setting like a cafe, or a quiet space like a library. This study compares four different layouts (eye level-close, eye level-far, below eye level-close, below eye level-far) of virtual displays regarding feelings of safety, perceived productivity, and social acceptability when working with MR in public. We test which layout is most preferred by users and seek to understand which factors affect users' layout preferences. The aim is to derive useful insights for designing better MR layouts. A field study in a public library was conducted using a within-subject design. While the participants interact with a layout, they are asked to work on a planning task. The results from a repeated measure ANOVA show a statistically significant effect on productivity but not on safety and social acceptability. Additionally, we report preferences expressed by the users regarding the layouts and using MR in public., Comment: Part of the 23rd annual IEEE International Symposium on Mixed and Augmented Reality (ISMAR 2024)

The EyeSight feature, introduced with the new Apple Vision Pro XR headset, promises to revolutionize user interaction by simulating real human eye expressions on a digital display. This feature could enhance XR devices' social acceptability and social presence when communicating with others outside the XR experience. In this pilot study, we explore the implications of the EyeSight feature by examining social acceptability, social presence, emotional responses, and technology acceptance. Eight participants engaged in conversational tasks in three conditions to contrast experiencing the Apple Vision Pro with EyeSight, the Meta Quest 3 as a reference XR headset, and a face-to-face setting. Our preliminary findings indicate that while the EyeSight feature improves perceptions of social presence and acceptability compared to the reference headsets, it does not match the social connectivity of direct human interactions., Comment: In 30th ACM Symposium on Virtual Reality Software and Technology (VRST 2024)

We present a hybrid approach combining isogeometric analysis with deep operator networks to solve electromagnetic scattering problems. The neural network takes a computer-aided design representation as input and predicts the electromagnetic field in a de Rham conforming B-spline basis such that for example the tangential continuity of the electric field is respected. The physical problem is included in the loss function during training. Our numerical results demonstrate that a trained network accurately predicts the electric field, showing convergence to the analytical solution with optimal rate. Additionally, training on a variety of geometries highlights the network's generalization capabilities, achieving small error increases when applied to new geometries not included in the training set.

In this article, we develop tools for computing $G$-crossed extensions of braided tensor categories. Their equivariantisations appear as categories of modules of fixed-point subalgebras (or orbifolds) of vertex operator algebras and are often difficult to determine. As the first tool, we show how the seminal work of Etingof, Nikshych and Ostrik on the uniqueness of $G$-crossed extensions can be used to determine the category of modules of orbifold vertex operator algebras. As an application, we determine the modular tensor category of the orbifold of a lattice vertex operator algebra under a lift of $-\mathrm{id}$ for a lattice with odd-order discriminant form. In that case, the de-equivariantisation is of Tambara-Yamagami type. As the second tool, we describe how $G$-crossed extensions and condensations by commutative algebras commute in a suitable sense. This leads to an effective approach to compute new $G$-crossed extensions. As one application, we produce the coherence data that are then used in arXiv:2411.12251 to define a generalisation of the Tambara-Yamagami categories with more than one simple object in the twisted sector. This also yields the modular tensor category of the orbifold of an arbitrary lattice vertex operator algebra under a lift of $-\mathrm{id}$. Finally, we sketch how to categorically approach the general problem of lattice orbifolds under lifts of arbitrary lattice involutions., Comment: 54 pages, LaTeX; the construction of the generalised Tambara-Yamagami categories was split off and can now be found in arXiv:2411.12251 to improve readability

Natural language explanations (NLEs) are vital for elucidating the reasoning behind large language model (LLM) decisions. Many techniques have been developed to generate NLEs using LLMs. However, like humans, LLMs might not always produce optimal NLEs on first attempt. Inspired by human learning processes, we introduce Cross-Refine, which employs role modeling by deploying two LLMs as generator and critic, respectively. The generator outputs a first NLE and then refines this initial explanation using feedback and suggestions provided by the critic. Cross-Refine does not require any supervised training data or additional training. We validate Cross-Refine across three NLP tasks using three state-of-the-art open-source LLMs through automatic and human evaluation. We select Self-Refine (Madaan et al., 2023) as the baseline, which only utilizes self-feedback to refine the explanations. Our findings from automatic evaluation and a user study indicate that Cross-Refine outperforms Self-Refine. Meanwhile, Cross-Refine can perform effectively with less powerful LLMs, whereas Self-Refine only yields strong results with ChatGPT. Additionally, we conduct an ablation study to assess the importance of feedback and suggestions. Both of them play an important role in refining explanations. We further evaluate Cross-Refine on a bilingual dataset in English and German., Comment: Accepted at COLING 2025; long paper

Visual Parameter Space Analysis (VPSA) enables domain scientists to explore input-output relationships of computational models. Existing VPSA applications often feature multi-view visualizations designed by visualization experts for a specific scenario, making it hard for domain scientists to adapt them to their problems without professional help. We present RSVP, the Rapid Suggestive Visualization Prototyping system encoding VPSA knowledge to enable domain scientists to prototype custom visualization dashboards tailored to their specific needs. The system implements a task-oriented, multi-view visualization recommendation strategy over a visualization design space optimized for VPSA to guide users in meeting their analytical demands. We derived the VPSA knowledge implemented in the system by conducting an extensive meta design study over the body of work on VPSA. We show how this process can be used to perform a data and task abstraction, extract a common visualization design space, and derive a task-oriented VisRec strategy. User studies indicate that the system is user-friendly and can uncover novel insights., Comment: 16 pages, 16 figures, 3 tables, accepted for publishing at TVCG Journal

Probabilistic relational models provide a well-established formalism to combine first-order logic and probabilistic models, thereby allowing to represent relationships between objects in a relational domain. At the same time, the field of artificial intelligence requires increasingly large amounts of relational training data for various machine learning tasks. Collecting real-world data, however, is often challenging due to privacy concerns, data protection regulations, high costs, and so on. To mitigate these challenges, the generation of synthetic data is a promising approach. In this paper, we solve the problem of generating synthetic relational data via probabilistic relational models. In particular, we propose a fully-fledged pipeline to go from relational database to probabilistic relational model, which can then be used to sample new synthetic relational data points from its underlying probability distribution. As part of our proposed pipeline, we introduce a learning algorithm to construct a probabilistic relational model from a given relational database., Comment: Accepted to the Proceedings of the 47th German Conference on Artificial Intelligence (KI 2024)

Dual encoder architectures like CLIP models map two types of inputs into a shared embedding space and learn similarities between them. However, it is not understood how such models compare two inputs. Here, we address this research gap with two contributions. First, we derive a method to attribute predictions of any differentiable dual encoder onto feature-pair interactions between its inputs. Second, we apply our method to CLIP-type models and show that they learn fine-grained correspondences between parts of captions and regions in images. They match objects across input modes and also account for mismatches. However, this visual-linguistic grounding ability heavily varies between object classes, depends on the training data distribution, and largely improves after in-domain training. Using our method we can identify knowledge gaps about specific object classes in individual models and can monitor their improvement upon fine-tuning.