Search

Existing works that predict bedform-induced hyporheic exchange flux (HEF) typically either assume a simplified streambed shape and corresponding sinusoidal head distribution or rely on costly computational fluid dynamics simulations. Experimental data have been lacking for the formulation of a priori prediction rules for hydraulic head and flux distributions induced by spatiotemporally heterogeneous natural bedforms because it has not previously been feasible to determine these in the laboratory. We address this problem, presenting a noninvasive technique for regularized inversion of photographic time series of dye front propagation in the hyporheic zone, compatible with arbitrarily-shaped, generally transient bedforms. We employ the technique to analyze three bench-scale flume experiments performed under different flow regimes, presenting a new data set of digitized bed profiles, corresponding head distributions, and dye fronts. To our knowledge, this is the first such data set collated for naturally-formed sand bedforms.

We present the results of a nearly decade-long photometric reverberation mapping (PRM) survey of the H$\alpha$ emission line in nearby ($0.01\lesssim z \lesssim0.05$) Seyfert-Galaxies using small ($15\,\mathrm{cm}-40\,\mathrm{cm}$) telescopes. Broad-band filters were used to trace the continuum emission, while narrow-band filters tracked the H$\alpha$-line signal. We introduce a new PRM formalism to determine the time delay between continuum and line emission using combinations of auto- and cross-correlation functions. We obtain robust delays for 33/80 objects, allowing us to estimate the broad-line region (BLR) size. Additionally, we measure multi-epoch delays for 6 objects whose scatter per source is smaller than the scatter in the BLR size-luminosity relation. Our study enhances the existing H$\alpha$ size-luminosity relation by adding high-quality results for 31 objects, whose nuclear luminosities were estimated using the flux-variation gradient method, resulting in a scatter of 0.26dex within our sample. The scatter reduces to 0.17dex when the 6 lowest luminosity sources are discarded, which is comparable to that found for the H$\beta$ line. Single-epoch spectra enable us to estimate black hole masses using the H$\alpha$ line and derive mass accretion rates from the iron-blend feature adjacent to H$\beta$. A similar trend, as previously reported for the H$\beta$ line, is implied whereby highly accreting objects tend to lie below the size-luminosity relation of the general population. Our work demonstrates the effectiveness of small telescopes in conducting high-fidelity PRM campaigns of prominent emission lines in bright active galactic nuclei.

This article proposes a new integration of linguistic anthropology and machine learning (ML) around convergent interests in both the underpinnings of language and making language technologies more socially responsible. While linguistic anthropology focuses on interpreting the cultural basis for human language use, the ML field of interpretability is concerned with uncovering the patterns that Large Language Models (LLMs) learn from human verbal behavior. Through the analysis of a conversation between a human user and an LLM-powered chatbot, we demonstrate the theoretical feasibility of a new, conjoint field of inquiry, cultural interpretability (CI). By focusing attention on the communicative competence involved in the way human users and AI chatbots co-produce meaning in the articulatory interface of human-computer interaction, CI emphasizes how the dynamic relationship between language and culture makes contextually sensitive, open-ended conversation possible. We suggest that, by examining how LLMs internally "represent" relationships between language and culture, CI can: (1) provide insight into long-standing linguistic anthropological questions about the patterning of those relationships; and (2) aid model developers and interface designers in improving value alignment between language models and stylistically diverse speakers and culturally diverse speech communities. Our discussion proposes three critical research axes: relativity, variation, and indexicality., Comment: Accepted for publication in Big Data & Society, November 2, 2024

The study of van der Waals heterostructures with an interlayer twist, known as "twistronics", has been instrumental in advancing contemporary condensed matter research. Most importantly, it has underpinned the emergence of a multitude of strongly-correlated phases, many of which derive from the topology of the physical system. Here, we explore the application of the twistronics paradigm in plasmonic systems with nontrivial topology, by creating a moir\'e skyrmion superlattice using two superimposed plasmonic skyrmion lattices, twisted at a "magic" angle. The complex electric field distribution of the moir\'e skyrmion superlattice is measured using time-resolved vector microscopy, revealing that each super-cell possesses very large topological invariants and harbors a "skyrmion bag", the size of which is controllable by the twist angle and center of rotation. Our work shows how twistronics leads to a diversity of topological features in optical fields, providing a new route to locally manipulate electromagnetic field distributions, which is crucial for future structured light-matter interaction.

In the current NISQ era, there is demand for functional quantum devices to solve relevant computational problems, which motivates a utilitarian perspective on device design: The goal is to create a device that is able to run a given algorithm with state-of-the-art performance. In this work, we use optimal control tools to derive the gate set required by a toy algorithm and, in tandem, explore the model space of superconducting quantum computer design, from dispersively coupled to stronger interacting qubits, to maximize gate fidelity. We employ perfect entangler theory to provide flexibility in the search for a two-qubit gate on a given platform and to compare designs with different entangling mechanisms, e.g., $\texttt{CPHASE}$ and $\sqrt{\texttt{iSWAP}}$. To ensure the applicability of our investigation, we limit ourselves to "simple" (i.e., sparse parametrization) pulses and quantify, where results differ from using the full complexity of piecewise constant controls., Comment: 7 pages, 5 figures

Open quantum systems are ubiquitous in nature and central to quantum technologies. A common description of their dynamics is given by the celebrated Lindblad master equation, which can be generalized to the non-Markovian scenario. In this work, we introduce the Non-Markovian Ensemble Propagation (NMEP) method, which extends the Monte Carlo Wave-Function (MCWF) method to the non-Markovian case in a simple and general manner. We demonstrate its accuracy and effectiveness in a selection of examples, and compare the results with either analytic expressions or direct numerical integration of the master equation., Comment: 16 pages total, 10 figures, 7 Pages of appendices

The local (z=0.0315) AGN Mrk 817, was monitored over more than 500 days with space-borne and ground-based instruments as part of a large international campaign AGN STORM 2. Here, we present a comprehensive analysis of the broad-band continuum variations using detailed modeling of the broad line region (BLR), several types of disk winds classified by their optical depth, and new numerical simulations. We find that diffuse continuum (DC) emission, with additional contributions from strong and broad emission lines, can explain the continuum lags observed in this source during high and low luminosity phases. Disk illumination by the variable X-ray corona contributes only a small fraction of the observed continuum lags. Our BLR models assume radiation pressure-confined clouds distributed over a distance of 2-122 light days. We present calculated mean-emissivity radii of many emission lines, and DC emission, and suggest a simple, transfer-function-dependent method that ties them to cross-correlation lag determinations. We do not find clear indications for large optical depth winds but identify the signature of lower column density winds. In particular, we associate the shortest observed continuum lags with a combination of tau(1 Ryd) approx. 2 wind and a partly shielded BLR. Even smaller optical depth winds may be associated with X-ray absorption features and with noticeable variations in the width and lags of several high ionization lines like HeII and CIV. Finally, we demonstrate the effect of torus dust emission on the observed lags in the i and z bands., Comment: 23 pages, 18 figures. Corrected typographical error in the title of the paper as it appeared in the Metadata

Following significant progress in the visualization and characterization of hybrid superconducting-semiconducting systems, greatly propelled by reports of Majorana zero modes in nanowire devices, considerable attention has been devoted to investigating the electronic structure at the buried superconducting-semiconducting interface and the nature of the induced superconducting correlations. The properties of that interface and the structure of the electronic wave functions that occupy it determine the functionality and the topological nature of the induced superconducting state. Here, we introduce a novel hybrid platform for proximity-inducing superconductivity in InAs$_{0.6}$Sb$_{0.4}$ nanowires, leveraging a unique architecture and material combination. By dispersing these nanowires over a superconducting Indium film we exploit Indium's high critical temperature of 3.7~K and the anticipated high spin-orbit and Zeeman couplings of InAs$_{0.6}$Sb$_{0.4}$. This design preserves the pristine top facet of the nanowires, making it highly compatible with scanning tunneling spectroscopy. Using this architecture we demonstrate that the mechanical contact supports Cooper-pair transparency as high as 90\%, comparable with epitaxial interfaces. The anisotropic angular response to an applied magnetic field shows the quasi-two-dimensional nature of the parent superconductivity in the Indium film and the quasi-one-dimensional nature of the induced superconductivity in the nanowires. Our platform offers robust and advantageous foundations for studying the emergence of topological superconductivity and the interplay of superconductivity and magnetism using atomic-scale spectroscopic tools., Comment: 30 pages, 20 figures

Quantum Electrodynamics in 2+1 dimensions (QED$_3$) with two Dirac fermions displays time reversal symmetry, nontrivial SPT phases and anomalies. The fate of this theory in its strongly coupled regime has been debated extensively. Surprisingly, we find that gluing together the phase diagrams of two standard Wilson-Fisher $O(4)$ theories suffices to reproduce all the SPT phases, anomalies, and semi-classical limits. A central mechanism behind it is ``SPT absorption''. The patching of the $O(4)$ transitions makes very concrete predictions for the behavior of the theory in its strongly coupled limits; for instance, the $\theta=\pi$ sigma model with $S^3$ topology appears due to monopole condensation., Comment: 6 pages, 3 figures, v2 submitted for publication

Memory has become the primary cost driver in cloud data centers. Yet, a significant portion of memory allocated to VMs in public clouds remains unused. To optimize this resource, "cold" memory can be reclaimed from VMs and stored on slower storage or compressed, enabling memory overcommit. Current overcommit systems rely on general-purpose OS swap mechanisms, which are not optimized for virtualized workloads, leading to missed memory-saving opportunities and ineffective use of optimizations like prefetchers. This paper introduces a userspace memory management framework designed for VMs. It enables custom policies that have full control over the virtual machines' memory using a simple userspace API, supports huge page-based swapping to satisfy VM performance requirements, is easy to deploy by leveraging Linux/KVM, and supports zero-copy I/O virtualization with shared VM memory. Our evaluation demonstrates that an overcommit system based on our framework outperforms the state-of-the-art solutions on both micro-benchmarks and commonly used cloud workloads. Specifically our implementation outperforms the Linux Kernel baseline implementation by up to 25% while saving a similar amount of memory. We also demonstrate the benefits of custom policies by implementing workload-specific reclaimers and prefetchers that save $10\%$ additional memory, improve performance in a limited memory scenario by 30% over the Linux baseline, and recover faster from hard limit releases., Comment: 13 pages, 13 figures

X-ray reverberation mapping is a powerful technique for probing the innermost accretion disk, whereas continuum reverberation mapping in the UV, optical, and infrared (UVOIR) reveals reprocessing by the rest of the accretion disk and broad-line region (BLR). We present the time lags of Mrk 817 as a function of temporal frequency measured from 14 months of high-cadence monitoring from Swift and ground-based telescopes, in addition to an XMM-Newton observation, as part of the AGN STORM 2 campaign. The XMM-Newton lags reveal the first detection of a soft lag in this source, consistent with reverberation from the innermost accretion flow. These results mark the first simultaneous measurement of X-ray reverberation and UVOIR disk reprocessing lags$\unicode{x2013}$effectively allowing us to map the entire accretion disk surrounding the black hole. Similar to previous continuum reverberation mapping campaigns, the UVOIR time lags arising at low temporal frequencies are longer than those expected from standard disk reprocessing by a factor of 2-3. The lags agree with the anticipated disk reverberation lags when isolating short-timescale variability, namely timescales shorter than the H$\beta$ lag. Modeling the lags requires additional reprocessing constrained at a radius consistent with the BLR size scale inferred from contemporaneous H$\beta$-lag measurements. When we divide the campaign light curves, the UVOIR lags show substantial variations, with longer lags measured when obscuration from an ionized outflow is greatest. We suggest that, when the obscurer is strongest, reprocessing by the BLR elongates the lags most significantly. As the wind weakens, the lags are dominated by shorter accretion disk lags., Comment: 28 pages, 11 figures, 3 tables. Accepted for publication in the Astrophysical Journal

Cloud observables such as precipitation efficiency and cloud lifetime are key quantities in weather and climate, but understanding their quantitative connection to initial conditions such as initial cloud water mass or droplet size remains challenging. Here we study the evolution of cloud droplets with a bin microphysics scheme, modeling both gravitational coagulation as well as fallout, and develop analytical formulae to describe the evolution of bulk cloud and rain water. We separate the dynamics into a mass-conserving and fallout-dominated regime, which reveals that the overall dynamics are governed by a single non-dimensional parameter $\mu$, the ratio of accretion and sedimentation time scales. Cloud observables from the simulations accordingly collapse as a function of $\mu$. We also find an unexpected relationship between cloud water and accumulated rain, and that fallout can be modeled with a bulk fall speed which is constant in time despite an evolving raindrop distribution., Comment: 34 pages, 18 figures

Prompts are how humans communicate with LLMs. Informative prompts are essential for guiding LLMs to produce the desired output. However, prompt engineering is often tedious and time-consuming, requiring significant expertise, limiting its widespread use. We propose Conversational Prompt Engineering (CPE), a user-friendly tool that helps users create personalized prompts for their specific tasks. CPE uses a chat model to briefly interact with users, helping them articulate their output preferences and integrating these into the prompt. The process includes two main stages: first, the model uses user-provided unlabeled data to generate data-driven questions and utilize user responses to shape the initial instruction. Then, the model shares the outputs generated by the instruction and uses user feedback to further refine the instruction and the outputs. The final result is a few-shot prompt, where the outputs approved by the user serve as few-shot examples. A user study on summarization tasks demonstrates the value of CPE in creating personalized, high-performing prompts. The results suggest that the zero-shot prompt obtained is comparable to its - much longer - few-shot counterpart, indicating significant savings in scenarios involving repetitive tasks with large text volumes.

Mass estimates of black holes (BHs) in the centers of Active Galactic Nuclei (AGN) often rely on the radius-luminosity relation. However, this relation, usually probed by reverberation mapping (RM), is poorly constrained in the high-luminosity and high-redshift ends due to the very long expected lag times. Multiply imaged AGN may thus offer a unique opportunity to explore the radius-luminosity relation at these ends. In addition to comprising several magnified images which enable a more efficient light-curve sampling, the time delay between multiple images of strongly lensed quasars can also aid in making such RM measurements feasible on reasonable timescales: If the time delay is, for example, of the order of the expected time lag, changes in the emission lines in the leading image can be observed around the same time as the changes in the continuum in the trailing image. In this work we probe the typical time-delay distribution in galaxy-cluster lenses and estimate the number of both high-mass ($\sim10^9-10^{10}$ M$_{\odot}$), and high-redshift ($z\gtrsim4-12$) quasars that are expected to be strongly lensed by clusters. We find that less than one, very massive (and luminous, L$_{UV}>10^{46.5}$ erg s$^{-1}$) multiply imaged quasar is expected across the sky down to 30 AB mag. Nonetheless, up to several dozen thousand M$_{BH}\sim10^{6}$-$10^{8}$ M$_{\odot}$ broad-line AGN at $z>4$ should be multiply imaged by galaxy clusters and detectable with JWST, hundreds with $\textit{Euclid}$ and several thousands with the $\textit{Roman}$ Space Telescope, across the whole sky. These could supply an important calibration for the BH mass scaling in the early Universe., Comment: To be submitted. Comments welcome. 14 pages, 3 figures, 1 table

Fine-tuning Large Language Models (LLMs) is an effective method to enhance their performance on downstream tasks. However, choosing the appropriate setting of tuning hyperparameters (HPs) is a labor-intensive and computationally expensive process. Here, we provide recommended HP configurations for practical use-cases that represent a better starting point for practitioners, when considering two SOTA LLMs and two commonly used tuning methods. We describe Coverage-based Search (CBS), a process for ranking HP configurations based on an offline extensive grid search, such that the top ranked configurations collectively provide a practical robust recommendation for a wide range of datasets and domains. We focus our experiments on Llama-3-8B and Mistral-7B, as well as full fine-tuning and LoRa, conducting a total of > 10,000 tuning experiments. Our results suggest that, in general, Llama-3-8B and LoRA should be preferred, when possible. Moreover, we show that for both models and tuning methods, exploring only a few HP configurations, as recommended by our analysis, can provide excellent results in practice, making this work a valuable resource for practitioners.

In recent years, notable advancements have been made in the domain of visual document understanding, with the prevailing architecture comprising a cascade of vision and language models. The text component can either be extracted explicitly with the use of external OCR models in OCR-based approaches, or alternatively, the vision model can be endowed with reading capabilities in OCR-free approaches. Typically, the queries to the model are input exclusively to the language component, necessitating the visual features to encompass the entire document. In this paper, we present VisFocus, an OCR-free method designed to better exploit the vision encoder's capacity by coupling it directly with the language prompt. To do so, we replace the down-sampling layers with layers that receive the input prompt and allow highlighting relevant parts of the document, while disregarding others. We pair the architecture enhancements with a novel pre-training task, using language masking on a snippet of the document text fed to the visual encoder in place of the prompt, to empower the model with focusing capabilities. Consequently, VisFocus learns to allocate its attention to text patches pertinent to the provided prompt. Our experiments demonstrate that this prompt-guided visual encoding approach significantly improves performance, achieving state-of-the-art results on various benchmarks., Comment: 32 pages, 18 figures

We observed the Seyfert 1 galaxy Mrk817 during an intensive multi-wavelength reverberation mapping campaign for 16 months. Here, we examine the behavior of narrow UV absorption lines seen in HST/COS spectra, both during the campaign and in other epochs extending over 14 years. We conclude that while the narrow absorption outflow system (at -3750 km/s with FWHM=177 km/s) responds to the variations of the UV continuum as modified by the X-ray obscurer, its total column density (logNH =19.5 cm-2) did not change across all epochs. The adjusted ionization parameter (scaled with respect to the variations in the Hydrogen ionizing continuum flux) is log UH =-1.0. The outflow is located at a distance smaller than 38 parsecs from the central source, which implies a hydrogen density of nH > 3000 cm-3. The absorption outflow system only covers the continuum emission source and not the broad emission line region, which suggests that its transverse size is small (< 1e16 cm), with potential cloud geometries ranging from spherical to elongated along the line of sight., Comment: 18 pages, 12 Figures, Accepted for publication in ApJ

The ability to map left ventricle (LV) myocardial motion using computed tomography angiography (CTA) is essential to diagnosing cardiovascular conditions and guiding interventional procedures. Due to their inherent locality, conventional neural networks typically have difficulty predicting subtle tangential movements, which considerably lessens the level of precision at which myocardium three-dimensional (3D) mapping can be performed. Using 3D optical flow techniques and Functional Maps (FMs), we present a comprehensive approach to address this problem. FMs are known for their capacity to capture global geometric features, thus providing a fuller understanding of 3D geometry. As an alternative to traditional segmentation-based priors, we employ surface-based two-dimensional (2D) constraints derived from spectral correspondence methods. Our 3D deep learning architecture, based on the ARFlow model, is optimized to handle complex 3D motion analysis tasks. By incorporating FMs, we can capture the subtle tangential movements of the myocardium surface precisely, hence significantly improving the accuracy of 3D mapping of the myocardium. The experimental results confirm the effectiveness of this method in enhancing myocardium motion analysis. This approach can contribute to improving cardiovascular diagnosis and treatment. Our code and additional resources are available at: https://shaharzuler.github.io/CardioSpectrumPage, Comment: This paper has been early accepted to MICCAI 2024, LNCS 15005, Springer, 2024

We report results of intensive time-resolved imaging photometry and synoptic deep imaging of the comet C/2012 S1 (ISON) performed in February 2013. The data were obtained at the Wise Observatory in Israel (WO), at the Himalayan Chandra Telescope (HCT) in India, and at the Polaris Observatory Association in California, USA. During this period, the comet's heliocentric distance changed from 4.9 to 4.6 AU, just within the orbit of Jupiter. We analyze these early images in an attempt to determine the nuclear rotation period, assuming that at these relatively large heliocentric distances it would be possible to detect the photometric modulation of a rotating nucleus against an underdeveloped coma. Since this is not evident in our February 2013 data, with more than 400 independent photometric measurements analyzed, we can only set upper limits of 0.05 mag for periodic brightness modulations. We discuss (and discount) a possible brightening event (minor outburst) that occurred on $15-16$ February 2013. We also present deep synoptic images of the comet, obtained by combining our exposures for each night, and analyze them. We find that during the period of our observations the comet exhibited a $\sim$$30^{\prime\prime}\simeq 60000$-km tail with no substructures visible and that this appearance did not change throughout our campaign. The comet, as indicated by a single spectroscopic measurement obtained during this observation period, showed a dust coma reflecting the solar light. Our observations indicate that during February 2013, comet ISON was relatively quiet, with the dust coma presumably hiding any light modulation by a spinning nucleus., Comment: 15 pages, 10 figures

In this paper, we take a new approach to autoregressive image generation that is based on two main ingredients. The first is wavelet image coding, which allows to tokenize the visual details of an image from coarse to fine details by ordering the information starting with the most significant bits of the most significant wavelet coefficients. The second is a variant of a language transformer whose architecture is re-designed and optimized for token sequences in this 'wavelet language'. The transformer learns the significant statistical correlations within a token sequence, which are the manifestations of well-known correlations between the wavelet subbands at various resolutions. We show experimental results with conditioning on the generation process., Comment: 16 pages, 10 figures

We examine the relationship between learnability and robust (or agnostic) learnability for the problem of distribution learning. We show that, contrary to other learning settings (e.g., PAC learning of function classes), realizable learnability of a class of probability distributions does not imply its agnostic learnability. We go on to examine what type of data corruption can disrupt the learnability of a distribution class and what is such learnability robust against. We show that realizable learnability of a class of distributions implies its robust learnability with respect to only additive corruption, but not against subtractive corruption. We also explore related implications in the context of compression schemes and differentially private learnability., Comment: In NeurIPS 2023

We present the results of the XMM-Newton and NuSTAR observations taken as part of the ongoing, intensive multi-wavelength monitoring program of the Seyfert 1 galaxy Mrk 817 by the AGN Space Telescope and Optical Reverberation Mapping 2 (AGN STORM 2) Project. The campaign revealed an unexpected and transient obscuring outflow, never before seen in this source. Of our four XMM-Newton/NuSTAR epochs, one fortuitously taken during a bright X-ray state has strong narrow absorption lines in the high-resolution grating spectra. From these absorption features, we determine that the obscurer is in fact a multi-phase ionized wind with an outflow velocity of $\sim$5200 km s$^{-1}$, and for the first time find evidence for a lower ionization component with the same velocity observed in absorption features in the contemporaneous HST spectra. This indicates that the UV absorption troughs may be due to dense clumps embedded in diffuse, higher ionization gas responsible for the X-ray absorption lines of the same velocity. We observe variability in the shape of the absorption lines on timescales of hours, placing the variable component at roughly 1000 $R_g$ if attributed to transverse motion along the line of sight. This estimate aligns with independent UV measurements of the distance to the obscurer suggesting an accretion disk wind at the inner broad line region. We estimate that it takes roughly 200 days for the outflow to travel from the disk to our line of sight, consistent with the timescale of the outflow's column density variations throughout the campaign.

We develop a method to generate prediction sets with a guaranteed coverage rate that is robust to corruptions in the training data, such as missing or noisy variables. Our approach builds on conformal prediction, a powerful framework to construct prediction sets that are valid under the i.i.d assumption. Importantly, naively applying conformal prediction does not provide reliable predictions in this setting, due to the distribution shift induced by the corruptions. To account for the distribution shift, we assume access to privileged information (PI). The PI is formulated as additional features that explain the distribution shift, however, they are only available during training and absent at test time. We approach this problem by introducing a novel generalization of weighted conformal prediction and support our method with theoretical coverage guarantees. Empirical experiments on both real and synthetic datasets indicate that our approach achieves a valid coverage rate and constructs more informative predictions compared to existing methods, which are not supported by theoretical guarantees.

Cuneiform tablets, emerging in ancient Mesopotamia around the late fourth millennium BCE, represent one of humanity's earliest writing systems. Characterized by wedge-shaped marks on clay tablets, these artifacts provided insight into Mesopotamian civilization across various domains. Traditionally, the analysis and dating of these tablets rely on subjective assessment of shape and writing style, leading to uncertainties in pinpointing their exact temporal origins. Recent advances in digitization have revolutionized the study of cuneiform by enhancing accessibility and analytical capabilities. Our research uniquely focuses on the silhouette of tablets as significant indicators of their historical periods, diverging from most studies that concentrate on textual content. Utilizing an unprecedented dataset of over 94,000 images from the Cuneiform Digital Library Initiative collection, we apply deep learning methods to classify cuneiform tablets, covering over 3,000 years of history. By leveraging statistical, computational techniques, and generative modeling through Variational Auto-Encoders (VAEs), we achieve substantial advancements in the automatic classification of these ancient documents, focusing on the tablets' silhouettes as key predictors. Our classification approach begins with a Decision Tree using height-to-width ratios and culminates with a ResNet50 model, achieving a 61% macro F1-score for tablet silhouettes. Moreover, we introduce novel VAE-powered tools to enhance explainability and enable researchers to explore changes in tablet shapes across different eras and genres. This research contributes to document analysis and diplomatics by demonstrating the value of large-scale data analysis combined with statistical methods. These insights offer valuable tools for historians and epigraphists, enriching our understanding of cuneiform tablets and the cultures that produced them., Comment: 24 pages, 18 figures

Conventional 2D pose estimation models are constrained by their design to specific object categories. This limits their applicability to predefined objects. To overcome these limitations, category-agnostic pose estimation (CAPE) emerged as a solution. CAPE aims to facilitate keypoint localization for diverse object categories using a unified model, which can generalize from minimal annotated support images. Recent CAPE works have produced object poses based on arbitrary keypoint definitions annotated on a user-provided support image. Our work departs from conventional CAPE methods, which require a support image, by adopting a text-based approach instead of the support image. Specifically, we use a pose-graph, where nodes represent keypoints that are described with text. This representation takes advantage of the abstraction of text descriptions and the structure imposed by the graph. Our approach effectively breaks symmetry, preserves structure, and improves occlusion handling. We validate our novel approach using the MP-100 benchmark, a comprehensive dataset spanning over 100 categories and 18,000 images. Under a 1-shot setting, our solution achieves a notable performance boost of 1.07\%, establishing a new state-of-the-art for CAPE. Additionally, we enrich the dataset by providing text description annotations, further enhancing its utility for future research.

What computational structure are we building into large language models when we train them on next-token prediction? Here, we present evidence that this structure is given by the meta-dynamics of belief updating over hidden states of the data-generating process. Leveraging the theory of optimal prediction, we anticipate and then find that belief states are linearly represented in the residual stream of transformers, even in cases where the predicted belief state geometry has highly nontrivial fractal structure. We investigate cases where the belief state geometry is represented in the final residual stream or distributed across the residual streams of multiple layers, providing a framework to explain these observations. Furthermore we demonstrate that the inferred belief states contain information about the entire future, beyond the local next-token prediction that the transformers are explicitly trained on. Our work provides a general framework connecting the structure of training data to the geometric structure of activations inside transformers.

We describe a novel subgradient following apparatus for calculating the optimum of convex problems with variational penalties. In this setting, we receive a sequence $y_i,\ldots,y_n$ and seek a smooth sequence $x_1,\ldots,x_n$. The smooth sequence attains the minimum Bregman divergence to an input sequence with additive variational penalties in the general form of $\sum_i g_i(x_{i+1}-x_i)$. We derive, as special cases of our apparatus, known algorithms for the fused lasso and isotonic regression. Our approach also facilitates new variational penalties such as non-smooth barrier functions. We next derive and analyze multivariate problems in which $\mathbf{x}_i,\mathbf{y}_i\in\mathbb{R}^d$ and variational penalties that depend on $\|\mathbf{x}_{i+1}-\mathbf{x}_i\|$. The norms we consider are $\ell_2$ and $\ell_\infty$ which promote group sparsity. Last but not least, we derive a lattice-based subgradient following for variational penalties characterized through the output of arbitrary convolutional filters. This paradigm yields efficient solvers for problems in which sparse high-order discrete derivatives such as acceleration and jerk are desirable.

Multi-participant discussions tend to unfold in a tree structure rather than a chain structure. Branching may occur for multiple reasons -- from the asynchronous nature of online platforms to a conscious decision by an interlocutor to disengage with part of the conversation. Predicting branching and understanding the reasons for creating new branches is important for many downstream tasks such as summarization and thread disentanglement and may help develop online spaces that encourage users to engage in online discussions in more meaningful ways. In this work, we define the novel task of branch prediction and propose GLOBS (Global Branching Score) -- a deep neural network model for predicting branching. GLOBS is evaluated on three large discussion forums from Reddit, achieving significant improvements over an array of competitive baselines and demonstrating better transferability. We affirm that structural, temporal, and linguistic features contribute to GLOBS success and find that branching is associated with a greater number of conversation participants and tends to occur in earlier levels of the conversation tree. We publicly release GLOBS and our implementation of all baseline models to allow reproducibility and promote further research on this important task.

3D scene registration is a fundamental problem in computer vision that seeks the best 6-DoF alignment between two scenes. This problem was extensively investigated in the case of point clouds and meshes, but there has been relatively limited work regarding Neural Radiance Fields (NeRF). In this paper, we consider the problem of rigid registration between two NeRFs when the position of the original cameras is not given. Our key novelty is the introduction of Viewshed Fields (VF), an implicit function that determines, for each 3D point, how likely it is to be viewed by the original cameras. We demonstrate how VF can help in the various stages of NeRF registration, with an extensive evaluation showing that VF-NeRF achieves SOTA results on various datasets with different capturing approaches such as LLFF and Objaverese.

We prove a polynomial Bogolyubov type lemma for the special linear group over finite fields. Specifically, we show that there exists an absolute constant $C>0,$ such that if $A$ is a density $\alpha$ subset of the special linear group, then the set $AA^{-1}AA^{-1}$ contains a subgroup $H$ of density $\alpha^C$. Moreover, this subgroup is isomorphic to a special linear group of a smaller rank. We also show that if $A$ is an approximate subgroups then it can be covered by the union of few cosets of $H$. Our proof makes use of the Gurevich--Howe notion of tensor rank, and of a strengthened Bonami type Lemma for global functions on the bilinear scheme. We also present applications to spectral bounds for global convolution operators, global product free sets, and covering numbers corresponding to global sets.

We present Jamba, a new base large language model based on a novel hybrid Transformer-Mamba mixture-of-experts (MoE) architecture. Specifically, Jamba interleaves blocks of Transformer and Mamba layers, enjoying the benefits of both model families. MoE is added in some of these layers to increase model capacity while keeping active parameter usage manageable. This flexible architecture allows resource- and objective-specific configurations. In the particular configuration we have implemented, we end up with a powerful model that fits in a single 80GB GPU. Built at large scale, Jamba provides high throughput and small memory footprint compared to vanilla Transformers, and at the same time state-of-the-art performance on standard language model benchmarks and long-context evaluations. Remarkably, the model presents strong results for up to 256K tokens context length. We study various architectural decisions, such as how to combine Transformer and Mamba layers, and how to mix experts, and show that some of them are crucial in large scale modeling. We also describe several interesting properties of these architectures which the training and evaluation of Jamba have revealed, and plan to release checkpoints from various ablation runs, to encourage further exploration of this novel architecture. We make the weights of our implementation of Jamba publicly available under a permissive license., Comment: Webpage: https://www.ai21.com/jamba

We consider type IIB string theory with $N$ D3 branes and various configurations of sevenbranes, such that the string coupling $g_s$ is fixed to a constant finite value. These are the simplest realizations of F-theory, and are holographically dual to rank $N$ Argyres-Douglas conformal field theories (CFTs) with $SU(2)$ and $SU(3)$ flavor groups, and Minahan-Nemeschansky CFTs with $E_6, E_7$ and $E_8$ flavor groups. We use the Seiberg-Witten curves of these theories to compute the mass deformed sphere free energy $F(m)$ at large $N$ in terms of novel matrix models with non-polynomial potentials. We show how $F(m)$ can be used along with the analytic bootstrap to fix the large $N$ expansion of flavor multiplet correlators in these CFTs, which are dual to scattering of gluons on $AdS_5\times S^3$, and in the flat space limit determine the effective theory of sevenbranes in F-theory. As a first step in this program, we use the matrix models to compute the $\log N$ term in $F(m)$ and thereby fix the logarithmic threshold in the $AdS_5\times S^3$ holographic correlator, which matches the flat space prediction., Comment: 51 pages, matches published version

We present DINO-Tracker -- a new framework for long-term dense tracking in video. The pillar of our approach is combining test-time training on a single video, with the powerful localized semantic features learned by a pre-trained DINO-ViT model. Specifically, our framework simultaneously adopts DINO's features to fit to the motion observations of the test video, while training a tracker that directly leverages the refined features. The entire framework is trained end-to-end using a combination of self-supervised losses, and regularization that allows us to retain and benefit from DINO's semantic prior. Extensive evaluation demonstrates that our method achieves state-of-the-art results on known benchmarks. DINO-tracker significantly outperforms self-supervised methods and is competitive with state-of-the-art supervised trackers, while outperforming them in challenging cases of tracking under long-term occlusions., Comment: Accepted to ECCV 2024. Project page: https://dino-tracker.github.io/

We compute the AdS Veneziano amplitude for type IIB gluon scattering in $AdS_5 \times S^3$ to all orders in $\alpha'$ in a small curvature expansion. This is achieved by combining a dispersion relation in the dual $4d$ $\mathcal{N}=2$ SCFT with an ansatz for the amplitude as a worldsheet integral in terms of multiple polylogarithms. The first curvature correction is fully fixed in this way and satisfies consistency checks in the high energy limit, the low energy expansion as previously fixed using supersymmetric localisation, and for the energy of massive string operators, which we independently compute using a semiclassical expansion. We also combine localisation with this first curvature correction to fix the unprotected $D^4F^4$ correction to the amplitude at finite curvature., Comment: 45 pages, 7 figures

Integrated photonics could bring transformative breakthroughs in computing, networking, imaging, sensing, and quantum information processing, enabled by increasingly sophisticated optical functionalities on a photonic chip. However, wideband optical isolators, which are essential for the robust operation of practically all optical systems, have been challenging to realize in integrated form due to the incompatibility of magnetic media with these circuit technologies. Here, we present the first-ever demonstration of an integrated non-magnetic optical isolator with terahertz-level optical bandwidth. The system is comprised of two acousto-optic frequency-shifting beam splitters which create a non-reciprocal multimode interferometer exhibiting high-contrast, nonreciprocal light transmission. We dramatically enhance the isolation bandwidth of this system by precisely dispersion balancing the paths of the interferometer. Using this approach, we demonstrate integrated nonmagnetic isolators with an optical contrast as high as 28 dB, insertion losses as low as -2.16 dB, and optical bandwidths as high as 2 THz (16 nm). We also show that the center frequency and direction of optical isolation are rapidly reconfigurable by tuning the relative phase of the microwave signals used to drive the acousto-optic beam splitters. With their CMOS compatibility, wideband operation, low losses, and rapid reconfigurability, such integrated isolators could address a key barrier to the integration of a wide range of photonic functionalities on a chip. Looking beyond the current demonstration, this bandwidth-scalable approach to nonmagnetic isolation opens the door to ultrawideband (>10 THz) isolators, which are needed to shrink state-of-the-art imaging, sensing, and communications systems into photonic integrated circuits.

The task of converting natural language queries into SQL queries is intricate, necessitating a blend of precise techniques for an accurate translation. The DIN-SQL (Decomposed-In-Context SQL) methodology represents a significant development in this domain. This paper introduces DFIN (Decomposed Focused-In-Context), an innovative extension of DIN-SQL that enhances Text-to-SQL conversion by addressing schema linking errors, which are a major source of inaccuracies. DFIN uniquely alternates between prompting techniques and Retrieval-Augmented Generation (RAG), adapting to the size and complexity of the database schema. A preprocessing phase embeds database definitions and leverages annotated files, akin to those in the BIRD dataset, facilitating the runtime retrieval of pertinent schema information. This strategy significantly reduces the token count for schema linking prompts, enabling the use of a standard GPT-4 model over its larger context variant, thus handling large-scale databases more effectively and economically. Our evaluation on the BIRD dataset, a challenging real-world benchmark, demonstrates that DFIN not only scales efficiently but also improves accuracy, achieving a score of 51.69. This improvement surpasses DIN-SQL method (the current third-place), which is the highest-ranked model employing in-context learning rather than fine-tuning, previously scoring 50.72. The advancement of DFIN underscores the evolving capabilities of in-context learning methodologies combined with advanced language models, offering a promising avenue for future research in complex Text-to-SQL conversion tasks.