Search

Large Language Models (LLMs) contain large amounts of facts about the world. These facts can become outdated over time, which has led to the development of knowledge editing methods (KEs) that can change specific facts in LLMs with limited side effects. This position paper argues that editing LLMs poses serious safety risks that have been largely overlooked. First, we note the fact that KEs are widely available, computationally inexpensive, highly performant, and stealthy makes them an attractive tool for malicious actors. Second, we discuss malicious use cases of KEs, showing how KEs can be easily adapted for a variety of malicious purposes. Third, we highlight vulnerabilities in the AI ecosystem that allow unrestricted uploading and downloading of updated models without verification. Fourth, we argue that a lack of social and institutional awareness exacerbates this risk, and discuss the implications for different stakeholders. We call on the community to (i) research tamper-resistant models and countermeasures against malicious model editing, and (ii) actively engage in securing the AI ecosystem.

Purpose: Measuring the ortho-positronium (oPs) lifetime in human tissue bears the potential of adding clinically relevant information about the tissue microenvironment to conventional positron emission tomography (PET). Through phantom measurements, we investigate the voxel-wise measurement of oPs lifetime using a commercial long-axial field-of-view (LAFOV) PET scanner. Methods: We prepared four samples with mixtures of Amberlite XAD4, a porous polymeric adsorbent, and water and added between 1.12 MBq and 1.44 MBq of $^{124}$I. The samples were scanned in two different setups: once with a couple of centimeters between each sample (15 minutes scan time) and once with all samples taped together (40 minutes scan time). For each scan, we determine the oPs lifetime for the full samples and at the voxel level. The voxel sizes under consideration are $10.0^3$ mm$^3$, $7.1^3$ mm$^3$ and $4.0^3$ mm$^3$. Results: Amberlite XAD4 allows the preparation of samples with distinct oPs lifetime. Using a Bayesian fitting procedure, the oPs lifetimes in the whole samples are $2.52 \pm 0.03$ ns, $2.37\pm 0.03$ ns, $2.27\pm0.04$ ns and $1.82\pm 0.02$ ns, respectively. The voxel-wise oPs lifetime fits showed that even with $4.0^3$ mm$^3$ voxels the samples are clearly distinguishable and a central voxels have good count statistics. However, the situation with the samples close together remains challenging with respect to the spatial distinction of regions with different oPs lifetimes. Conclusion: Our study shows that positronium lifetime imaging on a commercial LAFOV PET/CT should be feasible under clinical conditions using $^{124}$I.

Shortcut learning, i.e., a model's reliance on undesired features not directly relevant to the task, is a major challenge that severely limits the applications of machine learning algorithms, particularly when deploying them to assist in making sensitive decisions, such as in medical diagnostics. In this work, we leverage recent advancements in machine learning to create an unsupervised framework that is capable of both detecting and mitigating shortcut learning in transformers. We validate our method on multiple datasets. Results demonstrate that our framework significantly improves both worst-group accuracy (samples misclassified due to shortcuts) and average accuracy, while minimizing human annotation effort. Moreover, we demonstrate that the detected shortcuts are meaningful and informative to human experts, and that our framework is computationally efficient, allowing it to be run on consumer hardware.

A major goal of stochastic thermodynamics is to estimate the inevitable dissipation that accompanies particular observable phenomena in an otherwise not fully accessible system. Quantitative results are often formulated as lower bounds on the total entropy production, which capture the part of the total dissipation that can be determined based on the available data alone. In this work, we discuss the case of a continuous-time dynamics on a Markov network that is observed stroboscopically, i.e., at discrete points in time in regular intervals. We compare the standard approach of deriving a lower bound on the entropy production rate in the steady state to the less common method of reconstructing the generator from the observed propagators by taking the matrix logarithm. Provided that the timescale of the stroboscopic measurements is smaller than a critical value that can be determined from the available data, this latter method is able to recover all thermodynamic quantities like entropy production or cycle affinities and is therefore superior to the usual approach of deriving lower bounds. Beyond the critical value, we still obtain tight upper and lower bounds on these quantities that improve on extant methods. We conclude the comparison with numerical illustrations and a discussion of the requirements and limitations of both methods.

We present a new, cosmologically model-independent, statistical analysis of the Pantheon+ type Ia supernovae spectroscopic dataset, improving a standard methodology adopted by Lane et al. We use the Tripp equation for supernova standardisation alone, thereby avoiding any potential correlation in the stretch and colour distributions. We compare the standard homogeneous cosmological model, i.e., $\Lambda$CDM, and the timescape cosmology which invokes backreaction of inhomogeneities. Timescape, while statistically homogeneous and isotropic, departs from average Friedmann-Lema\^{\i}tre-Robertson-Walker evolution, and replaces dark energy by kinetic gravitational energy and its gradients, in explaining independent cosmological observations. When considering the entire Pantheon+ sample, we find very strong evidence ($\ln B> 5$) in favour of timescape over $\Lambda$CDM. Furthermore, even restricting the sample to redshifts beyond any conventional scale of statistical homogeneity, $z > 0.075$, timescape is preferred over $\Lambda$CDM with $\ln B> 1$. These results provide evidence for a need to revisit the foundations of theoretical and observational cosmology., Comment: 6 pages, 3 figures, 1 table; associated RAS press release RAS PR 24/33

For strongly continous semigroups on Hilbert spaces, we investigate admissibility properties of control and observation operators shifted along continuous scales of spaces built by means of either interpolation and extrapolation or functional calculus. Our results show equivalence of admissibility in, on the one hand, a fractional domain of the generator and, on the other hand, a (different, in general) quadratic interpolation space of the same "Sobolev order". Furthermore, such properties imply quantified resolvent bounds in the original state space topology. When the semigroup is a group, the resulting frequency-domain estimates are in fact equivalent to the aforementioned time-domain properties. In the case of systems with both control and observation, we are able to translate input-output regularity properties into high-frequency growth rates of operator-valued transfer functions. As an application, based on results by Lasiecka, Triggiani and Tataru on interior and boundary regularity of the wave equation under Neumann control, we derive optimal asymptotics for the Neumann-to-Dirichlet wave transfer function. With that in hand, we establish non-uniform energy decay rates for the wave equation posed in a rectangle and subject to Neumann damping on an arbitrary open subset of the boundary.

Multiplicity studies greatly benefit from focusing on M dwarfs because they are often paired in a variety of configurations with both stellar and substellar objects, including exoplanets. We aim to address the observed multiplicity of M dwarfs by conducting a systematic analysis using the latest available astrophotometric data. For every star in a sample of 2214 M dwarfs from the CARMENES catalogue, we investigated the existence of resolved and unresolved physical companions in the literature and in all-sky surveys, especially in Gaia DR3 data products. We covered a very wide range of separations, from known spectroscopic binaries in tight arrangements $\sim$0.01 au to remarkably separated ultra-wide pairs ($\sim$10$^5$ au). We identified 835 M dwarfs in 720 multiple systems, predominantly binaries. Thus, we propose 327 new binary candidates based on Gaia data. If these candidates are finally confirmed, we expect the multiplicity fraction of M dwarfs to be 40.3$^{+2.1}_{-2.0}$ %. When only considering the systems already identified, the multiplicity fraction is reduced to 27.8$^{+1.9}_{-1.8}$ %. This result is in line with most of the values published in the literature. We also identified M-dwarf multiple systems with FGK, white dwarf, ultra-cool dwarf, and exoplanet companions, as well as those in young stellar kinematic groups. We studied their physical separations, orbital periods, binding energies, and mass ratios. We argue that based on reliable astrometric data and spectroscopic investigations from the literature (even when considering detection biases), the multiplicity fraction of M dwarfs could still be significantly underestimated. This calls for further high-resolution follow-up studies to validate these findings.

Estimating entropy production from real observation data can be difficult due to finite resolution in both space and time and finite measurement statistics. We characterize the statistical error introduced by finite sample size and compare the performance of three different entropy estimators under these limitations for two different paradigmatic systems, a four-state Markov network and an augmented Michaelis-Menten reaction scheme. We consider the thermodynamic uncertainty relation, a waiting-time based estimator for resolved transitions and a waiting-time based estimator for blurred transitions in imperfect observation scenarios. For perfect measurement statistics and finite temporal resolution, the estimator based on resolved transitions performs best in all considered scenarios. The thermodynamic uncertainty relation gives a better estimate than the estimator based on blurred transitions at low driving affinities, whereas the latter performs better at high driving affinities. Furthermore, we find that a higher temporal and spatial resolution leads to slower convergence of measurement statistics, implying that for short measurement times, a lower resolution may be beneficial. Additionally, we identify a self-averaging effect for the waiting-time based entropy estimators that can reduce their variance for observations with finite statistics.

In the era of Extremely Large Telescopes, the current generation of 8-10m facilities are likely to remain competitive at ground-UV wavelengths for the foreseeable future. The Cassegrain U-Band Efficient Spectrograph (CUBES) has been designed to provide high instrumental efficiency ( $>$ 37\%) observations in the near UV (305-400 nm requirement, 300-420 nm goal) at a spectral resolving power of R $>$ 20, 000 (with a lower-resolution, sky-limited mode of R $\sim$ 7, 000). With the design focusing on maximizing the instrument throughput (ensuring a Signal to Noise Ratio -SNR- $\sim$ 20 per spectral resolution element at 313 nm for U $\sim$ 17.5 mag objects in 1h of observations), it will offer new possibilities in many fields of astrophysics: i) access to key lines of stellar spectra (e.g. lighter elements, in particular Beryllium), extragalactic studies (e.g. circumgalactic medium of distant galaxies, cosmic UV background) and follow-up of explosive transients. We present the CUBES instrument design, currently in Phase-C and approaching the final design review, summarizing the hardware architecture and interfaces between the different subsystems as well as the relevant technical requirements. We describe the optical, mechanical, electrical design of the different subsystems (from the telescope adapter and support structure, through the main opto-mechanical path, including calibration unit, detector devices and cryostat control, main control electronics), detailing peculiar instrument functions like the Active Flexure Compensation (AFC). Furthermore, we outline the AITV concept and the main instrument operations giving an overview of its software ecosystem. Installation at the VLT is planned for 2028-2029 and first science operations in late 2029.

We investigate the adsorption of molecular hydrogen on pristine zinc oxide (ZnO) platelets. The volumetric and gravimetric hydrogen storage capacities of the ZnO monolayers are evaluated in a broad range of thermodynamic conditions (i.e., for temperatures in the range 77 K < T < 450 K, and for external gas pressures up to 200 bar). The thermodynamic properties and the microscopic spatial distribution of the adsorbed hydrogen fluid are assessed within the density functional theory of liquids for quantum fluids at finite temperature (QLDFT), and the adsorption enthalphies are obtained by fitting the computed adsorption densities to the Toth model isotherm. Compared to graphene platelets, the ZnO sheets impose a rather tighter confinement to the motion of the hydrogen molecules parallel to the surface. The isosteric heat of adsorption approaches 3.2 kJ/mol in the low density regime. This quantity shows a fairly smooth dependence on the hydrogen uptake for temperatures below 100 K, while it is shown to depend quite sensitively on the adsorbate density above this temperature., Comment: 28 pages, 8 figures

One of the core challenges of research in quantum computing is concerned with the question whether quantum advantages can be found for near-term quantum circuits that have implications for practical applications. Motivated by this mindset, in this work, we prove an unconditional quantum advantage in the probably approximately correct (PAC) distribution learning framework with shallow quantum circuit hypotheses. We identify a meaningful generative distribution learning problem where constant-depth quantum circuits using one and two qubit gates (QNC^0) are superior compared to constant-depth bounded fan-in classical circuits (NC^0) as a choice for hypothesis classes. We hence prove a PAC distribution learning separation for shallow quantum circuits over shallow classical circuits. We do so by building on recent results by Bene Watts and Parham on unconditional quantum advantages for sampling tasks with shallow circuits, which we technically uplift to a hyperplane learning problem, identifying non-local correlations as the origin of the quantum advantage., Comment: 7 + 5 pages, 2 figures, added an acknowledgement

We study evolutionary equations in exponentially weighted $\mathrm{L}^{2}$-spaces as introduced by Picard in 2009. First, for a given evolutionary equation, we explicitly describe the $\nu$-adjoint system, which turns out to describe a system backwards in time. We prove well-posedness for the $\nu$-adjoint system. We then apply the thus obtained duality to introduce and study notions of null-controllability for evolutionary equations., Comment: 19 pages

Antiferromagnets exhibit rapid spin dynamics in a net zero magnetic background which enables novel spintronic applications and interrogation of many-body quantum phenomena. The layered antiferromagnet Sr$_2$IrO$_4$ hosts an exotic spin one-half Mott insulating state with an electronic gap arising from on-site Coulomb repulsion and strong spin-orbit coupling. This makes Sr$_2$IrO$_4$ an interesting candidate to interrogate dynamical attributes of the magnetic order using ultrafast laser pulses. We investigate the magnetization dynamics of Sr$_2$IrO$_4$ following circularly-polarized photoexcitation with below-gap mid-infrared (mid-IR -- 9 $\mu m$) and above-gap near-infrared (near-IR -- 1.3 $\mu m$) pulses. In both cases, we observe excitation of a zone-center coherent magnon mode featuring a 0.5 THz oscillation in the pump-induced Kerr-rotation signal. However, only below-gap excitation exhibits a helicity dependent response and linear (quadratic) scaling of the coherent magnon amplitude with excitation fluence (electric field). Moreover, below-gap excitation has a magnon generation efficiency that is at least two orders of magnitude greater in comparison to above-gap excitation. Our analysis indicates that the helicity dependence and enhanced generation efficiency arises from a unique one-photon two-magnon coupling mechanism for magnon generation. Thus, preferential spin-photon coupling without photoexcitation of electrons permits extremely efficient magnon generation. Our results reveal new possibilities for ultrafast control of antiferromagnets.

Aims. Our goals are to characterise the kinematic properties and to identify young and old stars among the M dwarfs of the CARMENES input catalogue. Methods. We compiled the spectral types, proper motions, distances, and radial velocities for 2187 M dwarfs. We used the public code SteParKin to derive their galactic space velocities and identify members in the different galactic populations. We also identified candidate members in young stellar kinematic groups, with ages ranging from 1 Ma to 800 Ma with SteParKin, LACEwING, and BANYAN {\Sigma}. We removed known close binaries and perform an analysis of kinematic, rotation, and activity indicators (rotational periods and projected velocities, Halpha, X-rays, and UV emission) for 1546 M dwarfs. We defined five rotation-activity-colour relations satisfied by young ({\tau} <= 800 Ma) stars. Results. We identified 191 young M dwarf candidates (~12%), 113 of which are newly recognised in this work. In this young sample, there are 118 very active stars based on H{\alpha} emission, fast rotation, and X-ray and UV emission excess. Of them, 27 have also strong magnetic fields, 9 of which are likely younger than 50 Ma. Additionally, there are 87 potentially young stars and 99 stars with a dubious youth classification, which may increase the fraction of young stars to an astounding 24%. Only one star out of the 2187 exhibits kinematics typical of the old Galactic halo. Conclusions. A combined analysis of kinematic and rotation-activity properties provides a robust method for identifying young M dwarfs from archival data. However, more observational efforts are needed to ascertain the true nature of numerous young star candidates in the field and, perhaps more importantly, to precisely quantify their age.

Superconductivity usually emerges from a metallic normal state which follows the Fermi-liquid paradigm. If, in contrast, the normal state is a fractionalized non-Fermi liquid, then pairing may either eliminate fractionalization via a Higgs-type mechanism leading to a conventional superconducting state, or pairing can occur in the presence of fractionalization. Here we discuss a simple model for the latter case: Using a combination of perturbation theory and functional renormalization group, we show that the Kitaev--Kondo lattice model displays a fractionalized superconducting phase at weak Kondo coupling. This phase is characterized by Cooper pairing of conventional electronic quasiparticles, coexisting with a spin-liquid background and topological order. Depending on the sign of the Kitaev coupling, we find the pairing to be either of chiral $d$-wave or $p$-wave type for extended doping regions around the van-Hove filling. We discuss applications and extensions., Comment: (7+7 pages, 3+7 figures)

Programming a quantum device describes the usage of quantum logic gates, agnostic of hardware specifics, to perform a sequence of operations with (typically) a computing or sensing task in mind. Such programs have been executed on digital quantum computers, which despite their noisy character, have shown the ability to optimize metrological functions, for example in the generation of spin squeezing and optimization of quantum Fisher information for signals manifesting as spin rotations in a quantum register. However, the qubits of these programmable quantum sensors are tightly spatially confined and therefore suboptimal for enclosing the kinds of large spacetime areas required for performing inertial sensing. In this work, we derive a set of quantum logic gates for a cold atom optical lattice interferometer that manipulates the momentum of atoms. Here, the operations are framed in terms of single qubit operations and mappings between qubit subspaces with internal levels given by the Bloch (crystal) eigenstates of the lattice. We describe how the quantum optimal control method of direct collocation is well suited for obtaining the modulation waveforms of the lattice which achieve these operations.

Algebraic spin liquids are quantum disordered phases of insulating magnets which exhibit fractionalized gapless excitations and power-law correlations. Quantum spin liquids in this category include the experimentally established 1D Luttinger liquid, as well as the U(1) Dirac spin liquid (DSL) which has been a focus of recent candidate materials searches. Most notably, several exchange-frustrated Heisenberg materials on the triangular lattice have shown evidence of the U(1) DSL. In this work, we measure the algebraic correlations of spin-singlet excitations in the $J_1$-$J_2$ antiferromagnetic Heisenberg model on the triangular lattice, prompting a detailed investigation of this model's stability under spin-phonon coupling using variational Monte Carlo. As seen before in 1D spin chains, we observe a low-temperature transition from a U(1) DSL to valence bond order and predict the parameter regime where the model realizes a stable DSL ground state. To achieve this, we employ a series of finite-size scaling Ans\"atze inspired by the low-energy DSL's conformal description in terms of quantum electrodynamics, and show that emergent monopole operators drive the instability. We compare the physics of this transition to the 1D Luttinger liquid throughout our analysis. We derive the regime of stability against spin-Peierls ordering and argue that the DSL ground state might still be achievable in candidate materials, despite its tendency to valence bond solid ordering.

Pre-trained Language Models (PLMs) encode various facts about the world at their pre-training phase as they are trained to predict the next or missing word in a sentence. There has a been an interest in quantifying and improving the amount of facts that can be extracted from PLMs, as they have been envisioned to act as soft knowledge bases, which can be queried in natural language. Different approaches exist to enhance fact retrieval from PLM. Recent work shows that the hidden states of PLMs can be leveraged to determine the truthfulness of the PLMs' inputs. Leveraging this finding to improve factual knowledge retrieval remains unexplored. In this work, we investigate the use of a helper model to improve fact retrieval. The helper model assesses the truthfulness of an input based on the corresponding hidden states representations from the PLMs. We evaluate this approach on several masked PLMs and show that it enhances fact retrieval by up to 33\%. Our findings highlight the potential of hidden states representations from PLMs in improving their factual knowledge retrieval.

In-context knowledge editing (IKE) enables efficient modification of large language model (LLM) outputs without parameter changes and at zero-cost. However, it can be misused to manipulate responses opaquely, e.g., insert misinformation or offensive content. Such malicious interventions could be incorporated into high-level wrapped APIs where the final input prompt is not shown to end-users. To address this issue, we investigate the detection and reversal of IKE-edits. First, we demonstrate that IKE-edits can be detected with high accuracy (F1 > 80\%) using only the top-10 output probabilities of the next token, even in a black-box setting, e.g. proprietary LLMs with limited output information. Further, we introduce the novel task of reversing IKE-edits using specially tuned reversal tokens. We explore using both continuous and discrete reversal tokens, achieving over 80\% accuracy in recovering original, unedited outputs across multiple LLMs. Our continuous reversal tokens prove particularly effective, with minimal impact on unedited prompts. Through analysis of output distributions, attention patterns, and token rankings, we provide insights into IKE's effects on LLMs and how reversal tokens mitigate them. This work represents a significant step towards enhancing LLM resilience against potential misuse of in-context editing, improving their transparency and trustworthiness.

We present the Merian Survey, an optical imaging survey optimized for studying the physical properties of bright star-forming dwarf galaxies. Merian is carried out with two medium-band filters ($N708$ and $N540$, centered at $708$ and $540$ nm), custom-built for the Dark Energy Camera (DECam) on the Blanco telescope. Merian covers $\sim 750\,\mathrm{deg}^2$ of equatorial fields, overlapping with the Hyper Suprime-Cam Subaru Strategic Program (HSC-SSP) wide, deep, and ultra-deep fields. When combined with the HSC-SSP imaging data ($grizy$), the new Merian DECam medium-band imaging allows for photometric redshift measurements via the detection of H$\rm\alpha$ and [OIII] line emission flux excess in the $N708$ and $N540$ filters, respectively, at $0.06

Generative Artificial Intelligence (GenAI) is becoming ubiquitous in our daily lives. The increase in computational power and data availability has led to a proliferation of both single- and multi-modal models. As the GenAI ecosystem matures, the need for extensible and model-agnostic risk identification frameworks is growing. To meet this need, we introduce the Python Risk Identification Toolkit (PyRIT), an open-source framework designed to enhance red teaming efforts in GenAI systems. PyRIT is a model- and platform-agnostic tool that enables red teamers to probe for and identify novel harms, risks, and jailbreaks in multimodal generative AI models. Its composable architecture facilitates the reuse of core building blocks and allows for extensibility to future models and modalities. This paper details the challenges specific to red teaming generative AI systems, the development and features of PyRIT, and its practical applications in real-world scenarios.

Reproducibility is essential for scientific research. However, in computer vision, achieving consistent results is challenging due to various factors. One influential, yet often unrecognized, factor is CUDA-induced randomness. Despite CUDA's advantages for accelerating algorithm execution on GPUs, if not controlled, its behavior across multiple executions remains non-deterministic. While reproducibility issues in ML being researched, the implications of CUDA-induced randomness in application are yet to be understood. Our investigation focuses on this randomness across one standard benchmark dataset and two real-world datasets in an isolated environment. Our results show that CUDA-induced randomness can account for differences up to 4.77% in performance scores. We find that managing this variability for reproducibility may entail increased runtime or reduce performance, but that disadvantages are not as significant as reported in previous studies.

We present a non-uniform analogue of the classical Datko-Pazy theorem. Our main result shows that an integrability condition imposed on orbits originating in a fractional domain of the generator (as opposed to all orbits) implies polynomial stability of a bounded $C_0$-semigroup. As an application of this result we establish polynomial stability of a semigroup under a certain non-uniform Lyapunov-type condition. We moreover give a new proof, under slightly weaker assumptions, of a recent result deducing polynomial stability from a certain non-uniform observability condition.

The wiretap channel is a well-studied problem in the physical layer security (PLS) literature. Although it is proven that the decoding error probability and information leakage can be made arbitrarily small in the asymptotic regime, further research on finite-blocklength codes is required on the path towards practical, secure communications systems. This work provides the first experimental characterization of a deep learning-based, finite-blocklength code construction for multi-tap fading wiretap channels without channel state information (CSI). In addition to the evaluation of the average probability of error and information leakage, we illustrate the influence of (i) the number of fading taps, (ii) differing variances of the fading coefficients and (iii) the seed selection for the hash function-based security layer.

Machine learning models are known to learn spurious correlations, i.e., features having strong relations with class labels but no causal relation. Relying on those correlations leads to poor performance in the data groups without these correlations and poor generalization ability. To improve the robustness of machine learning models to spurious correlations, we propose an approach to extract a subnetwork from a fully trained network that does not rely on spurious correlations. The subnetwork is found by the assumption that data points with the same spurious attribute will be close to each other in the representation space when training with ERM, then we employ supervised contrastive loss in a novel way to force models to unlearn the spurious connections. The increase in the worst-group performance of our approach contributes to strengthening the hypothesis that there exists a subnetwork in a fully trained dense network that is responsible for using only invariant features in classification tasks, therefore erasing the influence of spurious features even in the setup of multi spurious attributes and no prior knowledge of attributes labels.

Volumetric neuroimaging examinations like structural Magnetic Resonance Imaging (sMRI) are routinely applied to support the clinical diagnosis of dementia like Alzheimer's Disease (AD). Neuroradiologists examine 3D sMRI to detect and monitor abnormalities in brain morphology due to AD, like global and/or local brain atrophy and shape alteration of characteristic structures. There is a strong research interest in developing diagnostic systems based on Deep Learning (DL) models to analyse sMRI for AD. However, anatomical information extracted from an sMRI examination needs to be interpreted together with patient's age to distinguish AD patterns from the regular alteration due to a normal ageing process. In this context, part-prototype neural networks integrate the computational advantages of DL in an interpretable-by-design architecture and showed promising results in medical imaging applications. We present PIMPNet, the first interpretable multimodal model for 3D images and demographics applied to the binary classification of AD from 3D sMRI and patient's age. Despite age prototypes do not improve predictive performance compared to the single modality model, this lays the foundation for future work in the direction of the model's design and multimodal prototype training process, Comment: Accepted "late-breaking work" at XAI-2024

Recent innovations in language model training have demonstrated that it is possible to create highly performant models that are small enough to run on a smartphone. As these models are deployed in an increasing number of domains, it is critical to ensure that they are aligned with human preferences and safety considerations. In this report, we present our methodology for safety aligning the Phi-3 series of language models. We utilized a "break-fix" cycle, performing multiple rounds of dataset curation, safety post-training, benchmarking, red teaming, and vulnerability identification to cover a variety of harm areas in both single and multi-turn scenarios. Our results indicate that this approach iteratively improved the performance of the Phi-3 models across a wide range of responsible AI benchmarks. Finally, we include additional red teaming strategies and evaluations that were used to test the safety behavior of Phi-3.5-mini and Phi-3.5-MoE, which were optimized for multilingual capabilities.

We study the long-time asymptotic behaviour of a topologically non-trivial network of wave and heat equations. By analysing the simpler wave and the heat networks separately, and then applying recent results for abstract coupled systems, we establish energy decay at the rate $t^{-4}$ as $t\to\infty$ for all classical solutions.

The coupled transport of the charge and orbital angular momentum of electrons is at the heart of orbitronics. Here, we discuss the reciprocal relation between the direct and inverse orbital Hall effects (OHEs) in thin films. We argue that the conventional orbital current is ill-defined as it does not satisfy the reciprocal relation owing to non-conservation of the orbital angular momentum. We resolve the problem by adopting the definition of the so-called \emph{proper} orbital current, which is directly related to orbital accumulation. We prove the reciprocal relation between the \emph{global} response of orbital and charge currents. However, we show that their \emph{local} distributions are generally different, especially due to gigantic contributions at surfaces, which may lead to unintuitive results when charge and orbital currents are locally measured. We demonstrate our predictions by first-principles calculations on W(110) and Pt(111) thin films. In W(110), the direct and inverse OHEs are severely non-reciprocal locally in each layer although the total responses are exactly reciprocal. Interestingly, the SHEs are almost reciprocal locally in each layer. On the other hand, in Pt(111), both OHEs and SHEs are locally non-reciprocal, which we attribute to the pronounced spin-orbit interaction. We propose that the locally distinct responses may be used to distinguish the spin and orbital currents in experiments.

Variational quantum algorithms (VQAs) show potential for quantum advantage in the near term of quantum computing, but demand a level of accuracy that surpasses the current capabilities of NISQ devices. To systematically mitigate the impact of quantum device error on VQAs, we propose Clapton: Clifford-Assisted Problem Transformation for Error Mitigation in Variational Quantum Algorithms. Clapton leverages classically estimated good quantum states for a given VQA problem, classical simulable models of device noise, and the variational principle for VQAs. It applies transformations on the VQA problem's Hamiltonian to lower the energy estimates of known good VQA states in the presence of the modeled device noise. The Clapton hypothesis is that as long as the known good states of the VQA problem are close to the problem's ideal ground state and the device noise modeling is reasonably accurate (both of which are generally true), then the Clapton transformation substantially decreases the impact of device noise on the ground state of the VQA problem, thereby increasing the accuracy of the VQA solution. Clapton is built as an end-to-end application-to-device framework and achieves mean VQA initialization improvements of 1.7x to 3.7x, and up to a maximum of 13.3x, over the state-of-the-art baseline when evaluated for a variety of scientific applications from physics and chemistry on noise models and real quantum devices.

Deconfined quantum critical points (DQCPs) are putative phase transitions beyond the Landau paradigm with emergent fractionalized degrees of freedom. The original example of a DQCP is the spin-1/2 quantum antiferromagnet on the square lattice which features a second order transition between valence bond solid (VBS) and N\'eel order. The VBS order breaks a lattice symmetry, and the corresponding VBS order parameter may couple to lattice distortion modes (phonons) at appropriate momenta. We investigate a field-theoretic description of the DQCP in the presence of such a spin-lattice coupling. We show that treating phonons as classical lattice distortions leads to a relevant monopole-phonon interaction inducing an instability towards a distorted lattice by an analogous mechanism to the spin-Peierls instability in one dimension. Consequently, there is a breakdown of the DQCP which generally becomes a strong first-order transition. Taking into account the full quantum nature of the phonons, we argue that the continuous DQCP persists above a critical phonon frequency. Lastly, we comment on the connection to general gapless, deconfined gauge theories., Comment: 12 pages, 3 figures

For a network of discrete states with a periodically driven Markovian dynamics, we develop an inference scheme for an external observer who has access to some transitions. Based on waiting-time distributions between these transitions, the periodic probabilities of states connected by these observed transitions and their time-dependent transition rates can be inferred. Moreover, the smallest number of hidden transitions between accessible ones and some of their transition rates can be extracted. We prove and conjecture lower bounds on the total entropy production for such periodic stationary states. Even though our techniques are based on generalizations of known methods for steady states, we obtain original results for those as well.

Quantity and price risks are key uncertainties market participants face in electricity markets with increased volatility, for instance, due to high shares of renewables. From day ahead until real-time, there is a large variation in the best available information, leading to price changes that flexible assets, such as battery storage, can exploit economically. This study contributes to understanding how coordinated bidding strategies can enhance multi-market trading and large-scale energy storage integration. Our findings shed light on the complexities arising from interdependencies and the high-dimensional nature of the problem. We show how stochastic dual dynamic programming is a suitable solution technique for such an environment. We include the three markets of the frequency containment reserve, day-ahead, and intraday in stochastic modelling and develop a multi-stage stochastic program. Prices are represented in a multidimensional Markov Chain, following the scheduling of the markets and allowing for time-dependent randomness. Using the example of a battery storage in the German energy sector, we provide valuable insights into the technical aspects of our method and the economic feasibility of battery storage operation. We find that capacity reservation in the frequency containment reserve dominates over the battery's cycling in spot markets at the given resolution on prices in 2022. In an adjusted price environment, we find that coordination can yield an additional value of up to 12.5%.

Coverage-guided fuzz testing has received significant attention from the research community, with a strong focus on binary applications, greatly disregarding other targets, such as web applications. The importance of the World Wide Web in everyone's life cannot be overstated, and to this day, many web applications are developed in PHP. In this work, we address the challenges of applying coverage-guided fuzzing to PHP web applications and introduce PHUZZ, a modular fuzzing framework for PHP web applications. PHUZZ uses novel approaches to detect more client-side and server-side vulnerability classes than state-of-the-art related work, including SQL injections, remote command injections, insecure deserialization, path traversal, external entity injection, cross-site scripting, and open redirection. We evaluate PHUZZ on a diverse set of artificial and real-world web applications with known and unknown vulnerabilities, and compare it against a variety of state-of-the-art fuzzers. In order to show PHUZZ' effectiveness, we fuzz over 1,000 API endpoints of the 115 most popular WordPress plugins, resulting in over 20 security issues and 2 new CVE-IDs. Finally, we make the framework publicly available to motivate and encourage further research on web application fuzz testing., Comment: Preprint; Final version to be published in ASIA CCS '24: Proceedings of the 2024 ACM Asia Conference on Computer and Communications Security

We present an educational escape room in physics (ERP) covering the main disciplines of physics as taught in the education of neighboring sciences, in this case veterinary medicine, extended by a European board game system of victory points (VP). The puzzles in physics are mandatory to master the ERP requiring to be solved in a sequential order. In our study, we show the growth of knowledge by knowledge-transfer cards separated into disciplines of physics as demonstrated by critical exemplary questions of test groups. Tactical secondary puzzles, the so called \textit{Eurogame puzzles} (EPs), are able to be solved in parallel in any arbitrary and independent sequence, allowing a self-adjusted challenge autolevelled by the students by choice. The motivation is improved by a scoring list to solve, at least, a few optional EPs as additional game-like adventure and challenge collecting VP. A second part of the study demonstrates the attraction of the introduced European board game concept as an important tool for gamificiation and open-by-choice EPs not being necessarily related to physics directly. EPs are able to smooth the contrast of playing games and learning, providing an additional bridge to overcome the fear of physics. A strong advantage is the flexibility of the European board concept, choosing from a broad range of optional EPs to fascinate and motivate students taking into account individual preferences improving escape rooms by decision-making skills. The concept is able be applied to every escape room independent of the taught subject or topic., Comment: 20 pages, 9 figures

Several diatomic transition metal oxides, rare-earth metal oxides and fluorides have the unusual property that their bond dissociation energy is larger than their ionization energy. In these molecules, bound levels above the ionization energy can be populated via strong, resonant transitions from the ground state. The only relevant decay channel of these levels is autoionization; predissociation is energetically not possible and radiative decay is many orders of magnitude slower. Starting from translationally cold neutral molecules, translationally cold molecular ions can thus be produced with very high efficiency. By populating bound levels just above the ionization energy, internally cold molecular ions, exclusively occupying the lowest rotational level, are produced. This is experimentally shown here for the dysprosium monoxide molecule, DyO, for which the lowest bond dissociation energy is determined to be 0.0831(6) eV above the ionization energy.