Search

SDO-FM is a foundation model using data from NASA's Solar Dynamics Observatory (SDO) spacecraft; integrating three separate instruments to encapsulate the Sun's complex physical interactions into a multi-modal embedding space. This model can be used to streamline scientific investigations involving SDO by making the enormous datasets more computationally accessible for heliophysics research and enable investigations that require instrument fusion. We discuss four key components: an ingestion pipeline to create machine learning ready datasets, the model architecture and training approach, resultant embeddings and fine-tunable models, and finally downstream fine-tuned applications. A key component of this effort has been to include subject matter specialists at each stage of development; reviewing the scientific value and providing guidance for model architecture, dataset, and training paradigm decisions. This paper marks release of our pretrained models and embedding datasets, available to the community on Hugging Face and sdofm.org.

The unexpected discovery of intrinsic ferromagnetism in layered van der Waals materials has sparked interest in both their fundamental properties and their potential for novel applications. Recent studies suggest near room-temperature ferromagnetism in the pressurized van der Waals crystal CrGeTe$_3$. We perform a comprehensive experimental and theoretical investigation of magnetism and electronic correlations in CrGeTe$_3$, combining broad-frequency reflectivity measurements with density functional theory and dynamical mean-field theory calculations. Our experimental optical conductivity spectra trace the signatures of developing ferromagnetic order and of the insulator-to-metal transition (IMT) as a function of temperature and hydrostatic pressure. With increasing pressure, we observe the emergence of a mid-infrared feature in the optical conductivity, indicating the development of strong orbital-selective correlations in the high-pressure ferromagnetic phase. We find a peculiar relationship between the plasma frequency and Curie temperature of CrGeTe$_3$, which strongly suggests that a double-exchange mechanism is responsible for the observed near room-temperature ferromagnetism. In contrast to previous studies, our results clearly demonstrate the existence of an optical gap in the metallic phase, ruling out the collapse of the charge transfer gap under pressure., Comment: 8+15 pages, 4+20 figures

Using $pp$ collision data at $\sqrt{s}=13$ TeV, recorded by the LHCb experiment between 2016 and 2018 and corresponding to an integrated luminosity of $5.4$ fb$^{-1}$, the forward-backward asymmetry in the $pp \to Z/\gamma^{*} \to \mu^+\mu^-$ process is measured. The measurement is carried out in ten intervals of the difference between the muon pseudorapidities, within a fiducial region covering dimuon masses between $66$ and $116$ GeV, muon pseudorapidities between $2.0$ and $4.5$ and muon transverse momenta above $20$ GeV. These forward-backward asymmetries are compared with predictions, at next-to-leading order in the strong and electroweak couplings. The measured effective leptonic weak mixing angle is $\sin^2\theta_{\rm eff}^\ell = 0.23147 \pm 0.00044 \pm 0.00005 \pm 0.00023$, where the first uncertainty is statistical, the second arises from systematic uncertainties associated with the asymmetry measurement, and the third arises from uncertainties in the fit model used to extract $\sin^2\theta_{\rm eff}^\ell$ from the asymmetry measurement. This result is based on an arithmetic average of results using the CT18, MSHT20, and NNPDF31 parameterisations of the proton internal structure, and is consistent with previous measurements and with predictions from the global electroweak fit., Comment: All figures and tables, along with machine-readable versions and any supplementary material and additional information, are available at https://lbfence.cern.ch/alcm/public/analysis/full-details/3360/ (LHCb public pages)

The rising interest in leveraging higher-order interactions present in complex systems has led to a surge in more expressive models exploiting high-order structures in the data, especially in topological deep learning (TDL), which designs neural networks on high-order domains such as simplicial complexes. However, progress in this field is hindered by the scarcity of datasets for benchmarking these architectures. To address this gap, we introduce MANTRA, the first large-scale, diverse, and intrinsically high order dataset for benchmarking high-order models, comprising over 43,000 and 249,000 triangulations of surfaces and three-dimensional manifolds, respectively. With MANTRA, we assess several graph- and simplicial complex-based models on three topological classification tasks. We demonstrate that while simplicial complex-based neural networks generally outperform their graph-based counterparts in capturing simple topological invariants, they also struggle, suggesting a rethink of TDL. Thus, MANTRA serves as a benchmark for assessing and advancing topological methods, leading the way for more effective high-order models., Comment: 26 pages, 2 figures, 22 tables

The kinematic information of the extraplanar diffuse ionized gas (eDIG) around galaxies provides clues to the origin of the gas. The eDIG-CHANGES project studies the physical and kinematic properties of the eDIG around the CHANG-ES sample of nearby edge-on disk galaxies. We use a novel multi-slit narrow-band spectroscopy technique to obtain the spatial distribution of spectral properties of the ionized gas around NGC 891, which is often regarded as an analog of the Milky Way. We developed specific data reduction procedures for the multi-slit narrow-band spectroscopy data taken with the MDM 2.4m telescope. The data presented in this paper covers the H$\alpha$ and [N II]$\lambda\lambda$6548,6583\AA emission lines. The eDIG traced by the H$\alpha$ and [N II] lines shows an obvious asymmetric morphology, being brighter in the northeastern part of the galactic disk and extending a few kpc above and below the disk. Global variations in the [N II]/H$\alpha$ line ratio suggest additional heating mechanisms for the eDIG at large heights beyond photoionization. We also construct position-velocity (PV) diagrams of the eDIG based on our optical multi-slit spectroscopy data and compare them to similar PV diagrams constructed with the H I data. The dynamics of the two gas phases are generally consistent with each other. Modeling the rotation curves at different heights from the galactic mid-plane suggests a vertical negative gradient in turnover radius and maximum rotation velocity, with magnitudes of approximately $3$ kpc kpc$^{-1}$ and $22-25$ km s$^{-1}$ kpc$^{-1}$, respectively. Measured vertical gradients of the rotation curve parameters suggest significant differential rotation of the ionized gas in the halo, or often referred to as the lagging eDIG. Systematic study of the lagging eDIG in our eDIG-CHANGES project, will help us to better understand the dynamics of the ionized gas in the halo., Comment: 11 pages, 8 figures, accepted for publication in A&A

We present a purely deep neural network-based approach for estimating long memory parameters of time series models that incorporate the phenomenon of long-range dependence. Parameters, such as the Hurst exponent, are critical in characterizing the long-range dependence, roughness, and self-similarity of stochastic processes. The accurate and fast estimation of these parameters holds significant importance across various scientific disciplines, including finance, physics, and engineering. We harnessed efficient process generators to provide high-quality synthetic training data, enabling the training of scale-invariant 1D Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) models. Our neural models outperform conventional statistical methods, even those augmented with neural networks. The precision, speed, consistency, and robustness of our estimators are demonstrated through experiments involving fractional Brownian motion (fBm), the Autoregressive Fractionally Integrated Moving Average (ARFIMA) process, and the fractional Ornstein-Uhlenbeck (fOU) process. We believe that our work will inspire further research in the field of stochastic process modeling and parameter estimation using deep learning techniques., Comment: 14 pages, 16 figures, https://github.com/aielte-research/LMSParEst

We address the challenge of efficient auto-regressive generation in sequence prediction models by introducing FutureFill: a method for fast generation that applies to any sequence prediction algorithm based on convolutional operators. Our approach reduces the generation time requirement from linear to square root relative to the context length. Additionally, FutureFill requires a prefill cache sized only by the number of tokens generated, which is smaller than the cache requirements for standard convolutional and attention-based models. We validate our theoretical findings with experimental evidence demonstrating correctness and efficiency gains in a synthetic generation task.

Neurosymbolic (NeSy) artificial intelligence describes the combination of logic or rule-based techniques with neural networks. Compared to neural approaches, NeSy methods often possess enhanced interpretability, which is particularly promising for biomedical applications like drug discovery. However, since interpretability is broadly defined, there are no clear guidelines for assessing the biological plausibility of model interpretations. To assess interpretability in the context of drug discovery, we devise a novel prediction task, called drug mechanism-of-action (MoA) deconvolution, with an associated, tailored knowledge graph (KG), MoA-net. We then develop the MoA Retrieval System (MARS), a NeSy approach for drug discovery which leverages logical rules with learned rule weights. Using this interpretable feature alongside domain knowledge, we find that MARS and other NeSy approaches on KGs are susceptible to reasoning shortcuts, in which the prediction of true labels is driven by "degree-bias" rather than the domain-based rules. Subsequently, we demonstrate ways to identify and mitigate this. Thereafter, MARS achieves performance on par with current state-of-the-art models while producing model interpretations aligned with known MoAs., Comment: Under review. 10 pages, 5 supplementary pages. Corresponding code is here: https://github.com/laurendelong21/MARS and here: https://github.com/laurendelong21/MoA-Net

We address the problem of learning-augmented online caching in the scenario when each request is accompanied by a prediction of the next occurrence of the requested page. We improve currently known bounds on the competitive ratio of the BlindOracle algorithm, which evicts a page predicted to be requested last. We also prove a lower bound on the competitive ratio of any randomized algorithm and show that a combination of the BlindOracle with the Marker algorithm achieves a competitive ratio that is optimal up to some constant.

We study the depth of grade-school math (GSM) problem-solving capabilities of LLMs. To this end, we evaluate their performance on pairs of existing math word problems together so that the answer to the second problem depends on correctly answering the first problem. Our findings reveal a significant reasoning gap in most LLMs, that is performance difference between solving the compositional pairs and solving each question independently. This gap is more pronounced in smaller, more cost-efficient, and math-specialized models. Moreover, instruction-tuning recipes and code generation have varying effects across LLM sizes, while finetuning on GSM can lead to task overfitting. Our analysis indicates that large reasoning gaps are not because of test-set leakage, but due to distraction from additional context and poor second-hop reasoning. Overall, LLMs exhibit systematic differences in their reasoning abilities, despite what their performance on standard benchmarks indicates.

The practical use of text-to-image generation has evolved from simple, monolithic models to complex workflows that combine multiple specialized components. While workflow-based approaches can lead to improved image quality, crafting effective workflows requires significant expertise, owing to the large number of available components, their complex inter-dependence, and their dependence on the generation prompt. Here, we introduce the novel task of prompt-adaptive workflow generation, where the goal is to automatically tailor a workflow to each user prompt. We propose two LLM-based approaches to tackle this task: a tuning-based method that learns from user-preference data, and a training-free method that uses the LLM to select existing flows. Both approaches lead to improved image quality when compared to monolithic models or generic, prompt-independent workflows. Our work shows that prompt-dependent flow prediction offers a new pathway to improving text-to-image generation quality, complementing existing research directions in the field., Comment: Project website: https://comfygen-paper.github.io/

As the field of multi-agent reinforcement learning (MARL) progresses towards larger and more complex environments, achieving strong performance while maintaining memory efficiency and scalability to many agents becomes increasingly important. Although recent research has led to several advanced algorithms, to date, none fully address all of these key properties simultaneously. In this work, we introduce Sable, a novel and theoretically sound algorithm that adapts the retention mechanism from Retentive Networks to MARL. Sable's retention-based sequence modelling architecture allows for computationally efficient scaling to a large number of agents, as well as maintaining a long temporal context, making it well-suited for large-scale partially observable environments. Through extensive evaluations across six diverse environments, we demonstrate how Sable is able to significantly outperform existing state-of-the-art methods in the majority of tasks (34 out of 45, roughly 75\%). Furthermore, Sable demonstrates stable performance as we scale the number of agents, handling environments with more than a thousand agents while exhibiting a linear increase in memory usage. Finally, we conduct ablation studies to isolate the source of Sable's performance gains and confirm its efficient computational memory usage. Our results highlight Sable's performance and efficiency, positioning it as a leading approach to MARL at scale.

Cardiac magnetic resonance imaging (CMR), considered the gold standard for noninvasive cardiac assessment, is a diverse and complex modality requiring a wide variety of image processing tasks for comprehensive assessment of cardiac morphology and function. Advances in deep learning have enabled the development of state-of-the-art (SoTA) models for these tasks. However, model training is challenging due to data and label scarcity, especially in the less common imaging sequences. Moreover, each model is often trained for a specific task, with no connection between related tasks. In this work, we introduce a vision foundation model trained for CMR assessment, that is trained in a self-supervised fashion on 36 million CMR images. We then finetune the model in supervised way for 9 clinical tasks typical to a CMR workflow, across classification, segmentation, landmark localization, and pathology detection. We demonstrate improved accuracy and robustness across all tasks, over a range of available labeled dataset sizes. We also demonstrate improved few-shot learning with fewer labeled samples, a common challenge in medical image analyses. We achieve an out-of-box performance comparable to SoTA for most clinical tasks. The proposed method thus presents a resource-efficient, unified framework for CMR assessment, with the potential to accelerate the development of deep learning-based solutions for image analysis tasks, even with few annotated data available., Comment: 11 pages, 3 figures, 4 tables

The discovery of chiral helical magnetism (CHM) in Cr$_{1/3}$NbS$_2$ and the stabilization of a chiral soliton lattice (CSL) has attracted considerable interest in view of their potential technological applications. However, there is an ongoing debate regarding whether the sister compound, Mn$_{1/3}$NbS$_2$, which shares the same crystal structure, exhibits similar nontrivial properties which rely on the stabilization of the lack of inversion symmetry at the magnetic ion. In this study, we conduct a comprehensive investigation of the magnetically ordered states of both compounds, using $^{53}$Cr, $^{55}$Mn and $^{93}$Nb nuclear magnetic resonance. Our results, supported by density functional calculations, detect in a high-quality single crystal of Cr$_{1/3}$NbS$_2$ all the signatures of the monoaxial CHM in a magnetic field, identifying it as a textbook NMR case. The detailed understanding of this prototypic behavior provides a reference for Mn$_{1/3}$NbS$_2$. Despite the much larger density of specific defects in this second single crystal, we confirm the presence of a CHM phase in the Mn compound, characterized by a very large critical field for the forced ferromagnetic phase ($\approx 5$ T for H$\parallel\hat c$)., Comment: 10 pages, 7 figures, Supplemental MAterial

Kepler-51 is a $\lesssim 1\,\mathrm{Gyr}$-old Sun-like star hosting three transiting planets with radii $\approx 6$-$9\,R_\oplus$ and orbital periods $\approx 45$-$130\,\mathrm{days}$. Transit timing variations (TTVs) measured with past Kepler and Hubble Space Telescope (HST) observations have been successfully modeled by considering gravitational interactions between the three transiting planets, yielding low masses and low mean densities ($\lesssim 0.1\,\mathrm{g/cm^3}$) for all three planets. However, the transit time of the outermost transiting planet Kepler-51d recently measured by the James Webb Space Telescope (JWST) 10 years after the Kepler observations is significantly discrepant from the prediction made by the three-planet TTV model, which we confirmed with ground-based and follow-up HST observations. We show that the departure from the three-planet model is explained by including a fourth outer planet, Kepler-51e, in the TTV model. A wide range of masses ($\lesssim M_\mathrm{Jup}$) and orbital periods ($\lesssim 10\,\mathrm{yr}$) are possible for Kepler-51e. Nevertheless, all the coplanar solutions found from our brute-force search imply masses $\lesssim 10\,M_\oplus$ for the inner transiting planets. Thus their densities remain low, though with larger uncertainties than previously estimated. Unlike other possible solutions, the one in which Kepler-51e is around the $2:1$ mean motion resonance with Kepler-51d implies low orbital eccentricities ($\lesssim 0.05$) and comparable masses ($\sim 5\,M_\oplus$) for all four planets, as is seen in other compact multi-planet systems. This work demonstrates the importance of long-term follow-up of TTV systems for probing longer period planets in a system., Comment: 48 pages, 26 figures, accepted for publication in AJ

Real-time systems are intrinsic components of many pivotal applications, such as self-driving vehicles, aerospace and defense systems. The trend in these applications is to incorporate multiple tasks onto fewer, more powerful hardware platforms, e.g., multi-core systems, mainly for reducing cost and power consumption. Many real-time tasks, like control tasks, can tolerate occasional deadline misses due to robust algorithms. These tasks can be modeled using the weakly-hard model. Literature shows that leveraging the weakly-hard model can relax the over-provisioning associated with designed real-time systems. However, a wide-range of the research focuses on single-core platforms. Therefore, we strive to extend the state-of-the-art of scheduling weakly-hard real-time tasks to multi-core platforms. We present a global job-level fixed priority scheduling algorithm together with its schedulability analysis. The scheduling algorithm leverages the tolerable continuous deadline misses to assigning priorities to jobs. The proposed analysis extends the Response Time Analysis (RTA) for global scheduling to test the schedulability of tasks. Hence, our analysis scales with the number of tasks and number of cores because, unlike literature, it depends neither on Integer Linear Programming nor reachability trees. Schedulability analyses show that the schedulability ratio is improved by 40% comparing to the global Rate Monotonic (RM) scheduling and up to 60% more than the global EDF scheduling, which are the state-of-the-art schedulers on the RTEMS real-time operating system. Our evaluation on industrial embedded multi-core platform running RTEMS shows that the scheduling overhead of our proposal does not exceed 60 Nanosecond.

Current language models use subword-based tokenization algorithms like Byte Pair Encoding, which put their validity as models of linguistic representations into question. In this paper, we explore the potential of tokenization-free, phoneme- and grapheme-based language models. We demonstrate that small models based on the Llama architecture can achieve strong linguistic performance on standard syntactic and novel lexical/phonetic benchmarks when trained with character-level vocabularies. We further show that phoneme-based models without any graphemic biases almost match grapheme-based models in standard tasks and novel evaluations. Our findings suggest a promising direction for creating more linguistically plausible language models that are better suited for computational studies of language acquisition and processing.

Many recent methods aim to merge neural networks (NNs) with identical architectures trained on different tasks to obtain a single multi-task model. Most existing works tackle the simpler setup of merging NNs initialized from a common pre-trained network, where simple heuristics like weight averaging work well. This work targets a more challenging goal: merging large transformers trained on different tasks from distinct initializations. First, we demonstrate that traditional merging methods fail catastrophically in this setup. To overcome this challenge, we propose Foldable SuperNet Merge (FS-Merge), a method that optimizes a SuperNet to fuse the original models using a feature reconstruction loss. FS-Merge is simple, data-efficient, and capable of merging models of varying widths. We test FS-Merge against existing methods, including knowledge distillation, on MLPs and transformers across various settings, sizes, tasks, and modalities. FS-Merge consistently outperforms them, achieving SOTA results, particularly in limited data scenarios.

This work presents Fair4Free, a novel generative model to generate synthetic fair data using data-free distillation in the latent space. Fair4Free can work on the situation when the data is private or inaccessible. In our approach, we first train a teacher model to create fair representation and then distil the knowledge to a student model (using a smaller architecture). The process of distilling the student model is data-free, i.e. the student model does not have access to the training dataset while distilling. After the distillation, we use the distilled model to generate fair synthetic samples. Our extensive experiments show that our synthetic samples outperform state-of-the-art models in all three criteria (fairness, utility and synthetic quality) with a performance increase of 5% for fairness, 8% for utility and 12% in synthetic quality for both tabular and image datasets.

While "complexity science" has achieved significant successes in several interdisciplinary fields such as economics and biology, it is only a very recent observation that legal systems -- from the way legal texts are drafted and connected to the rest of the corpus, up to the level of how judges and courts reach decisions under a variety of conflicting inputs -- share several features with standard Complex Adaptive Systems. This review is meant as a gentle introduction to the use of quantitative tools and techniques of complexity science to describe, analyse, and tame the complex web of human interactions that the Law is supposed to regulate. We offer an overview of the main directions of research undertaken so far as well as an outlook for future research, and we argue that statistical physicists and complexity scientists should not ignore the opportunities offered by the cross-fertilisation between legal scholarship and complex-systems modelling., Comment: 7 pages, 4 figures, 2 tables - submitted as EPL Perspective

Flexoelectricity is universal in all dielectrics, effective at high temperatures, and a promising transduction technique for nanoelectromechanical systems (NEMS). However, as flexoelectricity is still in its early stages, many aspects require further investigation. Understanding how flexoelectricity depends on material parameters like crystallographic phase and how temperature might affect it, is important for selecting and optimizing the right material for technological applications. This work studies the influence of high-temperature annealing (and the consequent crystallization) in the flexoelectricity of hafnium oxide (HfO2), a material with significant technological relevance. We measure the flexoelectric coefficient for amorphous (not annealed) and annealed (slightly crystalline) phases of HfO2, with samples annealed in nitrogen or oxygen atmospheres. Our results indicate that the amorphous phase of HfO2 exhibits the highest flexoelectric coefficient (105 $\pm$ 10 pC/m), while annealed samples show a significant decrease, with the lowest value in nitrogen-annealed samples (26 $\pm$ 4 pC/m). Samples annealed in an oxygen atmosphere improve flexoelectric properties (54 $\pm$ 6 pC/m) compared to those annealed in nitrogen. Using cross-sectional imaging, X-ray diffraction, resonance frequency characterization, and relative permittivity measurements, we find that annealing promotes crystallization into the tetragonal phase and increases internal stress within the HfO2 layer, while most other parameters remain constant. We attribute the differences in flexoelectricity from the annealed samples to the quantity of oxygen vacancies in hafnium oxide. These oxygen vacancies in hafnium oxide seem to negatively affect the flexoelectric coefficient. This finding can be applied to optimize materials to enhance their flexoelectric properties.

Machine-learning based generation of process models from natural language text process descriptions provides a solution for the time-intensive and expensive process discovery phase. Many organizations have to carry out this phase, before they can utilize business process management and its benefits. Yet, research towards this is severely restrained by an apparent lack of large and high-quality datasets. This lack of data can be attributed to, among other things, an absence of proper tool assistance for dataset creation, resulting in high workloads and inferior data quality. We explore two assistance features to support dataset creation, a recommendation system for identifying process information in the text and visualization of the current state of already identified process information as a graphical business process model. A controlled user study with 31 participants shows that assisting dataset creators with recommendations lowers all aspects of workload, up to $-51.0\%$, and significantly improves annotation quality, up to $+38.9\%$. We make all data and code available to encourage further research on additional novel assistance strategies.

Recent advances in natural language processing and deep learning have accelerated the development of digital assistants. In conversational commerce, these assistants help customers find suitable products in online shops through natural language conversations. During the dialogue, the assistant identifies the customer's needs and preferences and subsequently suggests potentially relevant products. Traditional online shops often allow users to filter search results based on their preferences using facets. Selected facets can also serve as a reminder of how the product base was filtered. In conversational commerce, however, the absence of facets and the use of advanced natural language processing techniques can leave customers uncertain about how their input was processed by the system. This can hinder transparency and trust, which are critical factors influencing customers' purchase intentions. To address this issue, we propose a novel text-based digital assistant that, in the product assessment step, explains how specific product aspects relate to the user's previous utterances to enhance transparency and facilitate informed decision-making. We conducted a user study (N=135) and found a significant increase in user-perceived transparency when natural language explanations and highlighted text passages were provided, demonstrating their potential to extend system transparency to the product assessment step in conversational commerce.

Reward models are critical for aligning models to follow instructions, and are typically trained following one of two popular paradigms: Bradley-Terry style or Regression style. However, there is a lack of evidence that either approach is better than the other, when adequately matched for data. This is primarily because these approaches require data collected in different (but incompatible) formats, meaning that adequately matched data is not available in existing public datasets. To tackle this problem, we release preference annotations (designed for Bradley-Terry training) to complement existing ratings (designed for Regression style training) in the HelpSteer2 dataset. To improve data interpretability, preference annotations are accompanied with human-written justifications. Using this data, we conduct the first head-to-head comparison of Bradley-Terry and Regression models when adequately matched for data. Based on insights derived from such a comparison, we propose a novel approach to combine Bradley-Terry and Regression reward modeling. A Llama-3.1-70B-Instruct model tuned with this approach scores 94.1 on RewardBench, emerging top of more than 140 reward models as of 1 Oct 2024. We also demonstrate the effectiveness of this reward model at aligning models to follow instructions in RLHF. We open-source this dataset (CC-BY-4.0 license) at https://huggingface.co/datasets/nvidia/HelpSteer2 and openly release the trained Reward Model at https://huggingface.co/nvidia/Llama-3.1-Nemotron-70B-Reward, Comment: 26 pages, 3 figures

We present the results from spectroscopic and photometric analysis of 17 globular cluster (GC) candidates in the Irr II galaxy NGC 3077. The GC candidates were selected on the Hubble Space Telescope (HST) images and were cleaned of foreground Galactic stars using the GAIA parameters. We carried out aperture photometry using the multi-band archival images from SDSS, and 2MASS of all candidates, and low resolution (R= 1000) spectroscopic observations of 12 GC candidates and three suspected foreground stars using the OSIRIS/MOS mode at the Gran Telescopio Canarias (GTC). Age, metallicity and extinction values were determined both using spectroscopic and photometric data, independently. We find three of the 17 candidates are old (age >10 Gyr), metal-poor ([Fe/H]<-1.0 dex), massive (mass>10^5 Msun) GCs with characteristics similar to the classical GCs in the Milky Way. The rest are intermediate-age clusters (IACs) with typical ages of 3 to 4 Gyr, and in general metal-rich clusters. The radial velocities of both populations are within 100 km/s of the recessional velocity of the host galaxy. A relatively large population of IACs and low value of GC specific frequency (Sn=0.7) suggest that the pre-interaction galaxy was actively forming stars and star clusters, and is unlikely to be a dE as suggested in some previous works., Comment: Accepted in Astronomical Journal

This paper provides evidence that weak crossover effects differ in nature between matrix and relative clauses. Fukushima et al. (2024) provided similar evidence, showing that, when various non-structural factors were eliminated English speakers never accepted matrix weak crossover cases, but often accepted relative weak crossover ones. Those results were limited, however, by English word order, which lead to uncertainty as to whether this difference was due to the effects of linear precedence or syntactic structure. In this paper, to distinguish between these two possibilities, we conduct an experiment using Japanese, which lacks the word-order confound that English had. We find results that are qualitatively in line with Fukushima et al. (2024) suggesting that the relevant distinction is structural and not based simply on precedence., Comment: 18 pages, 17 figures, To appear in Proceedings of The Society of Modern Grammar (SMOG)'s International Conference on Syntax and Semantics (ICSS) 2024

Convolutional neural networks (CNNs) have shown great effectiveness in medical image segmentation. However, they may be limited in modeling large inter-subject variations in organ shapes and sizes and exploiting global long-range contextual information. This is because CNNs typically employ convolutions with fixed-sized local receptive fields and lack the mechanisms to utilize global information. To address these limitations, we developed Dynamic Multi-Resolution Convolution (DMRC) and Dynamic Multi-Scale Convolution (DMSC) modules. Both modules enhance the representation capabilities of single convolutions to capture varying scaled features and global contextual information. This is achieved in the DMRC module by employing a convolutional filter on images with different resolutions and subsequently utilizing dynamic mechanisms to model global inter-dependencies between features. In contrast, the DMSC module extracts features at different scales by employing convolutions with different kernel sizes and utilizing dynamic mechanisms to extract global contextual information. The utilization of convolutions with different kernel sizes in the DMSC module may increase computational complexity. To lessen this burden, we propose to use a lightweight design for convolution layers with a large kernel size. Thus, DMSC and DMRC modules are designed as lightweight drop-in replacements for single convolutions, and they can be easily integrated into general CNN architectures for end-to-end training. The segmentation network was proposed by incorporating our DMSC and DMRC modules into a standard U-Net architecture, termed Dynamic Multi-scale and Multi-resolution Convolution network (DMC-Net). The results demonstrate that our proposed DMSC and DMRC can enhance the representation capabilities of single convolutions and improve segmentation accuracy., Comment: 14 pages, 4 figures

Discovering mechanisms underlying the behaviors of complex, high dimensional, and nonlinear dynamical systems is a central goal of the natural and synthetic sciences. Breakthroughs in machine learning in concert with increasing capacities for computation and data collection have enabled the use of trajectory measurements for learning predictive models. However, rigorous approaches for interpreting mechanisms from such models remain elusive, and asymptotic prediction accuracy suffers if the model does not capture important state space structures (e.g., attracting invariant sets). These limitations are especially pressing for system-level behaviors such as whole-body locomotion, where discontinuous, transient, and multiscale phenomena are common and prior models are rare. To take the next step towards a theory and practice for dynamical inference of complex multiscale systems in biology and beyond, we introduce VERT, a framework for learning the attracting sets that characterize global system behavior without recourse to learning a global model. Our approach is based on an infinitesimal-local-global (ILG) framework for estimating the proximity of any sampled state to the attracting set, if one exists, with formal accuracy guarantees. We demonstrate our approach on phenomenological and physical oscillators with hierarchical and impulsive dynamics, finding sensitivity to both global and intermediate attractors composed in sequence and parallel. Application of VERT to human running kinematics data reveals insight into control modules that stabilize task-level dynamics, supporting a longstanding neuromechanical control hypothesis. The VERT framework thus enables rigorous inference of underlying dynamical structure even for systems where learning a global dynamics model is impractical or impossible.

Significant advancements have been made in understanding the physics of transition-edge sensors (TESs) over the past decade. However, key questions remain, particularly a detailed understanding of the current-dependent resistance of these detectors when biased within their superconducting transition. We use scanning superconducting quantum interference device (SQUID) microscopy (SSM) to image the local diamagnetic response of aluminum-manganese alloy (Al-Mn) transition-edge sensors (TESs) near their critical temperature of approximately 175 mK. By doing so, we gain insights into how the device dimensions influence TES transition width, which in turn affects device operation and informs optimal device design. Our images reveal that the Al-Mn thin film near the niobium (Nb) leads exhibits an excess diamagnetic response at temperatures several milli-Kelvin (mK) higher than the bulk of the film farther from the contacts. A possible origin of this behavior is a longitudinal proximity effect between the Nb and Al-Mn where the TES acts as a weak link between superconducting leads. We discuss how this effect shapes the temperature dependence of the resistance as the spacing between the leads decreases. This work demonstrates that magnetic imaging with SSM is a powerful tool for local characterization of superconducting detectors., Comment: 5 pages, 4 figures, Applied Superconductivity Conference (ASC 2024) proceedings, submitted to IEEE Transactions on Applied Superconductivity (IEEE TAS)

IRAC data have long implied that early ($z\gtrsim6$) galaxies often have very high specific star formation rates (sSFR$\gtrsim$30 Gyr$^{-1}$), but JWST data have shown that at least some early galaxies are forming stars far less vigorously. Here, we systematically analyze the recent star formation histories (SFHs) of a large ($N=368$) sample of $z\sim6$ Lyman-break galaxies (LBGs) spanning $-22\lesssim M_\mathrm{UV}\lesssim-16$ assembled from ACS+NIRCam imaging in the GOODS and Abell 2744 fields. We find that very low H$\alpha$-to-UV luminosity ratios ($L_\mathrm{H\alpha}/L_\mathrm{UV}$) and strong recent downturns in star formation rate (SFR) are $\approx$5$\times$ more common among the UV-faintest subset of our sample ($\langle M_\mathrm{UV}\rangle=-17.4$) compared to the brightest subset ($\langle M_\mathrm{UV}\rangle=-20.0$). The frequency of high $L_\mathrm{H\alpha}/L_\mathrm{UV}$ and strong recent SFR upturns is approximately constant with UV luminosity. We discuss how bursty SFHs naturally reproduce this much greater diversity in recent SFHs among very UV-faint galaxies. Using public NIRSpec/prism data, we newly confirm recent strong SFR downturns among three LBGs in our sample, and validate our photometric inferences on key SFH signatures among $z\sim6$ LBGs in general. Our results imply that early galaxies frequently cycle through phases of rapid stellar mass assembly and other periods of much slower growth. This yields huge ($\gtrsim$1-2 mag) fluctuations in $M_\mathrm{UV}$ on rapid ($\sim$10-30 Myr) timescales, helping explain the surprising abundance of $z>10$ galaxies. Finally, we caution that this burstiness causes all existing high-redshift samples (particularly line-selected samples) to be far less complete to galaxies with long recent phases of low sSFR than those currently undergoing a burst., Comment: Submitted to ApJ. 29 pages, 13 figures, 1 appendix

The scarcity of subspecialist medical expertise, particularly in rare, complex and life-threatening diseases, poses a significant challenge for healthcare delivery. This issue is particularly acute in cardiology where timely, accurate management determines outcomes. We explored the potential of AMIE (Articulate Medical Intelligence Explorer), a large language model (LLM)-based experimental AI system optimized for diagnostic dialogue, to potentially augment and support clinical decision-making in this challenging context. We curated a real-world dataset of 204 complex cases from a subspecialist cardiology practice, including results for electrocardiograms, echocardiograms, cardiac MRI, genetic tests, and cardiopulmonary stress tests. We developed a ten-domain evaluation rubric used by subspecialists to evaluate the quality of diagnosis and clinical management plans produced by general cardiologists or AMIE, the latter enhanced with web-search and self-critique capabilities. AMIE was rated superior to general cardiologists for 5 of the 10 domains (with preference ranging from 9% to 20%), and equivalent for the rest. Access to AMIE's response improved cardiologists' overall response quality in 63.7% of cases while lowering quality in just 3.4%. Cardiologists' responses with access to AMIE were superior to cardiologist responses without access to AMIE for all 10 domains. Qualitative examinations suggest AMIE and general cardiologist could complement each other, with AMIE thorough and sensitive, while general cardiologist concise and specific. Overall, our results suggest that specialized medical LLMs have the potential to augment general cardiologists' capabilities by bridging gaps in subspecialty expertise, though further research and validation are essential for wide clinical utility.

We present a novel deep learning pipeline to perform a model-independent, likelihood-free search for anomalous (i.e., non-background) events in the proposed next generation multi-ton scale liquid Xenon-based direct detection experiment, DARWIN. We train an anomaly detector comprising a variational autoencoder and a classifier on extensive, high-dimensional simulated detector response data and construct a one-dimensional anomaly score optimised to reject the background only hypothesis in the presence of an excess of non-background-like events. We benchmark the procedure with a sensitivity study that determines its power to reject the background-only hypothesis in the presence of an injected WIMP dark matter signal, outperforming the classical, likelihood-based background rejection test. We show that our neural networks learn relevant energy features of the events from low-level, high-dimensional detector outputs, without the need to compress this data into lower-dimensional observables, thus reducing computational effort and information loss. For the future, our approach lays the foundation for an efficient end-to-end pipeline that eliminates the need for many of the corrections and cuts that are traditionally part of the analysis chain, with the potential of achieving higher accuracy and significant reduction of analysis time., Comment: 10 Figures, 3 Tables, 23 Pages (incl. references)

Humans excel at abstracting data and constructing \emph{reusable} concepts, a capability lacking in current continual learning systems. The field of object-centric learning addresses this by developing abstract representations, or slots, from data without human supervision. Different methods have been proposed to tackle this task for images, whereas most are overly complex, non-differentiable, or poorly scalable. In this paper, we introduce a conceptually simple, fully-differentiable, non-iterative, and scalable method called SAMP Simplified Slot Attention with Max Pool Priors). It is implementable using only Convolution and MaxPool layers and an Attention layer. Our method encodes the input image with a Convolutional Neural Network and then uses a branch of alternating Convolution and MaxPool layers to create specialized sub-networks and extract primitive slots. These primitive slots are then used as queries for a Simplified Slot Attention over the encoded image. Despite its simplicity, our method is competitive or outperforms previous methods on standard benchmarks.

In this paper we study the (co)homology of Tanabe algebras, which are a family of subalgebras of the partition algebras exhibiting a Schur-Weyl duality with certain complex reflection groups. The homology of the partition algebras has been shown to be related to the homology of the symmetric groups by Boyd-Hepworth-Patzt and the results they obtain depend on a parameter. In all known results, the homology of a diagram algebra is dependent on one of two things: the invertibility of a parameter in the ground ring or the parity of the positive integer indexing the number of pairs of vertices. We show that the (co)homology of Tanabe algebras is isomorphic to the (co)homology of the symmetric groups and that this is independent of both the parameter and the parity of the index. To the best of our knowledge, this is the first example of a result of this sort. Along the way we will also study the (co)homology of the uniform block permutation algebra and the totally propagating partition algebra as well collecting cohomological analogues of known results for the homology of partition algebras and Jones annular algebras., Comment: 14 pages

UL 4600, the safety standard for autonomous products, mandates the use of Safety Performance Indicators (SPIs) to continuously ensure the validity of safety cases by monitoring and taking action when violations are identified. Despite numerous examples of concrete SPIs available in the standard and companion literature, their contribution rationale for achieving safety is often left implicit. In this paper, we present our initial work towards an argument pattern for the use of SPIs to ensure validity of safety cases throughout the entire lifecycle of the system. Our aim is to make the implicit argument behind using SPIs explicit, and based on this, to analyze the situations that can undermine confidence in the chosen set of SPIs. To maintain the confidence in SPIs' effectiveness, we propose an approach to continuously monitor their expected performance by using meta-SPIs.

Multi-fidelity Bayesian Optimization (MFBO) is a promising framework to speed up materials and molecular discovery as sources of information of different accuracies are at hand at increasing cost. Despite its potential use in chemical tasks, there is a lack of systematic evaluation of the many parameters playing a role in MFBO. In this work, we provide guidelines and recommendations to decide when to use MFBO in experimental settings. We investigate MFBO methods applied to molecules and materials problems. First, we test two different families of acquisition functions in two synthetic problems and study the effect of the informativeness and cost of the approximate function. We use our implementation and guidelines to benchmark three real discovery problems and compare them against their single-fidelity counterparts. Our results may help guide future efforts to implement MFBO as a routine tool in the chemical sciences.

Active learning (AL) is a widely used technique for optimizing data labeling in machine learning by iteratively selecting, labeling, and training on the most informative data. However, its integration with formal privacy-preserving methods, particularly differential privacy (DP), remains largely underexplored. While some works have explored differentially private AL for specialized scenarios like online learning, the fundamental challenge of combining AL with DP in standard learning settings has remained unaddressed, severely limiting AL's applicability in privacy-sensitive domains. This work addresses this gap by introducing differentially private active learning (DP-AL) for standard learning settings. We demonstrate that naively integrating DP-SGD training into AL presents substantial challenges in privacy budget allocation and data utilization. To overcome these challenges, we propose step amplification, which leverages individual sampling probabilities in batch creation to maximize data point participation in training steps, thus optimizing data utilization. Additionally, we investigate the effectiveness of various acquisition functions for data selection under privacy constraints, revealing that many commonly used functions become impractical. Our experiments on vision and natural language processing tasks show that DP-AL can improve performance for specific datasets and model architectures. However, our findings also highlight the limitations of AL in privacy-constrained environments, emphasizing the trade-offs between privacy, model accuracy, and data selection accuracy.

In this paper we obtain the existence of global attractors for the dynamical systems generated by weak solution of the three-dimensional Navier-Stokes equations with damping. We consider two cases, depending on the values of the parameters \b{eta},{\alpha} controlling the damping term and the viscosity {\mu}. First, for \b{eta} we define a multivalued dynamical systems and prove the existence of the global attractor as well. Second, for either \b{eta}>3 or \b{eta}=3, 4{\alpha}{\mu}>1 the weak solutions are unique and we prove that the global attractor for the corresponding semigroup is more regular. Also, we prove in this case that it is the global attractor for the semigroup generated by the strong solutions. Finally, some numerical simulations are performed.

The Multi-Object Search (MOS) problem involves navigating to a sequence of locations to maximize the likelihood of finding target objects while minimizing travel costs. In this paper, we introduce a novel approach to the MOS problem, called Finder, which leverages vision language models (VLMs) to locate multiple objects across diverse environments. Specifically, our approach introduces multi-channel score maps to track and reason about multiple objects simultaneously during navigation, along with a score fusion technique that combines scene-level and object-level semantic correlations. Experiments in both simulated and real-world settings showed that Finder outperforms existing methods using deep reinforcement learning and VLMs. Ablation and scalability studies further validated our design choices and robustness with increasing numbers of target objects, respectively. Website: https://find-all-my-things.github.io/, Comment: 12 pages, 6 figures, submitted to ICRA2025

The Earth's atmosphere is comprised of turbulent layers that result in speckled and blurry images from ground-based visible and infrared observations. Adaptive Optics (AO) systems are employed to measure the perturbed wavefront with a wavefront sensor (WFS) and correct for these distortions with a deformable mirror. Therefore, understanding and characterising the atmosphere is crucial for the design and functionality of AO systems. One parameter for characterizing the atmosphere is the atmospheric coherence time, which is a function of the effective wind velocity of the atmosphere. This parameter dictates how fast the AO system needs to correct for the atmosphere. If not fast enough, phenomena such as the wind butterfly effect can occur, hindering high-contrast coronographic imaging. This effect is a result of fast, strong, high-altitude turbulent layers. This paper presents two methods for estimating the effective wind velocity, using pseudo-open loop WFS slopes. The first method uses a spatial-temporal covariance map and the second uses the power spectral density of the defocus term. We show both simulated results and preliminary results from the Gemini Planet Imager AO telemetry., Comment: 7 pages, 7 figures, submitted to SPIE Astronomical + Telescopes 2024

This paper examines the performance of Random Forest models in forecasting short-term monthly inflation in Argentina, based on a database of monthly indicators since 1962. It is found that these models achieve forecast accuracy that is statistically comparable to the consensus of market analysts' expectations surveyed by the Central Bank of Argentina (BCRA) and to traditional econometric models. One advantage of Random Forest models is that, as they are non-parametric, they allow for the exploration of nonlinear effects in the predictive power of certain macroeconomic variables on inflation. Among other findings, the relative importance of the exchange rate gap in forecasting inflation increases when the gap between the parallel and official exchange rates exceeds 60%. The predictive power of the exchange rate on inflation rises when the BCRA's net international reserves are negative or close to zero (specifically, below USD 2 billion). The relative importance of inflation inertia and the nominal interest rate in forecasting the following month's inflation increases when the nominal levels of inflation and/or interest rates rise., Comment: in Spanish language, 13 pages, 8 figures

MAXI J1803-298 is a transient black hole candidate discovered in May of 2021 during an outburst that lasted several months. Multiple X-ray observations reveal recurring "dipping" intervals in several of its light curves, particularly during the hard/intermediate states, with a typical recurrence period of $\sim7\,\mathrm{hours}$. We report analysis of four NuSTAR observations of the source, supplemented with NICER data where available, over the duration of the outburst evolution covering the hard, intermediate and the soft states. Reflection spectroscopy reveals the black hole to be rapidly spinning ($a_*=0.990\pm{0.001}$) with a near edge-on viewing angle ($i=70\pm{1}\deg$). Additionally, we show that the light-curve dips are caused by photo-electric absorption from a moderately ionized absorber whose origin is not fully understood, although it is likely linked to material from the companion star impacting the outer edges of the accretion disk. We further detect absorption lines in some of the spectra, potentially associated with Fe XXV and Fe XXVI, indicative of disk winds with moderate to extreme velocities. During the intermediate state and just before transitioning into the soft state, the source showed a sudden flux increase which we found to be dominated by soft disk photons and consistent with the filling of the inner accretion disk, at the onset of state transition. In the soft state, we show that models of disk self-irradiation provide a better fit and a preferred explanation to the broadband reflection spectrum, consistent with previous studies of other accreting sources., Comment: 21 pages, 9 figures, accepted for publication in ApJ

It is common to reject undesired outputs of Large Language Models (LLMs); however, current methods to do so require an excessive amount of computation, or severely distort the distribution of outputs. We present a method to balance the distortion of the output distribution with computational efficiency, allowing for the generation of long sequences of text with difficult-to-satisfy constraints, with less amplification of low probability outputs compared to existing methods. We show through a series of experiments that the task-specific performance of our method is comparable to methods that do not distort the output distribution, while being much more computationally efficient., Comment: 9 pages main, 22 pages total

We study the problem of learning an approximate equilibrium in the offline multi-agent reinforcement learning (MARL) setting. We introduce a structural assumption -- the interaction rank -- and establish that functions with low interaction rank are significantly more robust to distribution shift compared to general ones. Leveraging this observation, we demonstrate that utilizing function classes with low interaction rank, when combined with regularization and no-regret learning, admits decentralized, computationally and statistically efficient learning in offline MARL. Our theoretical results are complemented by experiments that showcase the potential of critic architectures with low interaction rank in offline MARL, contrasting with commonly used single-agent value decomposition architectures.

Drug-releasing implants are gaining momentum in the treatment of glaucoma. Implants present however several limitations. Among these limitations, there is the inability to provide an even distribution of the drug in the trabecular meshwork. CFD simulations were used in this work to study the interplay between ocular fluid dynamics and drug diffusion to explore the options to achieve a homogeneous drug distribution. The analysis finds that the balance between convection and diffusion flux hinders mixing within the eye's anterior chamber. This results in highly localized drug delivery in the trabecular meshwork when using a gravity-driven location implant. The results also show that varying the size of the implant can help to solve the issue. Given the natural variability of the trabecular meshwork size amongst patients, implants with personalized size may become a potential solution. The location of the implant within the eye is also key to effective drug delivery. Natural laminar flow of aqueous humor within the eye's anterior chamber prevents achieving an even drug distribution at the target tissue for gravity-driven location implants. However, the iris-lens gap can be utilized as a natural mixer when placing the implant in the eye's posterior chamber, thus helping with effective delivery.

With the demand for scalable cryogenic microwave circuitry continuously rising, recently developed flexible microwave striplines offer the tantalyzing perspective of increasing the cabling density by an order of magnitude without thermally overloading the cryostat. We use a superconducting quantum circuit to test the thermalization of input flex cables with integrated $60\,$dB of attenuation distributed at various temperature stages. From the measured decoherence rate of a superconducting fluxonium qubit, we estimate a residual population of the readout resonator below $3.5\cdot10^{-3}$ photons and we measure a $0.28\,$ms thermalization time for the flexible stripline attenuators. Furthermore, we confirm that the qubit reaches an effective temperature of $26.4\,$mK, close to the base temperature of the cryostat, practically the same as when using a conventional semi-rigid coaxial cable setup.

We consider minimum time multicasting problems in directed and undirected graphs: given a root node and a subset of $t$ terminal nodes, multicasting seeks to find the minimum number of rounds within which all terminals can be informed with a message originating at the root. In each round, the telephone model we study allows the information to move via a matching from the informed nodes to the uninformed nodes. Since minimum time multicasting in digraphs is poorly understood compared to the undirected variant, we study an intermediate problem in undirected graphs that specifies a target $k < t$, and requires the only $k$ of the terminals be informed in the minimum number of rounds. For this problem, we improve implications of prior results and obtain an $\tilde{O}(t^{1/3})$ multiplicative approximation. For the directed version, we obtain an {\em additive} $\tilde{O}(k^{1/2})$ approximation algorithm (with a poly-logarithmic multiplicative factor). Our algorithms are based on reductions to the related problems of finding $k$-trees of minimum poise (sum of maximum degree and diameter) and applying a combination of greedy network decomposition techniques and set covering under partition matroid constraints., Comment: @APPROX24

The reasoning steps generated by LLMs might be incomplete, as they mimic logical leaps common in everyday communication found in their pre-training data: underlying rationales are frequently left implicit (unstated). To address this challenge, we introduce RATIONALYST, a model for process-supervision of reasoning based on pre-training on a vast collection of rationale annotations extracted from unlabeled data. We extract 79k rationales from web-scale unlabelled dataset (the Pile) and a combination of reasoning datasets with minimal human intervention. This web-scale pre-training for reasoning allows RATIONALYST to consistently generalize across diverse reasoning tasks, including mathematical, commonsense, scientific, and logical reasoning. Fine-tuned from LLaMa-3-8B, RATIONALYST improves the accuracy of reasoning by an average of 3.9% on 7 representative reasoning benchmarks. It also demonstrates superior performance compared to significantly larger verifiers like GPT-4 and similarly sized models fine-tuned on matching training sets., Comment: Our code, data, and model can be found at this repository: https://github.com/JHU-CLSP/Rationalyst

Galaxy clustering provides a powerful way to probe cosmology. This requires understanding of the background mean density of galaxy samples, which is estimated from the survey itself by averaging the observed galaxy number density over the angular position. The angle average includes not only the background mean density, but also the monopole fluctuation at each redshift. Here for the first time we compute the monopole fluctuations in galaxy surveys and investigate their impact on galaxy clustering. The monopole fluctuations vary as a function of redshift, and it is correlated with other fluctuations, affecting the two-point correlation function measurements. In an idealized all-sky survey, the rms fluctuation at $z=0.5$ can be as large as 7% of the two-point correlation function in amplitude at the BAO scale, and it becomes smaller than 1% at $z>2$. The monopole fluctuations are unavoidable, but they can be modeled. We discuss its relation to the integral constraint and the implications for the galaxy clustering analysis., Comment: 6 pages, 2 figures, submitted to ApJ Letters

We present a cost-effective approach for developing Automatic Speech Recognition (ASR) models for low-resource languages like Ika. We fine-tune the pretrained wav2vec 2.0 Massively Multilingual Speech Models on a high-quality speech dataset compiled from New Testament Bible translations in Ika. Our results show that fine-tuning multilingual pretrained models achieves a Word Error Rate (WER) of 0.5377 and Character Error Rate (CER) of 0.2651 with just over 1 hour of training data. The larger 1 billion parameter model outperforms the smaller 300 million parameter model due to its greater complexity and ability to store richer speech representations. However, we observe overfitting to the small training dataset, reducing generalizability. Our findings demonstrate the potential of leveraging multilingual pretrained models for low-resource languages. Future work should focus on expanding the dataset and exploring techniques to mitigate overfitting., Comment: 10 pages, 5 Figures This is a pre-release version