Search

Hoare-style inference rules for program constructs permit the copying of expressions and tests from program text into logical contexts. It is known that this requires care even for sequential programs but further issues arise for concurrent programs because of potential interference to the values of variables. The "rely-guarantee" approach does tackle the issue of recording acceptable interference and offers a way to provide safe inference rules. This paper shows how the algebraic presentation of rely-guarantee ideas can clarify and formalise the conditions for safely re-using expressions and tests from program text in logical contexts for reasoning about programs., Comment: 17 pages, 1 figure

The Gemini South telescope is now equipped with a new high-resolution spectrograph called GHOST (the Gemini High-resolution Optical SpecTrograph). This instrument provides high-efficiency, high-resolution spectra covering 347-1060 nm in a single exposure of either one or two targets simultaneously, along with precision radial velocity spectroscopy utilizing an internal calibration source. It can operate at a spectral element resolving power of either 76000 or 56000, and can reach a SNR$\sim$5 in a 1hr exposure on a V$\sim$20.8 mag target in median site seeing, and dark skies (per resolution element). GHOST was installed on-site in June 2022, and we report performance after full integration to queue operations in November 2023, in addition to scientific results enabled by the integration observing runs. These results demonstrate the ability to observe a wide variety of bright and faint targets with high efficiency and precision. With GHOST, new avenues to explore high-resolution spectroscopy have opened up to the astronomical community. These are described, along with the planned and potential upgrades to the instrument., Comment: Accepted in the Astronomical Journal; 26 pages, 24 figures and 4 tables

Feedback is widely recognized as an essential condition for Lyman continuum (LyC) escape in star-forming galaxies. However, the mechanisms by which galactic outflows clear neutral gas and dust remain unclear. In this paper, we model the Mg II 2796\r{A}, 2804\r{A} absorption + emission lines in 29 galaxies taken from the Low-z LyC Survey (LzLCS) to investigate the impact of (radiation + mechanical) feedback on LyC escape. Using constraints on Mg$^+$ and photoionization models, we map the outflows' neutral hydrogen content and predict $f_{esc}^{LyC}$ with a multiphase wind model. We measure mass, momentum, and energy loading factors for the neutral winds, which carry up to 10% of the momentum and 1% of the energy in SFR-based deposition rates. We use SED template fitting to determine the relative ages of stellar populations, allowing us to identify radiation feedback dominant systems. We then examine feedback related properties (stellar age, loading factors, etc.) under conditions that optimize feedback efficiency, specifically high star formation rate surface density and compact UV half-light radii. Our findings indicate that the strongest leakers are radiation feedback dominant, lack Mg II outflows, but have extended broad components in higher ionization lines like [O III] 5007\r{A}, as observed by Amor\'in et al. (2024). In contrast, galaxies experiencing supernovae feedback typically exhibit weaker $f_{esc}^{LyC}$ and show evidence of outflows in both Mg II and higher ionization lines. We attribute these findings to rapid or "catastrophic" cooling in the radiation-dominant systems, which, given the low metallicities in our sample, are likely experiencing delayed supernovae., Comment: 34 pages, 16 figures, 7 tables

One compelling use of citation networks is to characterize papers by their relationships to the surrounding literature. We propose a method to characterize papers by embedding them into two distinct "co-factor" spaces: one describing how papers send citations, and the other describing how papers receive citations. This approach presents several challenges. First, older documents cannot cite newer documents, and thus it is not clear that co-factors are even identifiable. We resolve this challenge by developing a co-factor model for asymmetric adjacency matrices with missing lower triangles and showing that identification is possible. We then frame estimation as a matrix completion problem and develop a specialized implementation of matrix completion because prior implementations are memory bound in our setting. Simulations show that our estimator has promising finite sample properties, and that naive approaches fail to recover latent co-factor structure. We leverage our estimator to investigate 237,794 papers published in statistics journals from 1898 to 2022, resulting in the most comprehensive topic model of the statistics literature to date. We find interpretable co-factors corresponding to many statistical subfields, including time series, variable selection, spatial methods, graphical models, GLM(M)s, causal inference, multiple testing, quantile regression, resampling, semi-parametrics, dimension reduction, and several more.

Solar eruptive events, including solar flares and coronal mass ejections (CMEs), are typically characterised by energetically significant X-ray emissions from flare-accelerated electrons and hot thermal plasmas. Occulted events, where the main flare is blocked by the solar limb, provide an opportunity to observe and analyse the X-ray emissions specifically associated with CMEs. This study investigates the X-ray and extreme ultraviolet (EUV) emissions associated with a large filament eruption and CME that occurred on February 15, 2022. This event was highly occulted from the three vantage points of Solar Orbiter, STEREO-A, and Earth. We utilised X-ray observations from the Spectrometer/Telescope for Imaging X-rays (STIX) and EUV observations from the Full Sun Imager (FSI) of the Extreme Ultraviolet Imager (EUI) on-board Solar Orbiter, supplemented by multi-viewpoint observations from STEREO-A/EUVI. This enabled a comprehensive analysis of the X-ray emissions in relation to the filament structure observed in EUV. We used STIX's imaging and spectroscopy capabilities to characterise the X-ray source associated with the eruption. Our analysis reveals that the X-ray emissions associated with the occulted eruption originated from an altitude exceeding 0.3Rsun above the main flare site. The X-ray time-profile showed a sharp increase and exponential decay, and consisted of both a hot thermal component at 17 MK and non-thermal emissions (>11.4 keV) characterised by an electron spectral index of 3.9. Imaging analysis showed an extended X-ray source that coincided with the EUV emission as observed from EUI, and was imaged until the source grew to a size larger than the imaging limit of STIX (180 arcsec). The findings demonstrate that STIX combined with EUI provides a unique and powerful tool for examining the energetic properties of the CME component of solar energetic eruptions., Comment: 9 pages, 8 figures, accepted A&A

Mapping stars and gas in nearby galaxies is fundamental for understanding their growth and the impact of their environment. This issue is addressed by comparing the stellar "edges" of galaxies $D_{\rm stellar}$, defined as the outermost diameter where in situ star formation significantly drops, with the gaseous distribution parameterized by the neutral atomic hydrogen diameter measured at 1 $M_{sun}$/pc$^2$, $D_{HI}$. By sampling a broad HI mass range $10^5 M_{sun} < M_{HI} < 10^{11} M_{sun}$, we find several dwarf galaxies with $M_{HI} < 10^9 M_{sun}$ from the field and Fornax Cluster which are distinguished by $D_{\rm stellar} >> D_{HI}$. For the cluster dwarfs, the average HI surface density near $D_{\rm stellar}$ is $\sim$0.3 $M_{sun}$/pc$^2$, reflecting the impact of quenching and outside-in gas removal from ram pressure and tidal interactions. In comparison, $D_{\rm stellar}/D_{HI}$ ranges between 0.5-2 in dwarf field galaxies, consistent with the expectations from stellar feedback. Only more massive disk galaxies in the field can thus be characterized by the common assumption that $D_{\rm stellar} \lesssim D_{HI}$. We discover a break in the $D_{\rm stellar}-M_{\rm stellar}$ relation at $m_{break} \sim 4\times10^8 M_{sun}$ that potentially differentiates the low mass regime where the influence of stellar feedback and environmental processes more prominently regulates the sizes of nearby galaxies. Our results highlight the importance of combining deep optical and HI imaging for understanding galaxy evolution., Comment: 19 pages, 10 figures, 2 tables. Key results in Figs. 4, 6, 7 and 8. Accepted for publication in The Astrophysical Journal

The ``2-process Model'' is a promising technique for interpreting stellar chemical abundance data from large-scale surveys (e.g., SDSS-IV/V, GALAH), enabling more quantitative empirical studies of differences in chemical enrichment history between galaxies without relying on detailed yield and evolution models. In this work, we fit 2-process model parameters to (1) a luminous giant Milky Way (MW) sample and (2) stars comprising the Sagittarius Dwarf Galaxy (Sgr). We then use these two sets of model parameters to predict the abundances of 14 elements of stars belonging to the MW and in five of its massive satellite galaxies, analyzing the residuals between the predicted and observed abundances. We find that the model fit to (1) results in large residuals (0.1-0.3 dex) for most metallicity-dependent elements in the metal-rich ([Mg/H] $>$ -0.8) stars of the satellite galaxies. However, the model fit to (2) results in small or no residuals for all elements across all satellite galaxies. Therefore, despite the wide variation in [X/Mg]-[Mg/H] abundance patterns of the satellite galaxies, the 2-process framework provides an accurate characterization of their abundance patterns across many elements, but these multi-element patterns are systematically different between the dwarf galaxy satellites and the MW disks. We consider a variety of scenarios for the origin of this difference, highlighting the possibility that a large inflow of pristine gas to the MW disk diluted the metallicity of star-forming gas without changing abundance ratios., Comment: 25 pages, 14 figures, accepted for publication in ApJ

Single-shot quantum error correction has the potential to speed up quantum computations by removing the need for multiple rounds of syndrome extraction in order to be fault-tolerant. Using Quantinuum's H2 trapped-ion quantum computer, we implement the [[33,1,4]] 4D surface code and perform the first experimental demonstration of single-shot quantum error correction with bare ancilla qubits. We conduct memory experiments comparing the 2D and 4D surface codes and find that despite differences in qubit use and syndrome extraction circuit depth, the 4D surface code matches or outperforms the 2D surface code in both the fault-tolerant and single-shot regimes.

We introduce Imagen 3, a latent diffusion model that generates high quality images from text prompts. We describe our quality and responsibility evaluations. Imagen 3 is preferred over other state-of-the-art (SOTA) models at the time of evaluation. In addition, we discuss issues around safety and representation, as well as methods we used to minimize the potential harm of our models.

Deep learning models are effective, yet brittle. Even carefully trained, their behavior tends to be hard to predict when confronted with out-of-distribution samples. In this work, our goal is to propose a simple yet effective solution to predict and describe via natural language potential failure modes of computer vision models. Given a pretrained model and a set of samples, our aim is to find sentences that accurately describe the visual conditions in which the model underperforms. In order to study this important topic and foster future research on it, we formalize the problem of Language-Based Error Explainability (LBEE) and propose a set of metrics to evaluate and compare different methods for this task. We propose solutions that operate in a joint vision-and-language embedding space, and can characterize through language descriptions model failures caused, e.g., by objects unseen during training or adverse visual conditions. We experiment with different tasks, such as classification under the presence of dataset bias and semantic segmentation in unseen environments, and show that the proposed methodology isolates nontrivial sentences associated with specific error causes. We hope our work will help practitioners better understand the behavior of models, increasing their overall safety and interpretability., Comment: Preprint of the eXCV Workshop paper (ECCV'24)

Developers of low-level systems code providing core functionality for operating systems and kernels must address hardware-level features of modern multicore architectures. A particular feature is pipelined "out-of-order execution" of the code as written, the effects of which are typically summarised as a "weak memory model" - a term which includes further complicating factors that may be introduced by compiler optimisations. In many cases, the nondeterminism inherent in weak memory models can be expressed as micro-parallelism, i.e., parallelism within threads and not just between them. Fortunately Jones' rely/guarantee reasoning provides a compositional method for shared-variable concurrency, whether that be in terms of communication between top-level threads or micro-parallelism within threads. In this paper we provide an in-depth verification of the lock algorithm used in the seL4 microkernel, using rely/guarantee to handle both interthread communication as well as micro-parallelism introduced by weak memory models.

This paper presents an estimation and control framework that enables the targeted reentry of a drag-modulated spacecraft in the presence of atmospheric density uncertainty. In particular, an extended Kalman filter (EKF) is used to estimate the in-flight density errors relative to the atmospheric density used to generate the nominal guidance trajectory. This information is leveraged within a model predictive control (MPC) strategy to improve tracking performance, reduce control effort, and increase robustness to actuator saturation compared to the state-of-the-art approach. The estimation and control framework is tested in a Monte Carlo simulation campaign with historical space weather data. These simulation efforts demonstrate that the proposed framework is able to stay within 100 km of the guidance trajectory at all points in time for 98.4% of cases. The remaining 1.6% of cases were pushed away from the guidance by large density errors, many due to significant solar storms and flares, that could not physically be compensated for by the drag control device. For the successful cases, the proposed framework was able to guide the spacecraft to the desired location at the entry interface altitude with a mean error of 12.1 km and 99.7% of cases below 100 km., Comment: Submitted to the Journal of Guidance, Control, and Dynamics

Tensor networks and quantum computation are two of the most powerful tools for the simulation of quantum many-body systems. Rather than viewing them as competing approaches, here we consider how these two methods can work in tandem. We introduce a novel algorithm that combines tensor networks and quantum computation to produce results that are more accurate than what could be achieved by either method used in isolation. Our algorithm is based on multiproduct formulas (MPF) - a technique that linearly combines Trotter product formulas to reduce algorithmic error. Our algorithm uses a quantum computer to calculate the expectation values and tensor networks to calculate the coefficients used in the linear combination. We present a detailed error analysis of the algorithm and demonstrate the full workflow on a one-dimensional quantum simulation problem on $50$ qubits using two IBM quantum computers: $ibm\_torino$ and $ibm\_kyiv$.

Fusing different sensor modalities can be a difficult task, particularly if they are asynchronous. Asynchronisation may arise due to long processing times or improper synchronisation during calibration, and there must exist a way to still utilise this previous information for the purpose of safe driving, and object detection in ego vehicle/ multi-agent trajectory prediction. Difficulties arise in the fact that the sensor modalities have captured information at different times and also at different positions in space. Therefore, they are not spatially nor temporally aligned. This paper will investigate the challenge of radar and LiDAR sensors being asynchronous relative to the camera sensors, for various time latencies. The spatial alignment will be resolved before lifting into BEV space via the transformation of the radar/LiDAR point clouds into the new ego frame coordinate system. Only after this can we concatenate the radar/LiDAR point cloud and lifted camera features. Temporal alignment will be remedied for radar data only, we will implement a novel method of inferring the future radar point positions using the velocity information. Our approach to resolving the issue of sensor asynchrony yields promising results. We demonstrate velocity information can drastically improve IoU for asynchronous datasets, as for a time latency of 360 milliseconds (ms), IoU improves from 49.54 to 53.63. Additionally, for a time latency of 550ms, the camera+radar (C+R) model outperforms the camera+LiDAR (C+L) model by 0.18 IoU. This is an advancement in utilising the often-neglected radar sensor modality, which is less favoured than LiDAR for autonomous driving purposes., Comment: This paper is a preprint of a paper submitted to the 26th Irish Machine Vision and Image Processing Conference (IMVIP 2024). If accepted, the copy of record will be available at IET Digital Library

The excess broadening of high-temperature spectral lines, long observed near the tops of flare arcades, is widely considered to result from magnetohydrodynamic (MHD) turbulence. According to different theories, plasma turbulence is also believed to be a candidate mechanism for particle acceleration during solar flares. However, the degree to which this broadening is connected to the acceleration of non-thermal electrons remains largely unexplored outside of recent work, and many observations have been limited by limited spatial resolution and cadence. Using the Interface Region Imaging Spectrometer (IRIS), we present spatially resolved observations of loop-top broadenings using hot (11MK) Fe XXI 1354.1 \r{A} line emission at ~9s cadence during the 2022 March 30 X1.3 flare. We find non-thermal velocities upwards of 65km/s that decay linearly with time, indicating the presence and subsequent dissipation of plasma turbulence. Moreover, the initial Fe XXI signal was found to be co-spatial and co-temporal with microwave emission measured by the Expanded Owens Valley Solar Array (EOVSA), placing a population of non-thermal electrons in the same region as the loop-top turbulence. Evidence of electron acceleration at this time is further supported by hard X-ray measurements from the Spectrometer/Telescope for Imaging X-rays (STIX) aboard Solar Orbiter. Using the decay of non-thermal broadenings as a proxy for turbulent dissipation, we found the rate of energy dissipation to be consistent with the power of non-thermal electrons deposited into the chromosphere, suggesting a possible connection between turbulence and electron acceleration.

The tensor-train (TT) format is a data-sparse tensor representation commonly used in high dimensional function approximations arising from computational and data sciences. Various sequential and parallel TT decomposition algorithms have been proposed for different tensor inputs and assumptions. In this paper, we propose subtensor parallel adaptive TT cross, which partitions a tensor onto distributed memory machines with multidimensional process grids, and constructs an TT approximation iteratively with tensor elements. We derive two iterative formulations for pivot selection and TT core construction under the distributed memory setting, conduct communication and scaling analysis of the algorithm, and illustrate its performance with multiple test experiments. These include up to 6D Hilbert tensors and tensors constructed from Maxwellian distribution functions that arise in kinetic theory. Our results demonstrate significant accuracy with greatly reduced storage requirements via the TT cross approximation. Furthermore, we demonstrate good to optimal strong and weak scaling performance for the proposed parallel algorithm.

With high equilibrium temperatures and tidally locked rotation, ultra-hot Jupiters (UHJs) are unique laboratories within which to probe extreme atmospheric physics and chemistry. In this paper, we present high-resolution dayside spectroscopy of the UHJ WASP-189b obtained with the new Gemini High-resolution Optical SpecTrograph (GHOST) at the Gemini South Observatory. The observations, which cover three hours of post-eclipse orbital phases, were obtained during the instrument's System Verification run. We detect the planet's atmosphere via the Doppler cross-correlation technique, and recover a detection of neutral iron in the planet's dayside atmosphere at a significance of 7.5$\sigma$ in the red-arm of the data, verifying the presence of a thermal inversion. We also investigate the presence of other species in the atmosphere and discuss the implications of model injection/recovery tests. These results represent the first atmospheric characterization of an exoplanet with GHOST's high-resolution mode, and demonstrate the potential of this new instrument in detecting and studying ultra-hot exoplanet atmospheres., Comment: 21 pages, 12 figures, 3 tables. Accepted for publication in The Astronomical Journal

While JWST has observed galaxies assembling as early as $z\sim14$, evidence of galaxies with significant old stellar populations in the Epoch of Reionisation (EoR) -- the descendants of these earliest galaxies -- are few and far between. Bursty star-formation histories (SFHs) have been invoked to explain the detectability of the earliest UV-bright galaxies, but also to interpret galaxies showing Balmer breaks without nebular emission lines. We present the first spectroscopic evidence of a $z\sim7.9$ galaxy, A2744-YD4, which shows a Balmer break and emission lines, indicating the presence of both a mature and young stellar population. The spectrum of A2744-YD4 shows peculiar emission line ratios suggesting a relatively low ionisation parameter and high gas-phase metallicity. A median stack of galaxies with similar emission line ratios reveals a clear Balmer break in their stacked spectrum. This suggests that a mature stellar population ($\sim 80$ Myr old) has produced a chemically enriched, disrupted interstellar medium. Based on SED-fitting and comparison to simulations, we conclude that the observed young stellar population is in fact the result of a rejuvenation event following a lull in star formation lasting $\sim 20$ Myr, making A2744-YD4 and our stack the first spectroscopic confirmation of galaxies that have rejuvenated following a mini-quenched phase. These rejuvenating galaxies appear to be in an exceptional evolutionary moment where they can be identified. Our analysis shows that a young stellar population of just $\sim 30 \%$ of the total stellar mass would erase the Balmer break. Hence, 'outshining' through bursty SFHs of galaxies in the early Universe is likely plaguing attempts to measure their stellar ages and masses accurately., Comment: 14 pages, 12 figures, 1 table. Submitted to MNRAS. Comments are welcome

Galaxy clusters are the most massive, gravitationally-bound structures in the Universe, emerging through hierarchical structure formation of large-scale dark matter and baryon overdensities. Early galaxy ``proto-clusters'' are believed to be important physical drivers of the overall cosmic star-formation rate density and serve as ``hotspots'' for the reionization of the intergalactic medium. Our understanding of the formation of these structures at the earliest cosmic epochs is, however, limited to sparse observations of their galaxy members, or based on phenomenological models and cosmological simulations. Here we report the detection of a massive neutral, atomic hydrogen (HI) gas reservoir permeating a galaxy proto-cluster at redshift $z=5.4$, observed one billion years after the Big Bang. The presence of this cold gas is revealed by strong damped Lyman-$\alpha$ absorption features observed in several background galaxy spectra taken with JWST/NIRSpec in close on-sky projection. While overall the sightlines probe a large range in HI column densities, $N_{\rm HI} = 10^{21.7}-10^{23.5}$ cm$^{-2}$, they are similar across nearby sightlines, demonstrating that they probe the same dense, neutral gas. This observation of a massive, large-scale overdensity of cold neutral gas challenges current large-scale cosmological simulations and has strong implications for the reionization topology of the Universe., Comment: Submitted

Due to the space radiation environment at L2, ESA's Euclid mission will be subject to a large amount of highly energetic particles over its lifetime. These particles can cause damage to the detectors by creating defects in the silicon lattice. These defects degrade the returned image in several ways, one example being a degradation of the Charge Transfer Efficiency, which appears as readout trails in the image data. This can be problematic for the Euclid VIS instrument, which aims to measure the shapes of galaxies to a very high degree of accuracy. Using a special clocking technique called trap pumping, the single defects in the CCDs can be detected and characterised. Being the first instrument in space with this capability, it will provide novel insights into the creation and evolution of radiation-induced defects and give input to the radiation damage correction of the scientific data. We present the status of the radiation damage of the Euclid VIS CCDs and how it has evolved over the first year in space., Comment: Submitted to the Proceedings for the SPIE Space Telescopes and Instrumentation 2024: Optical, Infrared, and Millimeter Wave, paper no. 13092-24, 10 pages, 6 figures

Exact unlearning was first introduced as a privacy mechanism that allowed a user to retract their data from machine learning models on request. Shortly after, inexact schemes were proposed to mitigate the impractical costs associated with exact unlearning. More recently unlearning is often discussed as an approach for removal of impermissible knowledge i.e. knowledge that the model should not possess such as unlicensed copyrighted, inaccurate, or malicious information. The promise is that if the model does not have a certain malicious capability, then it cannot be used for the associated malicious purpose. In this paper we revisit the paradigm in which unlearning is used for in Large Language Models (LLMs) and highlight an underlying inconsistency arising from in-context learning. Unlearning can be an effective control mechanism for the training phase, yet it does not prevent the model from performing an impermissible act during inference. We introduce a concept of ununlearning, where unlearned knowledge gets reintroduced in-context, effectively rendering the model capable of behaving as if it knows the forgotten knowledge. As a result, we argue that content filtering for impermissible knowledge will be required and even exact unlearning schemes are not enough for effective content regulation. We discuss feasibility of ununlearning for modern LLMs and examine broader implications.

Competing and intertwined orders are ubiquitous in strongly correlated electron systems, such as the charge, spin, and superconducting orders in the high-Tc cuprates. Recent scanning tunneling microscopy (STM) measurements provide evidence for a charge density wave (CDW) that coexists with superconductivity in the heavy Fermion metal UTe2. This CDW persists up to at least 7.5 K and, as a CDW breaks the translational symmetry of the lattice, its disappearance is necessarily accompanied by thermodynamic phase transition. Here, we report high-precision thermodynamic measurements of the elastic moduli of UTe2. We observe no signature of a phase transition in the elastic moduli down to a level of 1 part in 10^7, strongly implying the absence of bulk CDW order in UTe2. We suggest that the CDW and associated pair density wave (PDW) observed by STM may be confined to the surface of UTe2.

BlueMUSE is a blue-optimised, medium spectral resolution, panoramic integral field spectrograph under development for the Very Large Telescope (VLT). With an optimised transmission down to 350 nm, spectral resolution of R$\sim$3500 on average across the wavelength range, and a large FoV (1 arcmin$^2$), BlueMUSE will open up a new range of galactic and extragalactic science cases facilitated by its specific capabilities. The BlueMUSE consortium includes 9 institutes located in 7 countries and is led by the Centre de Recherche Astrophysique de Lyon (CRAL). The BlueMUSE project development is currently in Phase A, with an expected first light at the VLT in 2031. We introduce here the Top Level Requirements (TLRs) derived from the main science cases, and then present an overview of the BlueMUSE system and its subsystems fulfilling these TLRs. We specifically emphasize the tradeoffs that are made and the key distinctions compared to the MUSE instrument, upon which the system architecture is built., Comment: 20 pages, 10 figures, proceedings of the SPIE astronomical telescopes and instrumentation conference, Yokohama, 16-21 June

Reinforcement learning with human feedback (RLHF) has become the dominant method to align large models to user preferences. Unlike fine-tuning, for which there are many studies regarding training data memorization, it is not clear how memorization is affected by or introduced in the RLHF alignment process. Understanding this relationship is important as real user data may be collected and used to align large models; if user data is memorized during RLHF and later regurgitated, this could raise privacy concerns. In addition to RLHF, other methods such as Direct Preference Optimization (DPO) and $\Psi$PO have gained popularity for learning directly from human preferences, removing the need for optimizing intermediary reward models with reinforcement learning. In this work, we analyze how training data memorization can surface and propagate through each phase of RLHF and direct preference learning. We focus our study on code completion models, as code completion is one of the most popular use cases for large language models. We find that RLHF significantly decreases the chance that data used for reward modeling and reinforcement learning is memorized in comparison to directly fine-tuning on this data, but that examples already memorized during the fine-tuning stage of RLHF, will, in the majority of cases, remain memorized after RLHF. In contrast, we find that aligning by learning directly from human preference data via a special case of $\Psi$PO, Identity Preference Optimization (IPO), increases the likelihood that training data is regurgitated compared to RLHF. Our work suggests that RLHF, as opposed to direct preference learning, is a safer way to mitigate the risk of regurgitating sensitive preference data when aligning large language models. We find our conclusions are robust across multiple code completion datasets, tasks, and model scales.

Large Language Models (LLMs) have taken the field of AI by storm, but their adoption in the field of Artificial Life (ALife) has been, so far, relatively reserved. In this work we investigate the potential synergies between LLMs and ALife, drawing on a large body of research in the two fields. We explore the potential of LLMs as tools for ALife research, for example, as operators for evolutionary computation or the generation of open-ended environments. Reciprocally, principles of ALife, such as self-organization, collective intelligence and evolvability can provide an opportunity for shaping the development and functionalities of LLMs, leading to more adaptive and responsive models. By investigating this dynamic interplay, the paper aims to inspire innovative crossover approaches for both ALife and LLM research. Along the way, we examine the extent to which LLMs appear to increasingly exhibit properties such as emergence or collective intelligence, expanding beyond their original goal of generating text, and potentially redefining our perception of lifelike intelligence in artificial systems.

JWST is uncovering the properties of ever increasing numbers of galaxies at z>6, during the epoch of reionization. Connecting these observed populations to the process of reionization requires understanding how efficiently they produce Lyman continuum (LyC) photons and what fraction (fesc) of these photons escape into the intergalactic medium. By applying the Cox proportional hazards model, a survival analysis technique, to the Low-redshift Lyman Continuum Survey (LzLCS), we develop new, empirical, multivariate predictions for fesc. The models developed from the LzLCS reproduce the observed fesc for z~3 samples, which suggests that LyC emitters may share similar properties at low and high redshift. Our best-performing models for the z~3 galaxies include information about dust attenuation, ionization, and/or morphology. We then apply these models to z$\gtrsim$6 galaxies. For large photometric samples, we find a median predicted fesc=0.047-0.14. For smaller spectroscopic samples, which may include stronger emission line galaxies, we find that $\geq$33% of the galaxies have fesc >0.2, and we identify several candidate extreme leakers with fesc $\geq$0.5. The current samples show no strong trend between predicted fesc and UV magnitude, but limited spectroscopic information makes this result uncertain. Multivariate predictions can give significantly different results from single variable predictions, and the predicted fesc for high-redshift galaxies can differ significantly depending on whether star formation rate surface density or radius is used as a measure of galaxy morphology. We provide all parameters necessary to predict fesc for additional samples of high-redshift galaxies using these models., Comment: Accepted for publication in ApJ. 33 pages, 9 figures, 10 tables, plus appendix

To understand how galaxies reionized the universe, we must determine how the escape fraction of Lyman Continuum (LyC) photons (fesc) depends on galaxy properties. Using the z~0.3 Low-redshift Lyman Continuum Survey (LzLCS), we develop and analyze new multivariate predictors of fesc. These predictions use the Cox proportional hazards model, a survival analysis technique that incorporates both detections and upper limits. Our best model predicts the LzLCS fesc detections with a root-mean-square (RMS) scatter of 0.31 dex, better than single-variable correlations. According to ranking techniques, the most important predictors of fesc are the equivalent width (EW) of Lyman-series absorption lines and the UV dust attenuation, which track line-of-sight absorption due to HI and dust. The HI absorption EW is uniquely crucial for predicting fesc for the strongest LyC emitters, which show properties similar to weaker LyC emitters and whose high fesc may therefore result from favorable orientation. In the absence of HI information, star formation rate surface density ($\Sigma_{\rm SFR}$) and [O III]/[O II] ratio are the most predictive variables and highlight the connection between feedback and fesc. We generate a model suitable for z>6, which uses only the UV slope, $\Sigma_{\rm SFR}$, and [O III]/[O II]. We find that $\Sigma_{\rm SFR}$ is more important in predicting fesc at higher stellar masses, whereas [O III]/[O II] plays a greater role at lower masses. We also analyze predictions for other parameters, such as the ionizing-to-non ionizing flux ratio and Ly=alpha escape fraction. These multivariate models represent a promising tool for predicting fesc at high redshift., Comment: Accepted for publication in ApJ. 34 pages + appendix, 12 figures

Deep neural networks, costly to train and rich in intellectual property value, are increasingly threatened by model extraction attacks that compromise their confidentiality. Previous attacks have succeeded in reverse-engineering model parameters up to a precision of float64 for models trained on random data with at most three hidden layers using cryptanalytical techniques. However, the process was identified to be very time consuming and not feasible for larger and deeper models trained on standard benchmarks. Our study evaluates the feasibility of parameter extraction methods of Carlini et al. [1] further enhanced by Canales-Mart\'inez et al. [2] for models trained on standard benchmarks. We introduce a unified codebase that integrates previous methods and reveal that computational tools can significantly influence performance. We develop further optimisations to the end-to-end attack and improve the efficiency of extracting weight signs by up to 14.8 times compared to former methods through the identification of easier and harder to extract neurons. Contrary to prior assumptions, we identify extraction of weights, not extraction of weight signs, as the critical bottleneck. With our improvements, a 16,721 parameter model with 2 hidden layers trained on MNIST is extracted within only 98 minutes compared to at least 150 minutes previously. Finally, addressing methodological deficiencies observed in previous studies, we propose new ways of robust benchmarking for future model extraction attacks.

Visual Question Answering (VQA) models play a critical role in enhancing the perception capabilities of autonomous driving systems by allowing vehicles to analyze visual inputs alongside textual queries, fostering natural interaction and trust between the vehicle and its occupants or other road users. This study investigates the attention patterns of humans compared to a VQA model when answering driving-related questions, revealing disparities in the objects observed. We propose an approach integrating filters to optimize the model's attention mechanisms, prioritizing relevant objects and improving accuracy. Utilizing the LXMERT model for a case study, we compare attention patterns of the pre-trained and Filter Integrated models, alongside human answers using images from the NuImages dataset, gaining insights into feature prioritization. We evaluated the models using a Subjective scoring framework which shows that the integration of the feature encoder filter has enhanced the performance of the VQA model by refining its attention mechanisms.

We present the findings of the first NeurIPS competition on unlearning, which sought to stimulate the development of novel algorithms and initiate discussions on formal and robust evaluation methodologies. The competition was highly successful: nearly 1,200 teams from across the world participated, and a wealth of novel, imaginative solutions with different characteristics were contributed. In this paper, we analyze top solutions and delve into discussions on benchmarking unlearning, which itself is a research problem. The evaluation methodology we developed for the competition measures forgetting quality according to a formal notion of unlearning, while incorporating model utility for a holistic evaluation. We analyze the effectiveness of different instantiations of this evaluation framework vis-a-vis the associated compute cost, and discuss implications for standardizing evaluation. We find that the ranking of leading methods remains stable under several variations of this framework, pointing to avenues for reducing the cost of evaluation. Overall, our findings indicate progress in unlearning, with top-performing competition entries surpassing existing algorithms under our evaluation framework. We analyze trade-offs made by different algorithms and strengths or weaknesses in terms of generalizability to new datasets, paving the way for advancing both benchmarking and algorithm development in this important area.

In differentially private (DP) machine learning, the privacy guarantees of DP mechanisms are often reported and compared on the basis of a single $(\varepsilon, \delta)$-pair. This practice overlooks that DP guarantees can vary substantially even between mechanisms sharing a given $(\varepsilon, \delta)$, and potentially introduces privacy vulnerabilities which can remain undetected. This motivates the need for robust, rigorous methods for comparing DP guarantees in such cases. Here, we introduce the $\Delta$-divergence between mechanisms which quantifies the worst-case excess privacy vulnerability of choosing one mechanism over another in terms of $(\varepsilon, \delta)$, $f$-DP and in terms of a newly presented Bayesian interpretation. Moreover, as a generalisation of the Blackwell theorem, it is endowed with strong decision-theoretic foundations. Through application examples, we show that our techniques can facilitate informed decision-making and reveal gaps in the current understanding of privacy risks, as current practices in DP-SGD often result in choosing mechanisms with high excess privacy vulnerabilities., Comment: ICML 2024

The escape fraction of LyC ionizing radiation $f_{LyC}^{esc}$ is crucial for understanding reionization, yet impossible to measure at z $\gtrsim$ 5.3. Recently, studies have focused on calibrating indirect indicators of $f_{LyC}^{esc}$ at z $\sim$ 0.3, finding that Ly$\alpha$ is closely linked to it. What is still unclear is whether the LyC - Ly$\alpha$ relation evolves with redshift, and if Ly$\alpha$ is truly applicable as an $f_{LyC}^{esc}$ indicator during the reionization epoch. In this study, we investigate seven $-21 \lesssim M_{UV} \lesssim -19$ gravitationally lensed galaxies from the BELLS GALLERY survey at z $\sim$ 2.3. Our targets have rest-frame Ly$\alpha$ equivalent widths ranging from 10 \r{A} to 100 \r{A} and low dust content ($-2.5 \lesssim \beta \lesssim -1.9$), both indicative of high LyC escape. Surprisingly, direct estimates of $f_{LyC}^{esc}$ using Hubble Space Telescope imaging with F275W and F225W reveal that our targets are not LyC emitters, with absolute $f_{LyC}^{esc}$ $\lesssim$ 6.5% at 3$\sigma$ significance (with two sources having absolute $f_{LyC}^{esc}$(3$\sigma$) $\lesssim$ 10% and $\lesssim$ 16%). The low $f_{LyC}^{esc}$, coupled with the high Ly$\alpha$ escape fraction and equivalent width could potentially be attributed to the redshift evolution of the neutral hydrogen column density and dust content. Our results challenge previous studies based on local samples, suggesting that the extrapolation of z ~ 0 Ly$\alpha$-based LyC indirect estimators into the reionization epoch might not be correct., Comment: 19 Pages, 8 Figures. Submitted to ApJ

Most studies of quasi-periodic pulsations in solar flares have identified characteristic periods in the 5 - 300s range. Due to observational limitations there have been few attempts to probe the < 5s period regime and understand the prevalence of such short-period quasi-periodic pulsations. However, the Fermi Gamma-ray Burst Monitor (GBM) has observed approximately 1500 solar flares to date in high cadence 16 Hz burst mode, providing us with an opportunity to study short-period quasi-periodic pulsations at X-ray energies. We systematically analyse every solar flare observed by Fermi/GBM in burst mode, estimating the prevalence of quasi-periodic pulsations in multiple X-ray energy bands. To better understand these results, we complement this with analysis of synthetic solar flare lightcurves, both with and without oscillatory signals present. Using these synthetic lightcurves, we can understand the likely false alarm and true positive rates in the real solar GBM data. We do not find strong evidence for widespread short-period quasi-periodic pulsations, indicating either a low base occurrence rate of such signatures or that their typical signal-to-noise ratios must be low - less than 1 - in Fermi/GBM data. Finally, we present a selection of the most interesting potential quasi-periodic pulsation events that were identified in the GBM solar X-ray data., Comment: Accepted for publication in the Astrophysical Journal. 14 pages, 14 figures

Data sets with imbalanced class sizes, where one class size is much smaller than that of others, occur exceedingly often in many applications, including those with biological foundations, such as disease diagnosis and drug discovery. Therefore, it is extremely important to be able to identify data elements of classes of various sizes, as a failure to do so can result in heavy costs. Nonetheless, many data classification procedures do not perform well on imbalanced data sets as they often fail to detect elements belonging to underrepresented classes. In this work, we propose the BTDT-MBO algorithm, incorporating Merriman-Bence-Osher (MBO) approaches and a bidirectional transformer, as well as distance correlation and decision threshold adjustments, for data classification tasks on highly imbalanced molecular data sets, where the sizes of the classes vary greatly. The proposed technique not only integrates adjustments in the classification threshold for the MBO algorithm in order to help deal with the class imbalance, but also uses a bidirectional transformer procedure based on an attention mechanism for self-supervised learning. In addition, the model implements distance correlation as a weight function for the similarity graph-based framework on which the adjusted MBO algorithm operates. The proposed method is validated using six molecular data sets and compared to other related techniques. The computational experiments show that the proposed technique is superior to competing approaches even in the case of a high class imbalance ratio.

Gas kinematics affect the radiative transfer and escape of hydrogen Lyman-$\alpha$ (Ly$\alpha$) emission from galaxies. We investigate this interplay empirically by relating the ionised gas kinematics of 42 galaxies in the extended Ly$\alpha$ Reference Sample with their Ly$\alpha$ escape fractions, $f_\rm{esc}$, Ly$\alpha$ equivalent widths, $\rm{EW}_\rm{Ly\alpha}$, and Ly$\alpha$ luminosities, $L_\rm{Ly\alpha}$. To this aim we use PMAS integral-field spectroscopic observations of the Balmer-$\alpha$ line. We calculate shearing velocities, $v_\rm{shear}$, and intrinsic velocity dispersions, $\sigma_0^\rm{obs}$ (empirically corrected for beam-smearing effects), as global kinematical measures for each galaxy. The sample is characterised by highly turbulent motions and more than half of the sample shows dispersion dominated kinematics. We uncover clear trends between Ly$\alpha$ observables and global kinematical statistics. We discuss statistically the importance of $v_\rm{shear}$, $\sigma_0^\rm{obs}$, and $v_\rm{shear}/\sigma_0^\rm{obs}$ for regulating the Ly$\alpha$ observables in comparison to other galaxy parameters. It emerges that $\sigma_0^\rm{obs}$ is the dominating parameter for $\rm{EW}_\rm{Ly\alpha}$ and that is as important as nebular extinction, gas covering fraction, and ionising photon production efficiency in regulating $f_\rm{esc}$. A simple scenario where the starburst age is simultaneously regulating turbulence, $\rm{EW}_\rm{Ly\alpha}$, and $f_\rm{esc}$ does not find support by our observations. However, we show that the small scale distribution of dust appears to be influenced by turbulence in some galaxies. In support of our observational result we discuss how turbulence is theoretically expected to play a significant role in modulating $f_\rm{esc}$. (abridged), Comment: Accepted version for publication in A&A

Feedback driven massive outflows play a crucial role in galaxy evolution by regulating star formation and influencing the dynamics of surrounding media. Extracting outflow properties from spectral lines is a notoriously difficult process for a number of reasons, including the possibility that a substantial fraction of the outflow is carried by dense gas in a very narrow range in velocity. This gas can hide in spectra with insufficient resolution. Empirically motivated analysis based on the Apparent Optical Depth method, commonly used in the literature, neglects the contribution of this gas, and may therefore underestimate the true gas column density. More complex semi-analytical line transfer (e.g., SALT) models, on the other hand, allow for the presence of this gas by modeling the radial density and velocity of the outflows as power laws. Here we compare the two approaches to quantify the uncertainties in the inferences of outflow properties based on 1-D "down-the-barrel" using the UV spectra of the CLASSY galaxy sample. We find that empirical modeling may significantly underestimate the column densities relative to SALT analysis, particularly in the optically thick regime. We use simulations to show that the main reason for this discrepancy is the presence of large amount of dense material at low velocities, which can be hidden by the finite spectral resolution of the data. The SALT models in turn could over-estimate the column densities if the assumed power laws of the density profiles strong are not a property of actual outflows., Comment: Accepted to the Astrophysical Journal

Empirical evidence for a gap between the computational powers of classical and quantum computers has been provided by experiments that sample the output distributions of two-dimensional quantum circuits. Many attempts to close this gap have utilized classical simulations based on tensor network techniques, and their limitations shed light on the improvements to quantum hardware required to frustrate classical simulability. In particular, quantum computers having in excess of $\sim 50$ qubits are primarily vulnerable to classical simulation due to restrictions on their gate fidelity and their connectivity, the latter determining how many gates are required (and therefore how much infidelity is suffered) in generating highly-entangled states. Here, we describe recent hardware upgrades to Quantinuum's H2 quantum computer enabling it to operate on up to $56$ qubits with arbitrary connectivity and $99.843(5)\%$ two-qubit gate fidelity. Utilizing the flexible connectivity of H2, we present data from random circuit sampling in highly connected geometries, doing so at unprecedented fidelities and a scale that appears to be beyond the capabilities of state-of-the-art classical algorithms. The considerable difficulty of classically simulating H2 is likely limited only by qubit number, demonstrating the promise and scalability of the QCCD architecture as continued progress is made towards building larger machines., Comment: Includes minor updates to the text and an updated author list to include researchers who made technical contributions in upgrading the machine to 56 qubits but were left off the original version by mistake

Quantum error correction protects logical quantum information against environmental decoherence by encoding logical qubits into entangled states of physical qubits. One of the most important near-term challenges in building a scalable quantum computer is to reach the break-even point, where logical quantum circuits on error-corrected qubits achieve higher fidelity than equivalent circuits on uncorrected physical qubits. Using Quantinuum's H2 trapped-ion quantum processor, we encode the GHZ state in four logical qubits with fidelity $ 99.5 \pm 0.15 \% \le F \le 99.7 \pm 0.1\% $ (after postselecting on over 98% of outcomes). Using the same quantum processor, we can prepare an uncorrected GHZ state on four physical qubits with fidelity $97.8 \pm 0.2 \% \le F\le 98.7\pm 0.2\%$. The logical qubits are encoded in a $[\![ 25,4,3 ]\!]$ Tanner-transformed long-range-enhanced surface code. Logical entangling gates are implemented using simple swap operations. Our results are a first step towards realizing fault-tolerant quantum computation with logical qubits encoded in geometrically nonlocal quantum low-density parity check codes., Comment: 13 pages, 9 figures, 1 table. v2 changes: added classical simulation data and appendices

Modern Machine Learning models are expensive IP and business competitiveness often depends on keeping this IP confidential. This in turn restricts how these models are deployed -- for example it is unclear how to deploy a model on-device without inevitably leaking the underlying model. At the same time, confidential computing technologies such as Multi-Party Computation or Homomorphic encryption remain impractical for wide adoption. In this paper we take a different approach and investigate feasibility of ML-specific mechanisms that deter unauthorized model use by restricting the model to only be usable on specific hardware, making adoption on unauthorized hardware inconvenient. That way, even if IP is compromised, it cannot be trivially used without specialised hardware or major model adjustment. In a sense, we seek to enable cheap locking of machine learning models into specific hardware. We demonstrate that locking mechanisms are feasible by either targeting efficiency of model representations, such making models incompatible with quantisation, or tie the model's operation on specific characteristics of hardware, such as number of cycles for arithmetic operations. We demonstrate that locking comes with negligible work and latency overheads, while significantly restricting usability of the resultant model on unauthorized hardware., Comment: 10 pages, 2 figures of main text; 14 pages, 16 figures of appendices

The ability to shop independently, especially in grocery stores, is important for maintaining a high quality of life. This can be particularly challenging for people with visual impairments (PVI). Stores carry thousands of products, with approximately 30,000 new products introduced each year in the US market alone, presenting a challenge even for modern computer vision solutions. Through this work, we present a proof-of-concept socially assistive robotic system we call ShelfHelp, and propose novel technical solutions for enhancing instrumented canes traditionally meant for navigation tasks with additional capability within the domain of shopping. ShelfHelp includes a novel visual product locator algorithm designed for use in grocery stores and a novel planner that autonomously issues verbal manipulation guidance commands to guide the user during product retrieval. Through a human subjects study, we show the system's success in locating and providing effective manipulation guidance to retrieve desired products with novice users. We compare two autonomous verbal guidance modes achieving comparable performance to a human assistance baseline and present encouraging findings that validate our system's efficiency and effectiveness and through positive subjective metrics including competence, intelligence, and ease of use., Comment: 8 pages, 14 figures and charts

We claim that LLMs can be paired with formal analysis methods to provide accessible, relevant feedback for HRI tasks. While logic specifications are useful for defining and assessing a task, these representations are not easily interpreted by non-experts. Luckily, LLMs are adept at generating easy-to-understand text that explains difficult concepts. By integrating task assessment outcomes and other contextual information into an LLM prompt, we can effectively synthesize a useful set of recommendations for the learner to improve their performance., Comment: Presented at Human-LLM Interaction Workshop at HRI 2024

The workforce will need to continually upskill in order to meet the evolving demands of industry, especially working with robotic and autonomous systems. Current training methods are not scalable and do not adapt to the skills that learners already possess. In this work, we develop a system that automatically assesses learner skill in a quadrotor teleoperation task using temporal logic task specifications. This assessment is used to generate multimodal feedback based on the principles of effective formative feedback. Participants perceived the feedback positively. Those receiving formative feedback viewed the feedback as more actionable compared to receiving summary statistics. Participants in the multimodal feedback condition were more likely to achieve a safe landing and increased their safe landings more over the experiment compared to other feedback conditions. Finally, we identify themes to improve adaptive feedback and discuss and how training for complex psychomotor tasks can be integrated with learning theories., Comment: Under review at Human-Agent Interaction 2024 conference

The Astro2020 Decadal Survey recognizes time-domain astronomy as a key science area over the next decade and beyond. With over 30 years of HST data and the potential for 20 years of JWST operations, these flagship observatories offer an unparalleled prospect for a half-century of space-based observations in the time domain. To take best advantage of this opportunity, STScI charged a working group to solicit community input and formulate strategies to maximize the science return in time-domain astronomy from these two platforms. Here, the HST/JWST Long-Term Monitoring Working Group reports on the input we received and presents our recommendations to enhance the scientific return for time-domain astronomy from HST and JWST. We suggest changes in policies to enable and prioritize long-term science programs of high scientific value. As charged, we also develop recommendations based on community input for a JWST Director's Discretionary Time program to observe high-redshift transients., Comment: 17 pages, 3 figures, minor typos corrected, working group information at https://outerspace.stsci.edu/display/HPR/Long-term+variability+monitoring+strategies+for+HST+and+JWST