Search

We present the Cosmic Evolution Early Release Science (CEERS) Survey, a 77.2 hour Director's Discretionary Early Release Science Program. CEERS demonstrates, tests, and validates efficient extragalactic surveys using coordinated, overlapping parallel observations with the JWST instrument suite, including NIRCam and MIRI imaging, NIRSpec low (R~100) and medium (R~1000) resolution spectroscopy, and NIRCam slitless grism (R~1500) spectroscopy. CEERS targets the Hubble Space Telescope-observed region of the Extended Groth Strip (EGS) field, supported by a rich set of multiwavelength data. CEERS facilitated immediate community science in both of the extragalactic core JWST science drivers ``First Light" and ``Galaxy Assembly," including: 1) The discovery and characterization of large samples of galaxies at z >~ 10 from ~90 arcmin^2 of NIRCam imaging, constraining their abundance and physical nature; 2) Deep spectra of >1000 galaxies, including dozens of galaxies at 63; and 4) Characterizing galaxy mid-IR emission with MIRI to study dust-obscured star-formation and supermassive black hole growth at z~1-3. As a legacy product for the community, the CEERS team has provided several data releases, accompanied by detailed notes on the data reduction procedures and notebooks to aid in reproducibility. In addition to an overview of the survey and quality of the data, we provide science highlights from the first two years with CEERS data., Comment: 38 pages, 13 figures, 6 tables

Thriving communities depend on a strong early childhood education (ECE) system--one where both young children and members of the workforce are served and supported. In recent years, state government leaders have increasingly focused on changing qualifications for specific ECE roles, as increasing credential requirements has sometimes been associated with increasing quality. However, across the country, early childhood educators face significant barriers to economic security and continuing education--all while supporting children, parents, and their communities with specialized education services. This report examines the barriers that even highly qualified early child educators experience in the face of increased credential requirements. Featuring case studies of California and Washington, D.C., the report offers implications for policies that protect educators and advance racial and economic equity in the profession. Included with the report are an Executive Summary encapsulating the findings and a Technical Document describing the methodology, data collection, and quantitative analysis.

Many public workforce training programs lead to industry-recognized, third-party awarded credentials, but little research has been conducted on the economic benefits of these credentials in the labor market. This paper provides quasi-experimental evidence on the labor market returns to industry-recognized credentials connected to community college workforce noncredit training programs. Based on novel data that includes approximately 24,000 working-age adults enrolled in noncredit workforce training programs at the Virginia Community College System, we employ a comparative individual-level fixed effects model to estimate earnings premia net of fixed attributes and earnings time-trends. Our results indicate that earning an industry-recognized credential, on average, increases quarterly earnings by approximately $1,000 and the probability of being employed by 2.4 percentage points, although there is substantial heterogeneity in economic return across different program fields. Back-of-the-envelope calculations suggest that the earnings gains associated with the industry credential obtained through the noncredit workforce training would exceed program costs in just over half a year on average.

The declining cost of solar photovoltaics (PV) combined with strong federal and state-level incentives have resulted in a high number of residential solar PV installations in the US. However, these installations are concentrated in particular regions, such as California, and demographics, such as high-income Asian neighborhoods. This inequitable distribution creates an illusion that further increasing residential solar installations will become increasingly challenging. Furthermore, while the inequity in solar installations has received attention, no prior comprehensive work has been done on understanding whether our current trajectory of residential solar adoption is energy- and carbon-efficient. In this paper, we reveal the hidden energy and carbon cost of the inequitable distribution of existing installations. Using US-based data on carbon offset potential, the amount of avoided carbon emissions from using rooftop PV instead of electric grid energy, and the number of existing solar installations, we surprisingly observe that locations and demographics with a higher carbon offset potential have fewer existing installations. For instance, neighborhoods with relatively higher black population have 7.4% higher carbon offset potential than average but 36.7% fewer installations; lower-income neighborhoods have 14.7% higher potential and 47% fewer installations. We propose several equity- and carbon-aware solar siting strategies. In evaluating these strategies, we develop Sunsight, a toolkit that combines simulation/visualization tools and our relevant datasets, which we are releasing publicly. Our projections show that a multi-objective siting strategy can address two problems at once; namely, it can improve societal outcomes in terms of distributional equity and simultaneously improve the carbon-efficiency (i.e., climate impact) of current installation trends by up to 39.8%., Comment: 11 pages, 9 figures, E-Energy 2024

We predict the properties of stellar halos in galaxies of present-day virial mass $10^8 < M_{200} < 10^{12} {\rm M_\odot}$ by combining the GALFORM semi-analytic model of galaxy formation, the COCO cosmological N-body simulation, and the STINGS particle tagging technique. Galaxies in low mass halos have a wide range of stellar halo properties. Their diversity is much greater at fixed stellar mass than fixed virial mass. In the least massive DM halos capable of supporting galaxy formation, accreted mass fractions are < 10 per cent, and the typical density profile of accreted stars is similar to that of stars formed in situ. In low mass galaxies, the radial scale lengths of accreted stellar halos are smaller than those of stellar disks formed in situ, but the accreted component is still diffuse, not bulge-like. At the scale of galaxies like M33, the surface density of accreted stars exceeds that of the in situ component at ~15kpc; the accreted surface density at this radius is $3.5\lesssim \log_{10} M_\star/{\rm M_\odot} \mathrm{kpc^{-2}} \lesssim 6$, varying systematically with virial mass. We compare our predictions to observations and other cosmological simulations. A small but significant number of $\sim 10^{12} {\rm M_\odot}$ halos with exceptionally inefficient star formation -- "failed Milky Ways" -- are more prominent in our model than others; the true abundance of this population is a potential constraint on galaxy formation physics and could be measured by low surface brightness surveys targeting field galaxies with M33-like stellar mass. The stellar particle data for our simulation is publicly available., Comment: 25 pages, 21 figures (appendix 5 pages, 4 figures). Submitted to MNRAS. See appendix A and https://github.com/nthu-ga/stings-data for public simulation data

From $>1000$ orbits of HST imaging, we present deep homogeneous resolved star color-magnitude diagrams that reach the oldest main sequence turnoff and uniformly measured star formation histories (SFHs) of 36 dwarf galaxies ($-6 \ge M_V \ge -17$) associated with the M31 halo, and for 10 additional fields in M31, M33, and the Giant Stellar Stream. From our SFHs we find: i) the median stellar age and quenching epoch of M31 satellites correlate with galaxy luminosity and galactocentric distance. Satellite luminosity and present-day distance from M31 predict the satellite quenching epoch to within $1.8$ Gyr at all epochs. This tight relationship highlights the fundamental connection between satellite halo mass, environmental history, and star formation duration. ii) There is no difference between the median SFH of galaxies on and off the great plane of Andromeda satellites. iii) $\sim50$\% of our M31 satellites show prominent ancient star formation ($>12$ Gyr ago) followed by delayed quenching ($8-10$ Gyr ago), which is not commonly observed among the MW satellites. iv) A comparison with TNG50 and FIRE-2 simulated satellite dwarfs around M31-like hosts show that some of these trends (dependence of SFH on satellite luminosity) are reproduced in the simulations while others (dependence of SFH on galactocentric distance, presence of the delayed-quenching population) are weaker or absent. We provide all photometric catalogs and SFHs as High-Level Science Products on MAST., Comment: Accepted for publication on ApJ. 47 pages, 24 figures, 12 tables. Corresponding HLSP data can be retrieved at: https://archive.stsci.edu/hlsp/m31-satellites

The Neural Contextual Reinforcement Framework introduces an innovative approach to enhancing the logical coherence and structural consistency of text generated by large language models. Leveraging reinforcement learning principles, the framework integrates custom reward functions and dynamic context alignment mechanisms to address challenges inherent in maintaining long-range dependencies across extended sequences. The architecture incorporates multi-head attention layers and hierarchical encoding modules, enabling the model to produce outputs that align closely with human expectations of logical structure and semantic flow. Quantitative evaluations across diverse datasets demonstrate substantial improvements in coherence metrics, perplexity reduction, and semantic alignment, showcasing the framework's ability to outperform baseline models in both general and domain-specific tasks. Qualitative analyses further highlight the framework's capacity to generate text with improved narrative clarity and reduced redundancy, reflecting its effectiveness in balancing fluency with structural precision. In addition to its performance gains, the framework exhibits robustness in handling noisy input data and scalability across varying model sizes, reinforcing its versatility in practical applications. Experimental results reveal that optimal context window sizes significantly influence coherence outcomes, showing the importance of architectural flexibility in adapting to diverse linguistic structures. Cross-lingual performance evaluations affirm the framework's adaptability to multiple languages, extending its utility beyond monolingual contexts. Resource efficiency analyses indicate a reduction in computational overhead compared to traditional approaches, emphasizing the practicality of the framework for large-scale deployment.

Reliably generating structured outputs has become a critical capability for modern language model (LM) applications. Constrained decoding has emerged as the dominant technology across sectors for enforcing structured outputs during generation. Despite its growing adoption, little has been done with the systematic evaluation of the behaviors and performance of constrained decoding. Constrained decoding frameworks have standardized around JSON Schema as a structured data format, with most uses guaranteeing constraint compliance given a schema. However, there is poor understanding of the effectiveness of the methods in practice. We present an evaluation framework to assess constrained decoding approaches across three critical dimensions: efficiency in generating constraint-compliant outputs, coverage of diverse constraint types, and quality of the generated outputs. To facilitate this evaluation, we introduce JSONSchemaBench, a benchmark for constrained decoding comprising 10K real-world JSON schemas that encompass a wide range of constraints with varying complexity. We pair the benchmark with the existing official JSON Schema Test Suite and evaluate six state-of-the-art constrained decoding frameworks, including Guidance, Outlines, Llamacpp, XGrammar, OpenAI, and Gemini. Through extensive experiments, we gain insights into the capabilities and limitations of constrained decoding on structured generation with real-world JSON schemas. Our work provides actionable insights for improving constrained decoding frameworks and structured generation tasks, setting a new standard for evaluating constrained decoding and structured generation. We release JSONSchemaBench at https://github.com/guidance-ai/jsonschemabench

We consider `off-axis' electromagnetic wave propagation down the homogeneous direction of a low-loss two-dimensional periodic dielectric (or photonic crystal fibre) near the light line of the low-dielectric material constituent. Numerous physical and numerical experiments demonstrate the presence of photonic band gaps in the vicinity of this `critical' light line. We mathematically analyse the existence of photonic band gaps near the line and characterise them in terms of frequency gaps in the spectrum of the Maxwell equations restricted to the line. We apply the results to both one-dimensional photonic crystal fibres, and a genuinely two-dimensional photonic crystal fibres with `thin' inclusions, namely ``ARROW'' fibres. In the case of ARROW fibres, by an asymptotic analysis, in terms of the small inclusion parameter, we demonstrate the existence of low frequency photonic band gaps.

This paper presents an integrated system that transforms symbolic music scores into expressive piano performance audio. By combining a Transformer-based Expressive Performance Rendering (EPR) model with a fine-tuned neural MIDI synthesiser, our approach directly generates expressive audio performances from score inputs. To the best of our knowledge, this is the first system to offer a streamlined method for converting score MIDI files lacking expression control into rich, expressive piano performances. We conducted experiments using subsets of the ATEPP dataset, evaluating the system with both objective metrics and subjective listening tests. Our system not only accurately reconstructs human-like expressiveness, but also captures the acoustic ambience of environments such as concert halls and recording studios. Additionally, the proposed system demonstrates its ability to achieve musical expressiveness while ensuring good audio quality in its outputs., Comment: Accepted by ICASSP 2025

We present the strongest limits to date on the mixing angle, $\theta$, with which a new scalar particle, $S$, mixes with the Higgs field in the mass range $100$ $MeV

Science laboratory automation enables accelerated discovery in life sciences and materials. However, it requires interdisciplinary collaboration to address challenges such as robust and flexible autonomy, reproducibility, throughput, standardization, the role of human scientists, and ethics. This article highlights these issues, reflecting perspectives from leading experts in laboratory automation across different disciplines of the natural sciences.

Global solar photospheric magnetic maps play a critical role in solar and heliospheric physics research. Routine magnetograph measurements of the field occur only along the Sun-Earth line, leaving the far-side of the Sun unobserved. Surface Flux Transport (SFT) models attempt to mitigate this by modeling the surface evolution of the field. While such models have long been established in the community (with several releasing public full-Sun maps), none are open source. The Open Source Flux Transport (OFT) model seeks to fill this gap by providing an open and user-extensible SFT model that also builds on the knowledge of previous models with updated numerical and data acquisition/assimilation methods along with additional user-defined features. In this first of a series of papers on OFT, we introduce its computational core: the High-performance Flux Transport (HipFT) code (github.com/predsci/hipft). HipFT implements advection, diffusion, and data assimilation in a modular design that supports a variety of flow models and options. It can compute multiple realizations in a single run across model parameters to create ensembles of maps for uncertainty quantification and is high-performance through the use of multi-CPU and multi-GPU parallelism. HipFT is designed to enable users to easily write extensions, enhancing its flexibility and adaptability. We describe HipFT's model features, validations of its numerical methods, performance of its parallel and GPU-accelerated code implementation, analysis/post-processing options, and example use cases., Comment: 32 pages, 16 figures

We analyse a family of lattices considered by Conway and Sloane and show that the corresponding Epstein zeta function attains a minimum for the body-centred cubic lattice.

The programming landscape is nowadays being reshaped by the advent of Large Language Models (LLMs) able to automate code-related tasks related to code implementation (e.g., code completion) and comprehension (e.g., code summarization). Such a paradigm shift comes with a number of implications related to how software will be written, maintained, and evolved. Also, these LLMs are extremely expensive to train, posing questions on their sustainability over time. Given their training cost, their ability to generalize, namely their ability to work on task instances different from those on which they have been trained, is an aspect worth being investigated. Previous work already showed that transformer models can successfully support code completion in a cross-project setting. However, it is unclear whether LLM are able to generalize to inputs having lengths not seen during training. For example, it is known that training a model on short instances allows to substantially reduce the training cost. However, the extent to which such a model would provide good performance on sequences having lengths not seen during training is not known. Many recent works in Natural Language Processing (NLP) tackled this problem in the context of decoder-only LLMs, i.e., xPOS and ALiBi. To assess if these solutions extend to encoder-decoder LLMs usually adopted in the code-related tasks, we present a large empirical study evaluating this generalization property of these and other encoding schemes proposed in the literature, namely Sinusoidal, xPOS, ALiBi, and T5. We found that none of these solutions successfully generalize to unseen lengths and that the only safe solution is to ensure the representativeness in the training set of all lengths likely to be encountered at inference time., Comment: Accepted for publication at ICSME 2024

Material flow analyses (MFAs) provide insight into supply chain level opportunities for resource efficiency. MFAs can be represented as networks with nodes that represent materials, processes, sectors or locations. MFA network structure uncertainty (i.e., the existence or absence of flows between nodes) is pervasive and can undermine the reliability of the flow predictions. This article investigates MFA network structure uncertainty by proposing candidate node-and-flow structures and using Bayesian model selection to identify the most suitable structures and Bayesian model averaging to quantify the parametric mass flow uncertainty. The results of this holistic approach to MFA uncertainty are used in conjunction with the input-output (I/O) method to make risk-informed resource efficiency recommendation. These techniques are demonstrated using a case study on the U.S. steel sector where 16 candidate structures are considered. Model selection highlights 2 networks as most probable based on data collected from the United States Geological Survey and the World Steel Association. Using the I/O method, we then show that the construction sector accounts for the greatest mean share of domestic U.S. steel industry emissions while the automotive and steel products sectors have the highest mean emissions per unit of steel used in the end-use sectors. The uncertainty in the results is used to analyze which end-use sector should be the focus of demand reduction efforts under different appetites for risk. This article's methods generate holistic and transparent MFA uncertainty that account for structural uncertainty, enabling decisions whose outcomes are more robust to the uncertainty., Comment: 17 pages of main paper and 26 pages of supporting information

We characterize the permutations of $\mathbb{F}_q$ whose graph minimizes the number of collinear triples and describe the lexicographically-least one, affirming a conjecture of Cooper-Solymosi. This question is closely connected to Dudeney's No-3-in-a-Line problem, the Heilbronn triangle problem, and the structure of finite plane Kakeya sets. We discuss a connection with complete sets of mutually orthogonal latin squares and state a few open problems primarily about general finite affine planes., Comment: 8 pages, 0 figures

The Berry phase-related nontrivial electronic band geometries can significantly influence bulk and edge plasma waves, or plasmons, resulting in their non-reciprocal propagation and opening new opportunities for plasmonics. In the present work, we extend the hydrodynamic framework to describe the scattering of plasma waves in a Dirac electron fluid off a circular region with an induced nonzero anomalous Hall response, i.e. a Berry flux target. We demonstrate that the scattering has a giant asymmetry or skewness and exhibits a series of resonances. The latter appears due to a chiral non-topological trapped mode circulating the target. We discuss possible experimental realizations, including the surface of a topological insulator film and graphene irradiated by the circularly polarized beam., Comment: 10 pages, 5 figures

We construct $S^r$-colored knot Floer homologies and prove that they satisfy categorified recurrence relations. The associated Euler characteristic implies $q$-holonomicity of the corresponding sequence of colored Alexander polynomials, in analogy with the AJ conjecture for colored Jones polynomials.

As technology advances, the integration of physical, virtual, and social worlds has led to a complex landscape of ``Realities'' such as Virtual Reality (VR), Augmented Reality (AR), metaverse, spatial computing, and other emerging paradigms. This paper builds upon and refines the concept of eXtended Reality (XR) as the unifying framework that not only interpolates across these diverse realities but also extrapolates (extends) to create entirely new possibilities. XR is the ``physical spatial metaverse,'' bridging the physical world, the virtual world of artificial intelligence, and the social world of human interaction. These three worlds define the Socio-Cyber-Physical Taxonomy of XR that allows us to identify underexplored research areas such as Diminished Reality (DR), and chart future directions to {\bf advance technology for people and planet}. We highlight the six core properties of XR for applications in sustainability, healthcare, frontline work, and daily life. Central to this vision is the development of AI-driven wearable technologies, such as the smart eyeglass, that sustainably extend human capabilities., Comment: 9 pages, 9 figures, getting ready for the panel discussion at IEEE ICCE

We present the Active Galactic Nuclei (AGN) catalog from the fourth data release (HDR4) of the Hobby-Eberly Telescope Dark Energy Experiment Survey (HETDEX). HETDEX is an untargeted spectroscopic survey. HDR4 contains 345,874 Integral Field Unit (IFU) observations from January 2017 to August 2023 covering an effective area of 62.9 deg2. With no imaging pre-selection, our spectroscopic confirmed AGN sample includes low-luminosity AGN, narrow-line AGN, and/or red AGN down to g~25. This catalog has 15,940 AGN across the redshifts of z=0.1~4.6, giving a raw AGN number density of 253.4 deg-2. Among them, 10,499 (66%) have redshifts either confirmed by line pairs or matched to the Sloan Digital Sky Survey Quasar Catalog. For the remaining 5,441 AGN, 2,083 are single broad line AGN candidates, while the remaining 3,358 are single intermediate broad line (full width at half maximum, FWHM ~ 1200 km s-1) AGN candidates. A total of 4,060 (39%) of the 10,499 redshift-confirmed AGN have emission-line regions $3\sigma$ more extended than the image quality which could be strong outflows blowing into the outskirts of the host galaxies or ionized intergalactic medium., Comment: 15 pages, 7 figures, 2 tables, accepted by the Astrophysical Journal Supplement Series

Optical dynamic nuclear polarization (DNP) offers an attractive approach to enhancing the sensitivity of nuclear magnetic resonance (NMR) spectroscopy. Efficient, optically-generated electron polarization can be leveraged to operate across a broad range of temperatures and magnetic fields, making it particularly appealing for applications requiring high DNP efficiency or spatial resolution. While a large class of systems hold promise for optical DNP, many candidates display both variable electron polarizability and electron and nuclear T1 relaxation times as functions of magnetic field and temperature. This necessitates tools capable of studying DNP under diverse experimental conditions. To address this, we introduce a cryogenic field cycling instrument that facilitates optical DNP studies across a wide range of magnetic fields (10mT to 9.4T) and temperatures (10K to 300K). Continuous cryogen replenishment enables sustained, long-term operation. Additionally, the system supports the ability to manipulate and probe hyperpolarized nuclear spins via pulse sequences involving millions of RF pulses. We describe innovations in the device design and demonstrate its operation on a model system of 13C nuclear spins in diamond polarized through optically pumped nitrogen vacancy (NV) centers. We anticipate the use of the instrument for a broad range of optical DNP systems and studies., Comment: 10 figures, 11 pages (including references)

In Tennenholtz's program equilibrium, players of a game submit programs to play on their behalf. Each program receives the other programs' source code and outputs an action. This can model interactions involving AI agents, mutually transparent institutions, or commitments. Tennenholtz (2004) proves a folk theorem for program games, but the equilibria constructed are very brittle. We therefore consider simulation-based programs -- i.e., programs that work by running opponents' programs. These are relatively robust (in particular, two programs that act the same are treated the same) and are more practical than proof-based approaches. Oesterheld's (2019) $\epsilon$Grounded$\pi$Bot is such an approach. Unfortunately, it is not generally applicable to games of three or more players, and only allows for a limited range of equilibria in two player games. In this paper, we propose a generalisation to Oesterheld's (2019) $\epsilon$Grounded$\pi$Bot. We prove a folk theorem for our programs in a setting with access to a shared source of randomness. We then characterise their equilibria in a setting without shared randomness. Both with and without shared randomness, we achieve a much wider range of equilibria than Oesterheld's (2019) $\epsilon$Grounded$\pi$Bot. Finally, we explore the limits of simulation-based program equilibrium, showing that the Tennenholtz folk theorem cannot be attained by simulation-based programs without access to shared randomness.

This Letter presents an investigation of low-energy electron-neutrino interactions in the Fermilab Booster Neutrino Beam by the MicroBooNE experiment, motivated by the excess of electron-neutrino-like events observed by the MiniBooNE experiment. This is the first measurement to use data from all five years of operation of the MicroBooNE experiment, corresponding to an exposure of $1.11\times 10^{21}$ protons on target, a $70\%$ increase on past results. Two samples of electron neutrino interactions without visible pions are used, one with visible protons and one without any visible protons. MicroBooNE data is compared to two empirical models that modify the predicted rate of electron-neutrino interactions in different variables in the simulation to match the unfolded MiniBooNE low energy excess. In the first model, this unfolding is performed as a function of electron neutrino energy, while the second model aims to match the observed shower energy and angle distributions of the MiniBooNE excess. This measurement excludes an electron-like interpretation of the MiniBooNE excess based on these models at $> 99\%$ CL$_\mathrm{s}$ in all kinematic variables., Comment: 8 pages, 5 figures, 1 table

Encoder-only transformer models such as BERT offer a great performance-size tradeoff for retrieval and classification tasks with respect to larger decoder-only models. Despite being the workhorse of numerous production pipelines, there have been limited Pareto improvements to BERT since its release. In this paper, we introduce ModernBERT, bringing modern model optimizations to encoder-only models and representing a major Pareto improvement over older encoders. Trained on 2 trillion tokens with a native 8192 sequence length, ModernBERT models exhibit state-of-the-art results on a large pool of evaluations encompassing diverse classification tasks and both single and multi-vector retrieval on different domains (including code). In addition to strong downstream performance, ModernBERT is also the most speed and memory efficient encoder and is designed for inference on common GPUs.

Sparse static detector networks in urban environments can be used in efforts to detect illicit radioactive sources, such as stolen nuclear material or radioactive "dirty bombs". We use detailed simulations to evaluate multiple configurations of detector networks and their ability to detect sources moving through a $6\times6$ km$^2$ area of downtown Chicago. A detector network's probability of detecting a source increases with detector density but can also be increased with strategic node placement. We show that the ability to fuse correlated data from a source-carrying vehicle passing by multiple detectors can significantly contribute to the overall detection probability. In this paper we distinguish static sensor deployments operated as networks able to correlate signals between sensors, from deployments operated as arrays where each sensor is operated individually. In particular, we show that additional visual attributes of source-carrying vehicles, such as vehicle color and make, can greatly improve the ability of a detector network to detect illicit sources.

Text-to-image models are vulnerable to the stepwise "Divide-and-Conquer Attack" (DACA) that utilize a large language model to obfuscate inappropriate content in prompts by wrapping sensitive text in a benign narrative. To mitigate stepwise DACA attacks, we propose a two-layer method involving text summarization followed by binary classification. We assembled the Adversarial Text-to-Image Prompt (ATTIP) dataset ($N=940$), which contained DACA-obfuscated and non-obfuscated prompts. From the ATTIP dataset, we created two summarized versions: one generated by a small encoder model and the other by a large language model. Then, we used an encoder classifier and a GPT-4o classifier to perform content moderation on the summarized and unsummarized prompts. When compared with a classifier that operated over the unsummarized data, our method improved F1 score performance by 31%. Further, the highest recorded F1 score achieved (98%) was produced by the encoder classifier on a summarized ATTIP variant. This study indicates that pre-classification text summarization can inoculate content detection models against stepwise DACA obfuscations.

The growing number of satellite constellations in low Earth orbit (LEO) enhances global communications and Earth observation, and support of space commerce is a high priority of many governments. At the same time, the proliferation of satellites in LEO has negative effects on astronomical observations and research, and the preservation of the dark and quiet sky. These satellite constellations reflect sunlight onto optical telescopes, and their radio emission impacts radio observatories, jeopardising our access to essential scientific discoveries through astronomy. The changing visual appearance of the sky also impacts our cultural heritage and environment. Both ground-based observatories and space-based telescopes in LEO are affected, and there are no places on Earth that can escape the effects of satellite constellations given their global nature. The minimally disturbed dark and radio-quiet sky is crucial for conducting fundamental research in astronomy and important public services such as planetary defence, technology development, and high-precision geolocation. Some aspects of satellite deployment and operation are regulated by States and intergovernmental organisations. While regulatory agencies in some States have started to require operators to coordinate with their national astronomy agencies over impacts, mitigation of the impact of space objects on astronomical activities is not sufficiently regulated. To address this issue, the CPS urges States and the international community to take steps to protect the dark and quiet sky as specified in this paper., Comment: This position paper was developed by the IAU Centre for the Protection of the Dark and Quiet Sky from Satellite Constellation Interference (CPS). It can also be downloaded at the CPS website at https://cps.iau.org/news/cps-urges-action-in-first-recommendations-paper/

We introduce the Granite Guardian models, a suite of safeguards designed to provide risk detection for prompts and responses, enabling safe and responsible use in combination with any large language model (LLM). These models offer comprehensive coverage across multiple risk dimensions, including social bias, profanity, violence, sexual content, unethical behavior, jailbreaking, and hallucination-related risks such as context relevance, groundedness, and answer relevance for retrieval-augmented generation (RAG). Trained on a unique dataset combining human annotations from diverse sources and synthetic data, Granite Guardian models address risks typically overlooked by traditional risk detection models, such as jailbreaks and RAG-specific issues. With AUC scores of 0.871 and 0.854 on harmful content and RAG-hallucination-related benchmarks respectively, Granite Guardian is the most generalizable and competitive model available in the space. Released as open-source, Granite Guardian aims to promote responsible AI development across the community. https://github.com/ibm-granite/granite-guardian

We articulate fundamental mismatches between technical methods for machine unlearning in Generative AI, and documented aspirations for broader impact that these methods could have for law and policy. These aspirations are both numerous and varied, motivated by issues that pertain to privacy, copyright, safety, and more. For example, unlearning is often invoked as a solution for removing the effects of targeted information from a generative-AI model's parameters, e.g., a particular individual's personal data or in-copyright expression of Spiderman that was included in the model's training data. Unlearning is also proposed as a way to prevent a model from generating targeted types of information in its outputs, e.g., generations that closely resemble a particular individual's data or reflect the concept of "Spiderman." Both of these goals--the targeted removal of information from a model and the targeted suppression of information from a model's outputs--present various technical and substantive challenges. We provide a framework for thinking rigorously about these challenges, which enables us to be clear about why unlearning is not a general-purpose solution for circumscribing generative-AI model behavior in service of broader positive impact. We aim for conceptual clarity and to encourage more thoughtful communication among machine learning (ML), law, and policy experts who seek to develop and apply technical methods for compliance with policy objectives., Comment: Presented at the 2nd Workshop on Generative AI and Law at ICML (July 2024)

In an influential article Papadimitriou [FOCS 1991] proved that a local search algorithm called WalkSAT finds a satisfying assignment of a satisfiable 2-CNF with $n$ variables in $O(n^2)$ expected time. Variants of the WalkSAT algorithm have become a mainstay of practical SAT solving (e.g., [Hoos and St\"utzle 2000]). In the present article we analyse the expected running time of WalkSAT on random 2-SAT instances. Answering a question raised by Alekhnovich and Ben-Sasson [SICOMP 2007], we show that WalkSAT runs in linear expected time for all clause/variable densities up to the random 2-SAT satisfiability threshold.

The weak-to-strong generalization phenomenon is the driver for important machine learning applications including highly data-efficient learning and, most recently, performing superalignment. While decades of research have resulted in numerous algorithms that produce strong empirical performance, understanding what aspects of data enable weak-to-strong generalization has been understudied. We propose a simple data-centric mechanism that characterizes weak-to-strong generalization: the overlap density. Intuitively, generalization tracks the number of points that contain overlaps, i.e., both easy patterns (learnable by a weak model) and challenging patterns (only learnable by a stronger model), as with such points, weak predictions can be used to learn challenging patterns by stronger models. We provide a practical overlap detection algorithm to find such points in datasets and leverage them to learn, among multiple sources of data, which to query when seeking to maximize overlap density and thereby enhance weak-to-strong generalization. We present a theoretical result showing that the generalization benefit is a function of the overlap density and a regret bound for our data selection algorithm. Empirically, we validate the mechanism and the overlap detection algorithm on a wide array of settings., Comment: 39 pages

We present results from a Keck/DEIMOS survey to study satellite quenching in group environments at $z \sim 0.8$ within the Extended Groth Strip (EGS). We target $11$ groups in the EGS with extended X-ray emission. We obtain high-quality spectroscopic redshifts for group member candidates, extending to depths over an order of magnitude fainter than existing DEEP2/DEEP3 spectroscopy. This depth enables the first spectroscopic measurement of the satellite quiescent fraction down to stellar masses of $\sim 10^{9.5}~{\rm M}_{\odot}$ at this redshift. By combining an infall-based environmental quenching model, constrained by the observed quiescent fractions, with infall histories of simulated groups from the IllustrisTNG100-1-Dark simulation, we estimate environmental quenching timescales ($\tau_{\mathrm{quench}}$) for the observed group population. At high stellar masses (${M}_{\star}=10^{10.5}~{\rm M}_{\odot}$) we find that $\tau_{\mathrm{quench}} = 2.4\substack{+0.2 \\ -0.2}$ Gyr, which is consistent with previous estimates at this epoch. At lower stellar masses (${M}_{\star}=10^{9.5}~{\rm M}_{\odot}$), we find that $\tau_{\mathrm{quench}}=3.1\substack{+0.5 \\ -0.4}$ Gyr, which is shorter than prior estimates from photometry-based investigations. These timescales are consistent with satellite quenching via starvation, provided the hot gas envelope of infalling satellites is not stripped away. We find that the evolution in the quenching timescale between $0 \lt z \lt 1$ aligns with the evolution in the dynamical time of the host halo and the total cold gas depletion time. This suggests that the doubling of the quenching timescale in groups since $z\sim1$ could be related to the dynamical evolution of groups or a decrease in quenching efficiency via starvation with decreasing redshift., Comment: 10 figures; 2 tables; Accepted for publication in the Astrophysical Journal

This technical report presents Yi-Lightning, our latest flagship large language model (LLM). It achieves exceptional performance, ranking 6th overall on Chatbot Arena, with particularly strong results (2nd to 4th place) in specialized categories including Chinese, Math, Coding, and Hard Prompts. Yi-Lightning leverages an enhanced Mixture-of-Experts (MoE) architecture, featuring advanced expert segmentation and routing mechanisms coupled with optimized KV-caching techniques. Our development process encompasses comprehensive pre-training, supervised fine-tuning (SFT), and reinforcement learning from human feedback (RLHF), where we devise deliberate strategies for multi-stage training, synthetic data construction, and reward modeling. Furthermore, we implement RAISE (Responsible AI Safety Engine), a four-component framework to address safety issues across pre-training, post-training, and serving phases. Empowered by our scalable super-computing infrastructure, all these innovations substantially reduce training, deployment and inference costs while maintaining high-performance standards. With further evaluations on public academic benchmarks, Yi-Lightning demonstrates competitive performance against top-tier LLMs, while we observe a notable disparity between traditional, static benchmark results and real-world, dynamic human preferences. This observation prompts a critical reassessment of conventional benchmarks' utility in guiding the development of more intelligent and powerful AI systems for practical applications. Yi-Lightning is now available through our developer platform at https://platform.lingyiwanwu.com.

The valid measurement of generative AI (GenAI) systems' capabilities, risks, and impacts forms the bedrock of our ability to evaluate these systems. We introduce a shared standard for valid measurement that helps place many of the disparate-seeming evaluation practices in use today on a common footing. Our framework, grounded in measurement theory from the social sciences, extends the work of Adcock & Collier (2001) in which the authors formalized valid measurement of concepts in political science via three processes: systematizing background concepts, operationalizing systematized concepts via annotation procedures, and applying those procedures to instances. We argue that valid measurement of GenAI systems' capabilities, risks, and impacts, further requires systematizing, operationalizing, and applying not only the entailed concepts, but also the contexts of interest and the metrics used. This involves both descriptive reasoning about particular instances and inferential reasoning about underlying populations, which is the purview of statistics. By placing many disparate-seeming GenAI evaluation practices on a common footing, our framework enables individual evaluations to be better understood, interrogated for reliability and validity, and meaningfully compared. This is an important step in advancing GenAI evaluation practices toward more formalized and theoretically grounded processes -- i.e., toward a science of GenAI evaluations., Comment: NeurIPS 2024 Workshop on Statistical Foundations of LLMs and Foundation Models (SFLLM)

Computational fluid dynamics (CFD) simulations are crucial in automotive, aerospace, maritime and medical applications, but are limited by the complexity, cost and computational requirements of directly calculating the flow, often taking days of compute time. Machine-learning architectures, such as controlled generative adversarial networks (cGANs) hold significant potential in enhancing or replacing CFD investigations, due to cGANs ability to approximate the underlying data distribution of a dataset. Unlike traditional cGAN applications, where the entire image carries information, CFD data contains small regions of highly variant data, immersed in a large context of low variance that is of minimal importance. This renders most existing deep learning techniques that give equal importance to every portion of the data during training, inefficient. To mitigate this, a novel loss function is proposed called Gradient Mean Squared Error (GMSE) which automatically and dynamically identifies the regions of importance on a field-by-field basis, assigning appropriate weights according to the local variance. To assess the effectiveness of the proposed solution, three identical networks were trained; optimised with Mean Squared Error (MSE) loss, proposed GMSE loss and a dynamic variant of GMSE (DGMSE). The novel loss function resulted in faster loss convergence, correlating to reduced training time, whilst also displaying an 83.6% reduction in structural similarity error between the generated field and ground truth simulations, a 76.6% higher maximum rate of loss and an increased ability to fool a discriminator network. It is hoped that this loss function will enable accelerated machine learning within computational fluid dynamics., Comment: 37 pages, 13 figures, preprint submitted to Engineering Applications of Artificial Intelligence (EAAI)

The next generation of cosmological spectroscopic sky surveys will probe the distribution of matter across several Gigaparsecs (Gpc) or many billion light-years. In order to leverage the rich data in these new maps to gain a better understanding of the physics that shapes the large-scale structure of the cosmos, observed matter distributions must be compared to simulated mock skies. Small mock skies can be produced using precise, physics-driven hydrodynamical simulations. However, the need to capture small, kpc-scale density fluctuations in the intergalactic medium (IGM) places tight restrictions on the necessary minimum resolution of these simulations. Even on the most powerful supercomputers, it is impossible to run simulations of such high resolution in volumes comparable to what will be probed by future surveys, due to the vast quantity of data needed to store such a simulation in computer memory. However, it is possible to represent the essential features of these high-resolution simulations using orders of magnitude less memory. We present a hybrid approach that employs a physics-driven hydrodynamical simulation at a much lower-than-necessary resolution, followed by a data-driven, deep-learning Enhancement. This hybrid approach allows us to produce hydrodynamic mock skies that accurately capture small, kpc-scale features in the IGM but which span hundreds of Megaparsecs. We have produced such a volume which is roughly one Gigaparsec in diameter and examine its relevant large-scale statistical features, emphasizing certain properties that could not be captured by previous smaller simulations. We present this hydrodynamic volume as well as a companion n-body dark matter simulation and halo catalog which we are making publically available to the community for use in calibrating data pipelines for upcoming survey analyses.

Recently, a class of long-period radio transients (LPTs) has been discovered, exhibiting emission on timescales thousands of times longer than radio pulsars. Several models had been proposed implicating either a strong magnetic field neutron star, isolated white dwarf pulsar, or a white dwarf binary system with a low-mass companion. While several models for LPTs also predict X-ray emission, no LPTs have been detected in X-rays despite extensive searches. Here we report the discovery of an extremely bright LPT (10-20 Jy in radio), ASKAP J1832-0911, which has coincident radio and X-ray emission, both with a 44.2-minute period. The X-ray and radio luminosities are correlated and vary by several orders of magnitude. These properties are unique amongst known Galactic objects and require a new explanation. We consider a $\gtrsim0.5$ Myr old magnetar with a $\gtrsim 10^{13}$ G crustal field, or an extremely magnetised white dwarf in a binary system with a dwarf companion, to be plausible explanations for ASKAP J1832-0911, although both explanations pose significant challenges to formation and emission theories. The X-ray detection also establishes a new class of hour-scale periodic X-ray transients of luminosity $\sim10^{33}$ erg/s associated with exceptionally bright coherent radio emission., Comment: 52 pages, 17 figures, 3 tables

We examine the quiescent fractions of massive galaxies in six $z\gtrsim3$ spectroscopically-confirmed protoclusters in the COSMOS field, one of which is newly confirmed and presented here. We report the spectroscopic confirmation of MAGAZ3NE~J100143+023021 at $z=3.122^{+0.007}_{-0.004}$ by the Massive Ancient Galaxies At $z>3$ NEar-infrared (MAGAZ3NE) survey. MAGAZ3NE~J100143+023021 contains a total of 79 protocluster members (28 spectroscopic and 51 photometric). Three spectroscopically-confirmed members are star-forming ultra-massive galaxies ($\log(M_{\star}/{\rm M}_\odot)>11$; UMGs), the most massive of which has $\log(M_{\star}/{\rm M}_\odot)=11.15^{+0.05}_{-0.06}$. Combining Keck/MOSFIRE spectroscopy and the COSMOS2020 photometric catalog, we use a weighted Gaussian kernel density estimator to map the protocluster and measure its total mass $2.25^{+1.55}_{-0.65}\times10^{14}~{\rm M}_{\odot}$ in the dense ``core'' region. For each of the six COSMOS protoclusters, we compare the quiescent fraction to the status of the central UMG as star-forming or quiescent. We observe that galaxies in these protoclusters appear to obey galactic conformity: elevated quiescent fractions are found in protoclusters with $UVJ$ quiescent UMGs and low quiescent fractions are found in protoclusters containing $UVJ$ star-forming UMGs. This correlation of star-formation/quiescence in UMGs and the massive galaxies nearby in these protoclusters is the first evidence for the existence of galactic conformity at $z>3$. Despite disagreements over mechanisms behind conformity at low redshifts, its presence at these early cosmic times would provide strong constraints on the physics proposed to drive galactic conformity., Comment: 6 figures, 5 tables. Accepted for publication in the Astrophysical Journal

Uranium mononitride (UN) is a promising accident tolerant fuel due to its high fissile density and high thermal conductivity. In this study, we developed the first machine learning interatomic potentials for reliable atomic-scale modeling of UN at finite temperatures. We constructed a training set using density functional theory (DFT) calculations that was enriched through an active learning procedure and two neural network potentials were generated. We found that both potentials can reproduce some thermophysical properties of interest such as temperature dependent heat capacity. We also evaluated the energy of stoichiometric defect reactions and defect migration barriers and found close agreement with DFT values demonstrating that our potentials can be used for a description of defects in UN.

Large language model unlearning aims to remove harmful information that LLMs have learnt to prevent their use for malicious purposes. LLMU and RMU have been proposed as two methods for LLM unlearning, achieving impressive results on unlearning benchmarks. We study in detail the efficacy of these methods by evaluating their impact on general model capabilities on the WMDP benchmark as well as a biology benchmark we create. Our experiments show that RMU generally leads to better preservation of model capabilities, for similar or better unlearning. We further test the robustness of these methods and find that doing 5-shot prompting or rephrasing the question in simple ways can lead to an over ten-fold increase in accuracy on unlearning benchmarks. Finally, we show that training on unrelated data can almost completely recover pre-unlearning performance, demonstrating that these methods fail at truly unlearning. The code is available at: https://github.com/JaiDoshi/Knowledge-Erasure., Comment: 9 pages, 2 figures

Across academia, industry, and government, there is an increasing awareness that the measurement tasks involved in evaluating generative AI (GenAI) systems are especially difficult. We argue that these measurement tasks are highly reminiscent of measurement tasks found throughout the social sciences. With this in mind, we present a framework, grounded in measurement theory from the social sciences, for measuring concepts related to the capabilities, impacts, opportunities, and risks of GenAI systems. The framework distinguishes between four levels: the background concept, the systematized concept, the measurement instrument(s), and the instance-level measurements themselves. This four-level approach differs from the way measurement is typically done in ML, where researchers and practitioners appear to jump straight from background concepts to measurement instruments, with little to no explicit systematization in between. As well as surfacing assumptions, thereby making it easier to understand exactly what the resulting measurements do and do not mean, this framework has two important implications for evaluating evaluations: First, it can enable stakeholders from different worlds to participate in conceptual debates, broadening the expertise involved in evaluating GenAI systems. Second, it brings rigor to operational debates by offering a set of lenses for interrogating the validity of measurement instruments and their resulting measurements., Comment: NeurIPS 2024 Workshop on Evaluating Evaluations (EvalEval)

We introduce a dataset of natural-language questions in the decision theory of so-called Newcomb-like problems. Newcomb-like problems include, for instance, decision problems in which an agent interacts with a similar other agent, and thus has to reason about the fact that the other agent will likely reason in similar ways. Evaluating LLM reasoning about Newcomb-like problems is important because interactions between foundation-model-based agents will often be Newcomb-like. Some ways of reasoning about Newcomb-like problems may allow for greater cooperation between models. Our dataset contains both capabilities questions (i.e., questions with a unique, uncontroversially correct answer) and attitude questions (i.e., questions about which decision theorists would disagree). We use our dataset for an investigation of decision-theoretical capabilities and expressed attitudes and their interplay in existing models (different models by OpenAI, Anthropic, Meta, GDM, Reka, etc.), as well as models under simple prompt-based interventions. We find, among other things, that attitudes vary significantly between existing models; that high capabilities are associated with attitudes more favorable toward so-called evidential decision theory; and that attitudes are consistent across different types of questions., Comment: 48 pages, 15 figures; code and data at https://github.com/casparoe/newcomblike_questions_dataset; corrected error in funding acknowledgments

Inflammation of the umbilical cord can be seen as a result of ascending intrauterine infection or other inflammatory stimuli. Acute fetal inflammatory response (FIR) is characterized by infiltration of the umbilical cord by fetal neutrophils, and can be associated with neonatal sepsis or fetal inflammatory response syndrome. Recent advances in deep learning in digital pathology have demonstrated favorable performance across a wide range of clinical tasks, such as diagnosis and prognosis. In this study we classified FIR from whole slide images (WSI). We digitized 4100 histological slides of umbilical cord stained with hematoxylin and eosin(H&E) and extracted placental diagnoses from the electronic health record. We build models using attention-based whole slide learning models. We compared strategies between features extracted by a model (ConvNeXtXLarge) pretrained on non-medical images (ImageNet), and one pretrained using histopathology images (UNI). We trained multiple iterations of each model and combined them into an ensemble. The predictions from the ensemble of models trained using UNI achieved an overall balanced accuracy of 0.836 on the test dataset. In comparison, the ensembled predictions using ConvNeXtXLarge had a lower balanced accuracy of 0.7209. Heatmaps generated from top accuracy model appropriately highlighted arteritis in cases of FIR 2. In FIR 1, the highest performing model assigned high attention to areas of activated-appearing stroma in Wharton's Jelly. However, other high-performing models assigned attention to umbilical vessels. We developed models for diagnosis of FIR from placental histology images, helping reduce interobserver variability among pathologists. Future work may examine the utility of these models for identifying infants at risk of systemic inflammatory response or early onset neonatal sepsis.

We combine the power of blind integral field spectroscopy from the Hobby-Eberly Telescope (HET) Dark Energy Experiment (HETDEX) with sources detected by the Low Frequency Array (LOFAR) to construct the HETDEX-LOFAR Spectroscopic Redshift Catalog. Starting from the first data release of the LOFAR Two-metre Sky Survey (LoTSS), including a value-added catalog with photometric redshifts, we extracted 28,705 HETDEX spectra. Using an automatic classifying algorithm, we assigned each object a star, galaxy, or quasar label along with a velocity/redshift, with supplemental classifications coming from the continuum and emission line catalogs of the internal, fourth data release from HETDEX (HDR4). We measured 9,087 new redshifts; in combination with the value-added catalog, our final spectroscopic redshift sample is 9,710 sources. This new catalog contains the highest substantial fraction of LOFAR galaxies with spectroscopic redshift information; it improves archival spectroscopic redshifts, and facilitates research to determine the [O II] emission properties of radio galaxies from $0.0 < z < 0.5$, and the Ly$\alpha$ emission characteristics of both radio galaxies and quasars from $1.9 < z < 3.5$. Additionally, by combining the unique properties of LOFAR and HETDEX, we are able to measure star formation rates (SFR) and stellar masses. Using the Visible Integral-field Replicable Unit Spectrograph (VIRUS), we measure the emission lines of [O III], [Ne III], and [O II] and evaluate line-ratio diagnostics to determine whether the emission from these galaxies is dominated by AGN or star formation and fit a new SFR-L$_{150MHz}$ relationship., Comment: 21 pages, 17 figures, submitted to ApJ

Artificial intelligence (AI) tools are now prevalent in many knowledge work industries. As AI becomes more capable and interactive, there is a growing need for guidance on how to employ AI most effectively. The A2C framework (Tariq, Chhetri, Nepal & Paris, 2024) distinguishes three decision-making modes for engaging AI: automation (AI completes a task, including decision/action), augmentation (AI supports human to decide) and collaboration (iterative interaction between human and AI). However, selecting the appropriate mode for a specific application is not always straightforward. The goal of the present study was to compile and trial a simple set of criteria to support recommendations about appropriate A2C mode for a given application. Drawing on human factors and computer science literature, we identified key criteria related to elements of the task, impacts on worker and support needs. From these criteria we built a scoring rubric with recommendation for A2C mode. As a preliminary test of this approach, we applied the criteria to cognitive task analysis (CTA) outputs from three tasks in the science domain - genome annotation, biological collections curation and protein crystallization - which provided insights into worker decision points, challenges and expert strategies. This paper describes the method for connecting CTA to A2C, reflecting on the challenges and future directions., Comment: 27 pages

In Newcomb's problem, causal decision theory (CDT) recommends two-boxing and thus comes apart from evidential decision theory (EDT) and ex ante policy optimisation (which prescribe one-boxing). However, in Newcomb's problem, you should perhaps believe that with some probability you are in a simulation run by the predictor to determine whether to put a million dollars into the opaque box. If so, then causal decision theory might recommend one-boxing in order to cause the predictor to fill the opaque box. In this paper, we study generalisations of this approach. That is, we consider general Newcomblike problems and try to form reasonable self-locating beliefs under which CDT's recommendations align with an EDT-like notion of ex ante policy optimisation. We consider approaches in which we model the world as running simulations of the agent, and an approach not based on such models (which we call 'Generalised Generalised Thirding', or GGT). For each approach, we characterise the resulting CDT policies, and prove that under certain conditions, these include the ex ante optimal policies.

We investigate cosmological vacuum amplification of gravitational waves in dynamical Chern-Simons gravity. We develop a comprehensive framework to compute graviton production induced by the parity violating Pontryagin coupling and study its imprint on the stochastic gravitational wave background energy power spectrum. We explore gravitational vacuum amplification in four concrete scenarios for the evolution of the Chern-Simons pseudoscalar. We show that a parity-violating contribution dominates over an initially flat spectrum when the velocity of the pseudoscalar quickly interpolates between two asymptotically constant values or when it is nonvanishing and constant through a finite period of time. This is also the case when we parametrize the pseudoscalar evolution by a perfect fluid with radiation- and dust-like equations of state for large enough values of its energy density. The resulting spectra are compared with the sensitivity curves of current and future gravitational wave observational searches., Comment: 17 pages, 4 figures