Search

Uncertainty quantification is not yet widely adapted in the design process of engineering components despite its importance for achieving sustainable and resource-efficient structures. This is mainly due to two reasons: 1) Tracing the effect of uncertainty in engineering simulations is a computationally challenging task. This is especially true for inelastic simulations as the whole loading history influences the results. 2) Implementations of efficient schemes in standard finite element software are lacking. In this paper, we are tackling both problems. We are proposing a \rev{weakly}-intrusive implementation of the time-separated stochastic mechanics in the finite element software Abaqus. The time-separated stochastic mechanics is an efficient and accurate method for the uncertainty quantification of structures with inelastic material behavior. The method effectivly separates the stochastic but time-independent from the deterministic but time-dependent behavior. The resulting scheme consists only two deterministic finite element simulations for homogeneous material fluctuations in order to approximate the stochastic behavior. This brings down the computational cost compared to standard Monte Carlo simulations by at least two orders of magnitude while ensuring accurate solutions. In this paper, the implementation details in Abaqus and numerical comparisons are presented for the example of damage simulations.

We explore the structural and electronic properties of the bilayer nickelate La3Ni2O7 on LaAlO3(001) and SrTiO3(001) by using density functional theory including a Coulomb repulsion term. For La$_3$Ni$_2$O$_{7}$/LaAlO$_3$(001), we find that compressive strain and electron doping across the interface result in the unconventional occupation of the antibonding Ni $3d_{z^2}$ states. In sharp contrast, no charge transfer is observed for La$_3$Ni$_2$O$_{7}$/SrTiO$_3$(001). Surprisingly, tensile strain drives a metallization of the bonding Ni $3d_{z^2}$ states, rendering a Fermi surface topology resembling the superconducting bulk La$_3$Ni$_2$O$_{7}$ under high pressure. Concomitantly, significant octahedral rotations are retained. We discuss the fundamental differences between hydrostatic pressure and epitaxial strain regarding the crystal geometry and Ni $3d$ orbital occupation, and establish that strain provides a much stronger control over the Ni $e_g$ orbital polarization. The results suggest epitaxial La$_3$Ni$_2$O$_{7}$ under tensile strain as interesting system to provide novel insights into the physics of bilayer nickelates and possibly induce superconductivity without external pressure., Comment: 7 pages, 3 figures

We describe an algorithm to compute the stable multiplicity of a family of irreducible representations in the cohomology of ordered configuration space of the plane. Using this algorithm, we compute the stable multiplicities of all families of irreducibles given by Young diagrams with $23$ boxes or less up to cohomological degree $50$. In particular, this determines the stable cohomology in cohomological degrees $0 \leq i \leq 11$. We prove related qualitative results and formulate some conjectures., Comment: 16 pages, 0 figures

Many applications in traffic, civil engineering, or electrical engineering revolve around edge-level signals. Such signals can be categorized as inherently directed, for example, the water flow in a pipe network, and undirected, like the diameter of a pipe. Topological methods model edge signals with inherent direction by representing them relative to a so-called orientation assigned to each edge. These approaches can neither model undirected edge signals nor distinguish if an edge itself is directed or undirected. We address these shortcomings by (i) revising the notion of orientation equivariance to enable edge direction-aware topological models, (ii) proposing orientation invariance as an additional requirement to describe signals without inherent direction, and (iii) developing EIGN, an architecture composed of novel direction-aware edge-level graph shift operators, that provably fulfills the aforementioned desiderata. It is the first general-purpose topological GNN for edge-level signals that can model directed and undirected signals while distinguishing between directed and undirected edges. A comprehensive evaluation shows that EIGN outperforms prior work in edge-level tasks, for example, improving in RMSE on flow simulation tasks by up to 43.5%.

CMOS-based microelectronics are limited to ~150{\deg}C and therefore not suitable for the extreme high temperatures in aerospace, energy, and space applications. While wide bandgap semiconductors can provide high-temperature logic, nonvolatile memory devices at high temperatures have been challenging. In this work, we develop a nonvolatile electrochemical memory cell that stores and retains analog and digital information at temperatures as high as 600 {\deg}C. Through correlative electron microscopy, we show that this high-temperature information retention is a result of composition phase separation between the oxidized and reduced forms of amorphous tantalum oxide. This result demonstrates a memory concept that is resilient at extreme temperatures and reveals phase separation as the principal mechanism that enables nonvolatile information storage in these electrochemical memory cells., Comment: 19 pages, 4 figures

Stars presently identified in the bulge spheroid are probably very old, and their abundances can be interpreted as due to the fast chemical enrichment of the early Galactic bulge. The abundances of the iron-peak elements are important tracers of nucleosynthesis processes, in particular oxygen burning, silicon burning, the weak s-process, and alpha-rich freeze-out. Aims. The aim of this work is to derive the abundances of V, Cr, Mn, Co, Ni, and Cu in 58 bulge spheroid stars and to compare them with the results of a previous analysis of data from APOGEE. We selected the best lines for V, Cr, Mn, Co, Ni, and Cu located within the H-band of the spectrum, identifying the most suitable ones for abundance determination, and discarding severe blends. Using the stellar physical parameters available for our sample from the DR17 release of the APOGEE project, we derived the individual abundances through spectrum synthesis. We then complemented these measurements with similar results from different bulge field and globular cluster stars, in order to define the trends of the individual elements and compare with the results of chemical-evolution models. We verify that the H-band has useful lines for the derivation of the elements V, Cr, Mn, Co, Ni, and Cu in moderately metal-poor stars. The resulting abundances indicate that: V, Cr, and Ni vary in lockstep with Fe; Co tends to vary in lockstep with Fe, but could be showing a slight decrease with decreasing metallicity; and Mn and Cu decrease with decreasing metallicity. These behaviours are well reproduced by chemical-evolution models except for Cu, which appears to drop faster than the models predict for moderate metallicities. Finally, abundance indicators combined with kinematical and dynamical criteria appear to show that our 58 sample stars are likely to have originated in situ., Comment: 12 pages, Astronomy & Astrophysics, accepted on 9/October/2024

Purpose: Segmentation of the breast region in dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is essential for the automatic measurement of breast density and the quantitative analysis of imaging findings. This study aims to compare various deep learning methods to enhance whole breast segmentation and reduce computational costs as well as environmental effect for future research. Methods: We collected fifty-nine DCE-MRI scans from Stavanger University Hospital and, after preprocessing, analyzed fifty-eight scans. The preprocessing steps involved standardizing imaging protocols and resampling slices to ensure consistent volume across all patients. Using our novel approach, we defined new breast boundaries and generated corresponding segmentation masks. We evaluated seven deep learning models for segmentation namely UNet, UNet++, DenseNet, FCNResNet50, FCNResNet101, DeepLabv3ResNet50, and DeepLabv3ResNet101. To ensure robust model validation, we employed 10-fold cross-validation, dividing the dataset into ten subsets, training on nine, and validating on the remaining one, rotating this process to use all subsets for validation. Results: The models demonstrated significant potential across multiple metrics. UNet++ achieved the highest performance in Dice score, while UNet excelled in validation and generalizability. FCNResNet50, notable for its lower carbon footprint and reasonable inference time, emerged as a robust model following UNet++. In boundary detection, both UNet and UNet++ outperformed other models, with DeepLabv3ResNet also delivering competitive results.

Let $d,n\in \mathbb{N}$ be such that $d=\omega(1)$, and $d\le n^{1-a}$ for some constant $a>0$. Consider a $d$-regular graph $G=(V, E)$ and the random graph process that starts with the empty graph $G(0)$ and at each step $G(i)$ is obtained from $G(i-1)$ by adding uniformly at random a new edge from $E$. We show that if $G$ satisfies some (very) mild global edge-expansion, and an almost optimal edge-expansion of sets up to order $O(d\log n)$, then for any constant $k\in \mathbb{N}$ in the random graph process on $G$, typically the hitting times of minimum degree at least $k$ and of $k$-connectedness are equal. This, in particular, covers both $d$-regular high dimensional product graphs and pseudo-random graphs, and confirms a conjecture of Joos from 2015. We further demonstrate that this result is tight in the sense that there are $d$-regular $n$-vertex graphs with optimal edge-expansion of sets up to order $\Omega(d)$, for which the probability threshold of minimum degree at least one is different than the probability threshold of connectivity., Comment: 11 pages

Visual search is a fundamental natural task for humans and other animals. We investigated the decision processes humans use when searching briefly presented displays having well-separated potential target-object locations. Performance was compared with the Bayesian-optimal decision process under the assumption that the information from the different potential target locations is statistically independent. Surprisingly, humans performed slightly better than optimal, despite humans' substantial loss of sensitivity in the fovea, and the implausibility of the human brain replicating the optimal computations. We show that three factors can quantitatively explain these seemingly paradoxical results. Most importantly, simple and fixed heuristic decision rules reach near optimal search performance. Secondly, foveal neglect primarily affects only the central potential target location. Finally, spatially correlated neural noise causes search performance to exceed that predicted for independent noise. These findings have far-reaching implications for understanding visual search tasks and other identification tasks in humans and other animals.

The usage of quick response (QR) codes was limited in the pre-era of the COVID-19 pandemic. Due to the widespread and frequent application since then, this opened up an attractive phishing opportunity for malicious actors. They trick users into scanning the codes and redirecting them to malicious websites. In order to explore whether phishing with QR codes is another successful attack vector, we conducted a real-world phishing campaign with two different QR code variants at a research campus. The first version was rather plain, whereas the second version was more professionally designed and included the possibility to win a voucher. After the study was completed, a qualitative survey on phishing and QR codes was conducted to verify the results of the phishing campaign. Both, the phishing campaign and the survey, show that a professional design receives more attention. They also illustrate that QR codes are used more frequently by curious users because of their easy functionality. Although the results confirm that technical-savvy users are more aware of the risks, they also underpin the malicious potential for non-technical-savvy users and suggest further work regarding countermeasures.

Over the past decade, adversarial training has emerged as one of the few reliable methods for enhancing model robustness against adversarial attacks [Szegedy et al., 2014, Madry et al., 2018, Xhonneux et al., 2024], while many alternative approaches have failed to withstand rigorous subsequent evaluations. Recently, an alternative defense mechanism, namely "circuit breakers" [Zou et al., 2024], has shown promising results for aligning LLMs. In this report, we show that the robustness claims of "Improving Alignment and Robustness with Circuit Breakers" against unconstraint continuous attacks in the embedding space of the input tokens may be overestimated [Zou et al., 2024]. Specifically, we demonstrate that by implementing a few simple changes to embedding space attacks [Schwinn et al., 2024a,b], we achieve 100% attack success rate (ASR) against circuit breaker models. Without conducting any further hyperparameter tuning, these adjustments increase the ASR by more than 80% compared to the original evaluation. Code is accessible at: https://github.com/SchwinnL/circuit-breakers-eval

Existing studies have shown that Graph Neural Networks (GNNs) are vulnerable to adversarial attacks. Even though Graph Transformers (GTs) surpassed Message-Passing GNNs on several benchmarks, their adversarial robustness properties are unexplored. However, attacking GTs is challenging due to their Positional Encodings (PEs) and special attention mechanisms which can be difficult to differentiate. We overcome these challenges by targeting three representative architectures based on (1) random-walk PEs, (2) pair-wise-shortest-path PEs, and (3) spectral PEs - and propose the first adaptive attacks for GTs. We leverage our attacks to evaluate robustness to (a) structure perturbations on node classification; and (b) node injection attacks for (fake-news) graph classification. Our evaluation reveals that they can be catastrophically fragile and underlines our work's importance and the necessity for adaptive attacks.

Majority bootstrap percolation is a monotone cellular automata that can be thought of as a model of infection spreading in networks. Starting with an initially infected set, new vertices become infected once more than half of their neighbours are infected. The average case behaviour of this process was studied on the $n$-dimensional hypercube by Balogh, Bollob\'{a}s and Morris, who showed that there is a phase transition as the typical density of the initially infected set increases: For small enough densities the spread of infection is typically local, whereas for large enough densities typically the whole graph eventually becomes infected. Perhaps surprisingly, they showed that the critical window in which this phase transition occurs is bounded away from $1/2$, and they gave bounds on its width on a finer scale. In this paper we consider the majority bootstrap percolation process on a class of high-dimensional geometric graphs which includes many of the graph families on which percolation processes are typically considered, such as grids, tori and Hamming graphs, as well as other well-studied families of graphs such as (bipartite) Kneser graphs, including the odd graph and the middle layer graph. We show similar quantitative behaviour in terms of the location and width of the critical window for the majority bootstrap percolation process on this class of graphs., Comment: 36 pages

The ESO public survey VISTA Variables in the V\'ia L\'actea (VVV) surveyed the inner Galactic bulge and the adjacent southern Galactic disk from $2009-2015$. Upon its conclusion, the complementary VVV eXtended (VVVX) survey has expanded both the temporal as well as spatial coverage of the original VVV area, widening it from $562$ to $1700$ sq. deg., as well as providing additional epochs in $JHK_{\rm s}$ filters from $2016-2023$. With the completion of VVVX observations during the first semester of 2023, we present here the observing strategy, a description of data quality and access, and the legacy of VVVX. VVVX took $\sim 2000$ hours, covering about 4% of the sky in the bulge and southern disk. VVVX covered most of the gaps left between the VVV and the VISTA Hemisphere Survey (VHS) areas and extended the VVV time baseline in the obscured regions affected by high extinction and hence hidden from optical observations. VVVX provides a deep $JHK_{\rm s}$ catalogue of $\gtrsim 1.5\times10^9$ point sources, as well as a $K_{\rm s}$ band catalogue of $\sim 10^7$ variable sources. Within the existing VVV area, we produced a $5D$ map of the surveyed region by combining positions, distances, and proper motions of well-understood distance indicators such as red clump stars, RR Lyrae, and Cepheid variables. In March 2023 we successfully finished the VVVX survey observations that started in 2016, an accomplishment for ESO Paranal Observatory upon 4200 hours of observations for VVV+VVVX. The VVV+VVVX catalogues complement those from the Gaia mission at low Galactic latitudes and provide spectroscopic targets for the forthcoming ESO high-multiplex spectrographs MOONS and 4MOST., Comment: 17 pages, 11 figures (+ appendix). Accepted for publication in Astronomy and Astrophysics in section 14: Catalogs and data

Predictions made by graph neural networks (GNNs) usually lack interpretability due to their complex computational behavior and the abstract nature of graphs. In an attempt to tackle this, many GNN explanation methods have emerged. Their goal is to explain a model's predictions and thereby obtain trust when GNN models are deployed in decision critical applications. Most GNN explanation methods work in a post-hoc manner and provide explanations in the form of a small subset of important edges and/or nodes. In this paper we demonstrate that these explanations can unfortunately not be trusted, as common GNN explanation methods turn out to be highly susceptible to adversarial perturbations. That is, even small perturbations of the original graph structure that preserve the model's predictions may yield drastically different explanations. This calls into question the trustworthiness and practical utility of post-hoc explanation methods for GNNs. To be able to attack GNN explanation models, we devise a novel attack method dubbed \textit{GXAttack}, the first \textit{optimization-based} adversarial white-box attack method for post-hoc GNN explanations under such settings. Due to the devastating effectiveness of our attack, we call for an adversarial evaluation of future GNN explainers to demonstrate their robustness. For reproducibility, our code is available via GitHub.

Spatial Message Passing Graph Neural Networks (MPGNNs) are widely used for learning on graph-structured data. However, key limitations of l-step MPGNNs are that their "receptive field" is typically limited to the l-hop neighborhood of a node and that information exchange between distant nodes is limited by over-squashing. Motivated by these limitations, we propose Spatio-Spectral Graph Neural Networks (S$^2$GNNs) -- a new modeling paradigm for Graph Neural Networks (GNNs) that synergistically combines spatially and spectrally parametrized graph filters. Parameterizing filters partially in the frequency domain enables global yet efficient information propagation. We show that S$^2$GNNs vanquish over-squashing and yield strictly tighter approximation-theoretic error bounds than MPGNNs. Further, rethinking graph convolutions at a fundamental level unlocks new design spaces. For example, S$^2$GNNs allow for free positional encodings that make them strictly more expressive than the 1-Weisfeiler-Lehman (WL) test. Moreover, to obtain general-purpose S$^2$GNNs, we propose spectrally parametrized filters for directed graphs. S$^2$GNNs outperform spatial MPGNNs, graph transformers, and graph rewirings, e.g., on the peptide long-range benchmark tasks, and are competitive with state-of-the-art sequence modeling. On a 40 GB GPU, S$^2$GNNs scale to millions of nodes., Comment: 46 pages, 27 figures, 12 tables

The scale-up of autonomous vehicles depends heavily on their ability to deal with anomalies, such as rare objects on the road. In order to handle such situations, it is necessary to detect anomalies in the first place. Anomaly detection for autonomous driving has made great progress in the past years but suffers from poorly designed benchmarks with a strong focus on camera data. In this work, we propose AnoVox, the largest benchmark for ANOmaly detection in autonomous driving to date. AnoVox incorporates large-scale multimodal sensor data and spatial VOXel ground truth, allowing for the comparison of methods independent of their used sensor. We propose a formal definition of normality and provide a compliant training dataset. AnoVox is the first benchmark to contain both content and temporal anomalies., Comment: Daniel Bogdoll, Iramm Hamdard, and Lukas Namgyu R\"o{\ss}ler contributed equally. Accepted for publication at ECCV 2024 W-CODA workshop

This paper presents a large 6~m x 3~m aperture 7200 MEMS microphone array. The array is designed so that sub-arrays with optimized point spread functions can be used for beamforming and thus, enable the research of source directivity in wind tunnel facilities. The total array consists of modular 800 microphone panels, each consisting of four unique PCB board designs. This modular architecture allows for the time-synchronized measurement of an arbitrary number of panels and thus, aperture size and total number of sensors. The panels can be installed without a gap so that the array's microphone pattern avoids high sidelobes in the point spread function. The array's capabilities are evaluated on a 1:9.5 airframe half model in an open wind tunnel at DNW-NWB. The total source emission is quantified and the directivity is evaluated with beamforming. Additional far-field microphones are employed to validate the results., Comment: 30th AIAA/CEAS Aeroacoustics Conference

For $t \in \mathbb{N}$ and every $i\in[t]$, let $H_i$ be a $d_i$-regular connected graph, with $1<|V(H_i)|\le C$ for some integer $C\ge 2$. Let $G=\square_{i=1}^tH_i$ be the Cartesian product of $H_1, \ldots, H_t$. We show that if $t\ge 5C\log_2C$ then $G$ contains a (nearly-)perfect matching. Then, considering the random graph process on $G$, we generalise the result of Bollob\'as on the binary hypercube $Q^t$, showing that with high probability, the hitting times for minimum degree one, connectivity, and the existence of a (nearly-)perfect matching in the $G$-random-process are the same. We develop several tools which may be of independent interest in a more general setting of the typical existence of a perfect matching under percolation., Comment: Minor corrections

Dataspaces have recently gained adoption across various sectors, including traditionally less digitized domains such as culture. Leveraging Semantic Web technologies helps to make dataspaces FAIR, but their complexity poses a significant challenge to the adoption of dataspaces and increases their cost. The advent of Large Language Models (LLMs) raises the question of how these models can support the adoption of FAIR dataspaces. In this work, we demonstrate the potential of LLMs in dataspaces with a concrete example. We also derive a research agenda for exploring this emerging field., Comment: 8 pages. Preprint. Under review

Current LLM alignment methods are readily broken through specifically crafted adversarial prompts. While crafting adversarial prompts using discrete optimization is highly effective, such attacks typically use more than 100,000 LLM calls. This high computational cost makes them unsuitable for, e.g., quantitative analyses and adversarial training. To remedy this, we revisit Projected Gradient Descent (PGD) on the continuously relaxed input prompt. Although previous attempts with ordinary gradient-based attacks largely failed, we show that carefully controlling the error introduced by the continuous relaxation tremendously boosts their efficacy. Our PGD for LLMs is up to one order of magnitude faster than state-of-the-art discrete optimization to achieve the same devastating attack results.

In the present work we revisit the shock wave dynamics in a granular chain with precompression. By approximating the model by an $\alpha$-Fermi-Pasta-Ulam-Tsingou chain, we leverage the connection of the latter in the strain variable formulation to two separate integrable models, one continuum, namely the KdV equation, and one discrete, namely the Toda lattice. We bring to bear the Whitham modulation theory analysis of such integrable systems and the analytical approximation of their dispersive shock waves in order to provide, through the lens of the reductive connection to the granular crystal, an approximation to the shock wave of the granular problem. A detailed numerical comparison of the original granular chain and its approximate integrable-system based dispersive shocks proves very favorable in a wide parametric range. The gradual deviations between (approximate) theory and numerical computation, as amplitude parameters of the solution increase are quantified and discussed., Comment: 12 pages, 4 figures

Star clusters were historically considered simple stellar populations, with all stars sharing the same age and initial chemical composition. However, the presence of chemical anomalies in globular clusters (GCs), called multiple stellar populations (MPs), has challenged star formation theories in dense environments. Literature studies show that mass, metallicity, and age are likely controlling parameters for the manifestation of MPs. Identifying the limit between clusters with/without MPs in physical parameter space is crucial to reveal the driving mechanism behind their presence. In this study, we look for MP signals in Whiting 1, traditionally considered a young GC. Using the Magellan telescope, we obtained low-resolution spectra within $\rm \lambda\lambda = 3850-5500 \r{A}$ for eight giants of Whiting 1. We measured the C and N abundances from the CN and CH spectral indices. C and N abundances have variations comparable with their measurement errors ($\sim0.1$ dex), suggesting that MPs are absent from Whiting 1. Combining these findings with literature studies, we propose a limit in the metallicity vs. cluster compactness index parameter space, which relatively clearly separates star clusters with/without MPs (GCs/open clusters). This limit is physically motivated. On a larger scale, the galactic environment determines cluster compactness and metallicity, leading to metal-rich, diffuse, old clusters formed ex situ. Our proposed limit also impacts our understanding of the formation of the Sagittarius dwarf galaxy: star clusters formed after the first starburst (age$\lesssim 8-10$ Gyr). These clusters are simple stellar populations because the enriched galactic environment is no longer suitable for MP formation., Comment: 11 pages, 4 figures, SCPMA accepted

Integrating deep learning with the search for new electron-phonon superconductors represents a burgeoning field of research, where the primary challenge lies in the computational intensity of calculating the electron-phonon spectral function, $\alpha^2F(\omega)$, the essential ingredient of Midgal-Eliashberg theory of superconductivity. To overcome this challenge, we adopt a two-step approach. First, we compute $\alpha^2F(\omega)$ for 818 dynamically stable materials. We then train a deep-learning model to predict $\alpha^2F(\omega)$, using an unconventional training strategy to temper the model's overfitting, enhancing predictions. Specifically, we train a Bootstrapped Ensemble of Tempered Equivariant graph neural NETworks (BETE-NET), obtaining an MAE of 0.21, 45 K, and 43 K for the Eliashberg moments derived from $\alpha^2F(\omega)$: $\lambda$, $\omega_{\log}$, and $\omega_{2}$, respectively, yielding an MAE of 2.5 K for the critical temperature, $T_c$. Further, we incorporate domain knowledge of the site-projected phonon density of states to impose inductive bias into the model's node attributes and enhance predictions. This methodological innovation decreases the MAE to 0.18, 29 K, and 28 K, respectively, yielding an MAE of 2.1 K for $T_c$. We illustrate the practical application of our model in high-throughput screening for high-$T_c$ materials. The model demonstrates an average precision nearly five times higher than random screening, highlighting the potential of ML in accelerating superconductor discovery. BETE-NET accelerates the search for high-$T_c$ superconductors while setting a precedent for applying ML in materials discovery, particularly when data is limited., Comment: 12 pages, 5 figures, 1 table

We explore the optical properties of La3Ni2O7 bilayer nickelates by using density functional theory including a Coulomb repulsion term. Convincing agreement with recent experimental ambient-pressure spectra is achieved for U=3eV, which permits tracing the microscopic origin of the characteristic features. Simultaneous consistency with angle-resolved photoemission spectroscopy and x-ray diffraction suggests the notion of rather moderate electronic correlations in this novel high-Tc superconductor. Oxygen vacancies form predominantly at the inner apical sites and renormalize the optical spectrum quantitatively, while the released electrons are largely accommodated by a defect state. We show that the structural transition occurring under high pressure coincides with a significant enhancement of the Drude weight and a reduction of the out-of-plane interband contribution that act as a fingerprint of the emerging hole pocket. We further calculate the optical spectra for various possible magnetic phases including spin-density waves and discuss the results in the context of experiment. Finally, we investigate the role of the 2-2 versus 1-3 layer stacking and compare the bilayer nickelate to La4Ni3O10, La3Ni2O6, and NdNiO2, unveiling general trends in the optical spectrum as a function of the formal Ni valence in Ruddlesden-Popper versus reduced Ruddlesden-Popper nickelates., Comment: 11 pages, 5 figures

It is well-known that deep learning models are vulnerable to small input perturbations. Such perturbed instances are called adversarial examples. Adversarial examples are commonly crafted to fool a model either at training time (poisoning) or test time (evasion). In this work, we study the symbiosis of poisoning and evasion. We show that combining both threat models can substantially improve the devastating efficacy of adversarial attacks. Specifically, we study the robustness of Graph Neural Networks (GNNs) under structure perturbations and devise a memory-efficient adaptive end-to-end attack for the novel threat model using first-order optimization., Comment: NeurIPS 2023 New Frontiers in Graph Learning Workshop (NeurIPS GLFrontiers 2023)