Search

Large language models (LLMs) have demonstrated remarkable capabilities, but their adoption is limited by high computational costs during inference. While increasing parameter counts enhances accuracy, it also widens the gap between state-of-the-art capabilities and practical deployability. We present Puzzle, a framework to accelerate LLM inference on specific hardware while preserving their capabilities. Through an innovative application of neural architecture search (NAS) at an unprecedented scale, Puzzle systematically optimizes models with tens of billions of parameters under hardware constraints. Our approach utilizes blockwise local knowledge distillation (BLD) for parallel architecture exploration and employs mixed-integer programming for precise constraint optimization. We demonstrate the real-world impact of our framework through Llama-3.1-Nemotron-51B-Instruct (Nemotron-51B), a publicly available model derived from Llama-3.1-70B-Instruct. Nemotron-51B achieves a 2.17x inference throughput speedup, fitting on a single NVIDIA H100 GPU while preserving 98.4% of the original model's capabilities. Nemotron-51B currently stands as the most accurate language model capable of inference on a single GPU with large batch sizes. Remarkably, this transformation required just 45B training tokens, compared to over 15T tokens used for the 70B model it was derived from. This establishes a new paradigm where powerful models can be optimized for efficient deployment with only negligible compromise of their capabilities, demonstrating that inference performance, not parameter count alone, should guide model selection. With the release of Nemotron-51B and the presentation of the Puzzle framework, we provide practitioners immediate access to state-of-the-art language modeling capabilities at significantly reduced computational costs.

Malicious URL (Uniform Resource Locator) classification is a pivotal aspect of Cybersecurity, offering defense against web-based threats. Despite deep learning's promise in this area, its advancement is hindered by two main challenges: the scarcity of comprehensive, open-source datasets and the limitations of existing models, which either lack real-time capabilities or exhibit suboptimal performance. In order to address these gaps, we introduce a novel, multi-class dataset for malicious URL classification, distinguishing between benign, phishing and malicious URLs, named DeepURLBench. The data has been rigorously cleansed and structured, providing a superior alternative to existing datasets. Notably, the multi-class approach enhances the performance of deep learning models, as compared to a standard binary classification approach. Additionally, we propose improvements to string-based URL classifiers, applying these enhancements to URLNet. Key among these is the integration of DNS-derived features, which enrich the model's capabilities and lead to notable performance gains while preserving real-time runtime efficiency-achieving an effective balance for cybersecurity applications.

Scintillation, the process of converting high-energy radiation to detectable visible light, is pivotal in advanced technologies spanning from medical diagnostics to fundamental scientific research. Despite significant advancements toward faster and more efficient scintillators, there remains a fundamental limit arising from the intrinsic properties of scintillating materials. The scintillation process culminates in spontaneous emission of visible light, which is restricted in rate by the oscillator strength of individual emission centers. Here, we observe a novel collective emission phenomenon under X-ray excitation, breaking this limit and accelerating the emission. Our observation reveals that strong interactions between simultaneously excited coupled perovskite quantum dots can create collective radioluminescence. This effect is characterized by a spectral shift and an enhanced rate of emission, with an average lifetime of 230 ps, 14 times faster than their room temperature spontaneous emission. It has been established that such quantum dots exhibit superfluorescence under UV excitation. However, X-ray superfluorescence is inherently different, as each high-energy photon creates multiple synchronized excitation events, triggered by a photoelectron and resulting in even faster emission rates, a larger spectral shift, and a broader spectrum. This observation is consistent with a quantum-optical analysis explaining both the UV-driven and X-ray-driven effects. We use a Hanbury-Brown-Twiss g^(2) ({\tau}) setup to analyze the temperature-dependent temporal response of these scintillators. Collective radioluminescence breaks the limit of scintillation lifetime based on spontaneous emission and could dramatically improve time-of-flight detector performance, introducing quantum enhancements to scintillation science.

Hanbury Brown and Twiss (HBT) interferometry is a milestone experiment that transformed our understanding of the nature of light. The concept was demonstrated in 1956 to measure the radii of stars through photon coincidence detection. This form of coincidence detection later became a cornerstone of modern quantum optics. Here we connect HBT interferometry to the physics of scintillation, the process of spontaneous light emission upon excitation by high-energy particles, such as x-rays. Our work reveals intrinsic photon bunching in the scintillation process, which we utilize to elucidate its underlying light emission mechanisms. Specifically, g^((2) ) ({\tau}) enables the quantitative extraction of scintillation lifetime and light yield, showing their dependence on temperature and X-ray flux as well. This approach provides a characterization method that we benchmark on a wide gamut of scintillators, including rare-earth-doped garnets and perovskite nanocrystals. Our method is particularly important for nano- and micro-scale scintillators, whose properties are challenging to quantify by conventional means: We extract the scintillation properties in perovskite nanocrystals of only a few hundreds of nanometers, observing strong photon bunching (g^((2) ) (0)>50). Our research paves the way for broader use of photon-coincidence measurement and methods from quantum optics in studying materials with complex optical properties in extremes regions of the electromagnetic spectrum.

Ongoing efforts that span over decades show a rise of AI methods for accelerating scientific discovery, yet accelerating discovery in mathematics remains a persistent challenge for AI. Specifically, AI methods were not effective in creation of formulas for mathematical constants because each such formula must be correct for infinite digits of precision, with "near-true" formulas providing no insight toward the correct ones. Consequently, formula discovery lacks a clear distance metric needed to guide automated discovery in this realm. In this work, we propose a systematic methodology for categorization, characterization, and pattern identification of such formulas. The key to our methodology is introducing metrics based on the convergence dynamics of the formulas, rather than on the numerical value of the formula. These metrics enable the first automated clustering of mathematical formulas. We demonstrate this methodology on Polynomial Continued Fraction formulas, which are ubiquitous in their intrinsic connections to mathematical constants, and generalize many mathematical functions and structures. We test our methodology on a set of 1,768,900 such formulas, identifying many known formulas for mathematical constants, and discover previously unknown formulas for $\pi$, $\ln(2)$, Gauss', and Lemniscate's constants. The uncovered patterns enable a direct generalization of individual formulas to infinite families, unveiling rich mathematical structures. This success paves the way towards a generative model that creates formulas fulfilling specified mathematical properties, accelerating the rate of discovery of useful formulas., Comment: 8 figures, 5 tables, 28 pages including the supplementary information. For a 5-minute video abstract see https://recorder-v3.slideslive.com/#/share?share=97010&s=c47967e3-d585-453c-a4dd-a4fa7955dba3 . Code can be found at https://github.com/RamanujanMachine/Blind-Delta-Algorithm

Compact laboratory-scale X-ray sources still rely on the same fundamental principles as in the first X-ray tubes developed more than a century ago. In recent years, significant research and development have focused on large-scale X-ray sources such as synchrotrons and free-electron lasers, leading to the generation of high-brightness coherent X-rays. However, the large size and high costs of such sources prevent their widespread use. The quest for a compact and coherent Xray source has long been a critical objective in modern physics, gaining further importance in recent years for industrial applications and fundamental scientific research. Here, we review the physical mechanisms governing compact coherent X-ray generation. Of current interest are coherent periodic interactions of free electrons in crystalline materials, creating hard X-rays via a mechanism known as parametric X-ray radiation (PXR). Over the past decade, X-ray sources leveraging this mechanism have demonstrated state-of-the-art tunability, directionality, and broad spatial coherence, enabling X-ray phase-contrast imaging on a compact scale. The coming years are expected to show substantial miniaturization of compact X-ray sources, facilitated by progress in electron beam technologies. This review compares the most promising mechanisms used for hard-X-ray generation, contrasting parametric X-ray radiation with inverse Compton scattering and characteristic radiation from a liquid-jet anode. We cover the most recent advancements, including the development of new materials, innovative geometrical designs, and specialized optimization techniques, aiming toward X-ray flux levels suitable for medical imaging and X-ray spectroscopy in compact scales.

The collapse of social contexts has been amplified by digital infrastructures but surprisingly received insufficient attention from Web privacy scholars. Users are persistently identified within and across distinct web contexts, in varying degrees, through and by different websites and trackers, losing the ability to maintain a fragmented identity. To systematically evaluate this structural privacy harm we operationalize the theory of Privacy as Contextual Integrity and measure persistent user identification within and between distinct Web contexts. We crawl the top-700 popular websites across the contexts of health, finance, news & media, LGBTQ, eCommerce, adult, and education websites, for 27 days, to learn how persistent browser identification via third-party cookies and JavaScript fingerprinting is diffused within and between web contexts. Past work measured Web tracking in bulk, highlighting the volume of trackers and tracking techniques. These measurements miss a crucial privacy implication of Web tracking - the collapse of online contexts. Our findings reveal how persistent browser identification varies between and within contexts, diffusing user IDs to different distances, contrasting known tracking distributions across websites, and conducted as a joint or separate effort via cookie IDs and JS fingerprinting. Our network analysis can inform the construction of browser storage containers to protect users against real-time context collapse. This is a first modest step in measuring Web privacy as contextual integrity, opening new avenues for contextual Web privacy research., Comment: 19 pages, 12 figures

Vision-language models (VLMs) excel at extracting and reasoning about information from images. Yet, their capacity to leverage internal knowledge about specific entities remains underexplored. This work investigates the disparity in model performance when answering factual questions about an entity described in text versus depicted in an image. Our results reveal a significant accuracy drop --averaging 19%-- when the entity is presented visually instead of textually. We hypothesize that this decline arises from limitations in how information flows from image tokens to query tokens. We use mechanistic interpretability tools to reveal that, although image tokens are preprocessed by the vision encoder, meaningful information flow from these tokens occurs only in the much deeper layers. Furthermore, critical image processing happens in the language model's middle layers, allowing few layers for consecutive reasoning, highlighting a potential inefficiency in how the model utilizes its layers for reasoning. These insights shed light on the internal mechanics of VLMs and offer pathways for enhancing their reasoning capabilities.

Far from equilibrium, universal dynamics prevails in many different situations, from pattern coarsening to turbulence, while the understanding of its microscopic laws and classification has remained unsatisfactory. Here, universal scaling evolution reflected by coarsening in a multi-component Bose gas is traced back to a low-energy effective theory given by a double sine-Gordon (DSG) model for a single real scalar field. Our experimental observations of a rubidium spinor BEC support the applicability of this model. Evaluating the scaling evolution according to the low-energy effective model demonstrates the universal aspects of the scaling characteristics as compared with that of the spinor gas. The difference between diffusion-type and sub-diffusive scaling is shown to be connected to the occupation of minima within the DSG potential. Our results point to a path towards a microscopic description and classification of pattern coarsening in many-body systems., Comment: 16pp., 8 figs., minor corrections

Event relation detection is a fundamental NLP task, leveraged in many downstream applications, whose modeling requires datasets annotated with event relations of various types. However, systematic and complete annotation of these relations is costly and challenging, due to the quadratic number of event pairs that need to be considered. Consequently, many current event relation datasets lack systematicity and completeness. In response, we introduce \textit{EventFull}, the first tool that supports consistent, complete and efficient annotation of temporal, causal and coreference relations via a unified and synergetic process. A pilot study demonstrates that EventFull accelerates and simplifies the annotation process while yielding high inter-annotator agreement.

Fundamental mathematical constants appear in nearly every field of science, from physics to biology. Formulas that connect different constants often bring great insight by hinting at connections between previously disparate fields. Discoveries of such relations, however, have remained scarce events, relying on sporadic strokes of creativity by human mathematicians. Recent developments of algorithms for automated conjecture generation have accelerated the discovery of formulas for specific constants. Yet, the discovery of connections between constants has not been addressed. In this paper, we present the first library dedicated to mathematical constants and their interrelations. This library can serve as a central repository of knowledge for scientists from different areas, and as a collaborative platform for development of new algorithms. The library is based on a new representation that we propose for organizing the formulas of mathematical constants: a hypergraph, with each node representing a constant and each edge representing a formula. Using this representation, we propose and demonstrate a systematic approach for automatically enriching this library using PSLQ, an integer relation algorithm based on QR decomposition and lattice construction. During its development and testing, our strategy led to the discovery of 75 previously unknown connections between constants, including a new formula for the `first continued fraction' constant $C_1$, novel formulas for natural logarithms, and new formulas connecting $\pi$ and $e$. The latter formulas generalize a century-old relation between $\pi$ and $e$ by Ramanujan, which until now was considered a singular formula and is now found to be part of a broader mathematical structure. The code supporting this library is a public, open-source API that can serve researchers in experimental mathematics and other fields of science., Comment: 20 pages, 7 figures

We present analytical results for the distribution of first return (FR) times of non-backtracking random walks (NBWs) on undirected configuration model networks consisting of $N$ nodes with degree distribution $P(k)$. We focus on the case in which the network consists of a single connected component. Starting from a random initial node $i$ at time $t=0$, an NBW hops into a random neighbor of $i$ at time $t=1$ and at each subsequent step it continues to hop into a random neighbor of its current node, excluding the previous node. We calculate the tail distribution $P ( T_{\rm FR} > t )$ of first return times from a random node to itself. It is found that $P ( T_{\rm FR} > t )$ is given by a discrete Laplace transform of the degree distribution $P(k)$. This result exemplifies the relation between structural properties of a network, captured by the degree distribution, and properties of dynamical processes taking place on the network. Using the tail-sum formula, we calculate the mean first return time ${\mathbb E}[ T_{\rm FR} ]$. Surprisingly, ${\mathbb E}[ T_{\rm FR} ]$ coincides with the result obtained from the Kac's lemma that applies to classical random walks (RWs). We also calculate the variance ${\rm Var}(T_{\rm FR})$, which accounts for the variability of first return times between different NBW trajectories. We apply this formalism to random regular graphs, Erdos-R\'enyi networks and configuration model networks with exponential and power-law degree distributions and obtain closed-form expressions for $P ( T_{\rm FR} > t )$ as well as its mean and variance. These results provide useful insight on the advantages of NBWs over classical RWs in network exploration, sampling and search processes., Comment: 28 pages, 9 figures

How to properly conduct human evaluations for text summarization is a longstanding challenge. The Pyramid human evaluation protocol, which assesses content selection by breaking the reference summary into sub-units and verifying their presence in the system summary, has been widely adopted. However, it suffers from a lack of systematicity in the definition and granularity of the sub-units. We address these problems by proposing QAPyramid, which decomposes each reference summary into finer-grained question-answer (QA) pairs according to the QA-SRL framework. We collect QA-SRL annotations for reference summaries from CNN/DM and evaluate 10 summarization systems, resulting in 8.9K QA-level annotations. We show that, compared to Pyramid, QAPyramid provides more systematic and fine-grained content selection evaluation while maintaining high inter-annotator agreement without needing expert annotations. Furthermore, we propose metrics that automate the evaluation pipeline and achieve higher correlations with QAPyramid than other widely adopted metrics, allowing future work to accurately and efficiently benchmark summarization systems., Comment: The first two authors contributed equally. Code: https://github.com/ZhangShiyue/QAPyramid

\emph{Batch normalization} is a successful building block of neural network architectures. Yet, it is not well understood. A neural network layer with batch normalization comprises three components that affect the representation induced by the network: \emph{recentering} the mean of the representation to zero, \emph{rescaling} the variance of the representation to one, and finally applying a \emph{non-linearity}. Our work follows the work of Hadi Daneshmand, Amir Joudaki, Francis Bach [NeurIPS~'21], which studied deep \emph{linear} neural networks with only the rescaling stage between layers at initialization. In our work, we present an analysis of the other two key components of networks with batch normalization, namely, the recentering and the non-linearity. When these two components are present, we observe a curious behavior at initialization. Through the layers, the representation of the batch converges to a single cluster except for an odd data point that breaks far away from the cluster in an orthogonal direction. We shed light on this behavior from two perspectives: (1) we analyze the geometrical evolution of a simplified indicative model; (2) we prove a stability result for the aforementioned~configuration.

Artificial Expert Intelligence (AEI) seeks to transcend the limitations of both Artificial General Intelligence (AGI) and narrow AI by integrating domain-specific expertise with critical, precise reasoning capabilities akin to those of top human experts. Existing AI systems often excel at predefined tasks but struggle with adaptability and precision in novel problem-solving. To overcome this, AEI introduces a framework for ``Probably Approximately Correct (PAC) Reasoning". This paradigm provides robust theoretical guarantees for reliably decomposing complex problems, with a practical mechanism for controlling reasoning precision. In reference to the division of human thought into System 1 for intuitive thinking and System 2 for reflective reasoning~\citep{tversky1974judgment}, we refer to this new type of reasoning as System 3 for precise reasoning, inspired by the rigor of the scientific method. AEI thus establishes a foundation for error-bounded, inference-time learning.

We introduce a diffusion model for Gaussian Splats, SplatDiffusion, to enable generation of three-dimensional structures from single images, addressing the ill-posed nature of lifting 2D inputs to 3D. Existing methods rely on deterministic, feed-forward predictions, which limit their ability to handle the inherent ambiguity of 3D inference from 2D data. Diffusion models have recently shown promise as powerful generative models for 3D data, including Gaussian splats; however, standard diffusion frameworks typically require the target signal and denoised signal to be in the same modality, which is challenging given the scarcity of 3D data. To overcome this, we propose a novel training strategy that decouples the denoised modality from the supervision modality. By using a deterministic model as a noisy teacher to create the noised signal and transitioning from single-step to multi-step denoising supervised by an image rendering loss, our approach significantly enhances performance compared to the deterministic teacher. Additionally, our method is flexible, as it can learn from various 3D Gaussian Splat (3DGS) teachers with minimal adaptation; we demonstrate this by surpassing the performance of two different deterministic models as teachers, highlighting the potential generalizability of our framework. Our approach further incorporates a guidance mechanism to aggregate information from multiple views, enhancing reconstruction quality when more than one view is available. Experimental results on object-level and scene-level datasets demonstrate the effectiveness of our framework.

Rapid progress in precision nanofabrication and atomic design over the past 50 years has ushered in a succession of transformative eras for molding the generation and flow of light. The use of nanoscale and atomic features to design light sources and optical elements-encapsulated by the term nanophotonics-has led to new fundamental science and innovative technologies across the entire electromagnetic spectrum, with substantial emphasis on the microwave to visible regimes. In this review, we pay special attention to the impact and potential of nanophotonics in a relatively exotic yet technologically disruptive regime: high-energy particles such as X-ray photons and free electrons-where nanostructures and atomic design open the doors to unprecedented technologies in quantum science and versatile X-ray sources and optics. As the practical generation of X-rays is intrinsically linked to the existence of energetic free or quasi-free-electrons, our review will also capture related phenomena and technologies that combine free electrons with nanophotonics, including free-electron-driven nanophotonics at other photon energies. In particular, we delve into the demonstration and study of quantum recoil in the X-ray regime, the study of nanomaterial design and free-electron wave shaping as means to enhance and control X-ray radiation, examine the free-electron generation enabled by nanophotonics, and analyze the high-harmonic generation by quasi-free electrons. We also discuss applications of quantum nanophotonics for X-rays and free electrons, including nanostructure waveguides for X-rays, photon pair enhanced X-ray imaging, mirrors, and lenses for X-rays, among others.

Drug discovery typically consists of multiple steps, including identifying a target protein key to a disease's etiology, validating that interacting with this target could prevent symptoms or cure the disease, discovering a small molecule or biologic therapeutic to interact with it, and optimizing the candidate molecule through a complex landscape of required properties. Drug discovery related tasks often involve prediction and generation while considering multiple entities that potentially interact, which poses a challenge for typical AI models. For this purpose we present MAMMAL - Molecular Aligned Multi-Modal Architecture and Language - a method that we applied to create a versatile multi-task multi-align foundation model that learns from large-scale biological datasets (2 billion samples) across diverse modalities, including proteins, small molecules, and genes. We introduce a prompt syntax that supports a wide range of classification, regression, and generation tasks. It allows combining different modalities and entity types as inputs and/or outputs. Our model handles combinations of tokens and scalars and enables the generation of small molecules and proteins, property prediction, and transcriptomic lab test predictions. We evaluated the model on 11 diverse downstream tasks spanning different steps within a typical drug discovery pipeline, where it reaches new SOTA in 9 tasks and is comparable to SOTA in 2 tasks. This performance is achieved while using a unified architecture serving all tasks, in contrast to the original SOTA performance achieved using tailored architectures. The model code and pretrained weights are publicly available at https://github.com/BiomedSciAI/biomed-multi-alignment and https://huggingface.co/ibm/biomed.omics.bl.sm.ma-ted-458m.

Significant variations of delays among connecting neurons cause an inevitable disadvantage of asynchronous brain dynamics compared to synchronous deep learning. However, this study demonstrates that this disadvantage can be converted into a computational advantage using a network with a single output and M multiple delays between successive layers, thereby generating a polynomial time-series outputs with M. The proposed role of delay in brain dynamics (RoDiB) model, is capable of learning increasing number of classified labels using a fixed architecture, and overcomes the inflexibility of the brain to update the learning architecture using additional neurons and connections. Moreover, the achievable accuracies of the RoDiB system are comparable with those of its counterpart tunable single delay architectures with M outputs. Further, the accuracies are significantly enhanced when the number of output labels exceeds its fully connected input size. The results are mainly obtained using simulations of VGG-6 on CIFAR datasets and also include multiple label inputs. However, currently only a small fraction of the abundant number of RoDiB outputs is utilized, thereby suggesting its potential for advanced computational power yet to be discovered., Comment: 18 pages, 3 figures, 2 tables

Recent efforts in fine-tuning language models often rely on automatic data selection, commonly using Nearest Neighbors retrieval from large datasets. However, we theoretically show that this approach tends to select redundant data, limiting its effectiveness or even hurting performance. To address this, we introduce SIFT, a data selection algorithm designed to reduce uncertainty about the model's response given a prompt, which unifies ideas from retrieval and active learning. Whereas Nearest Neighbor retrieval typically fails in the presence of information duplication, SIFT accounts for information duplication and optimizes the overall information gain of the selected examples. We focus our evaluations on fine-tuning at test-time for prompt-specific language modeling on the Pile dataset, and show that SIFT consistently outperforms Nearest Neighbor retrieval, with minimal computational overhead. Moreover, we show that our uncertainty estimates can predict the performance gain of test-time fine-tuning, and use this to develop an adaptive algorithm that invests test-time compute proportional to realized performance gains. We provide the $\texttt{activeft}$ (Active Fine-Tuning) library which can be used as a drop-in replacement for Nearest Neighbor retrieval.

Recent advancements in Web agents have introduced novel architectures and benchmarks showcasing progress in autonomous web navigation and interaction. However, most existing benchmarks prioritize effectiveness and accuracy, overlooking factors like safety and trustworthiness which are essential for deploying web agents in enterprise settings. We present STWebAgentBench, a benchmark designed to evaluate web agents safety and trustworthiness across six critical dimensions, essential for reliability in enterprise applications. This benchmark is grounded in a detailed framework that defines safe and trustworthy (ST) agent behavior. Our work extends WebArena with safety templates and evaluation functions to assess safety policy compliance rigorously. We introduce the Completion Under Policy to measure task success while adhering to policies, alongside the Risk Ratio, which quantifies policy violations across dimensions, providing actionable insights to address safety gaps. Our evaluation reveals that current SOTA agents struggle with policy adherence and cannot yet be relied upon for critical business applications. We open-source this benchmark and invite the community to contribute, with the goal of fostering a new generation of safer, more trustworthy AI agents. All code, data, environment reproduction resources, and video demonstrations are available at https://sites.google.com/view/st-webagentbench/home.

This paper explores the characterization of equivariant linear layers for representations of permutations and related groups. Unlike traditional approaches, which address these problems using parameter-sharing, we consider an alternative methodology based on irreducible representations and Schur's lemma. Using this methodology, we obtain an alternative derivation for existing models like DeepSets, 2-IGN graph equivariant networks, and Deep Weight Space (DWS) networks. The derivation for DWS networks is significantly simpler than that of previous results. Next, we extend our approach to unaligned symmetric sets, where equivariance to the wreath product of groups is required. Previous works have addressed this problem in a rather restrictive setting, in which almost all wreath equivariant layers are Siamese. In contrast, we give a full characterization of layers in this case and show that there is a vast number of additional non-Siamese layers in some settings. We also show empirically that these additional non-Siamese layers can improve performance in tasks like graph anomaly detection, weight space alignment, and learning Wasserstein distances. Our code is available at \href{https://github.com/yonatansverdlov/Irreducible-Representations-of-Deep-Weight-Spaces}{GitHub}.

There has been an increasing interest in detecting hallucinations in model-generated texts, both manually and automatically, at varying levels of granularity. However, most existing methods fail to precisely pinpoint the errors. In this work, we introduce QASemConsistency, a new formalism for localizing factual inconsistencies in attributable text generation, at a fine-grained level. Drawing inspiration from Neo-Davidsonian formal semantics, we propose decomposing the generated text into minimal predicate-argument level propositions, expressed as simple question-answer (QA) pairs, and assess whether each individual QA pair is supported by a trusted reference text. As each QA pair corresponds to a single semantic relation between a predicate and an argument, QASemConsistency effectively localizes the unsupported information. We first demonstrate the effectiveness of the QASemConsistency methodology for human annotation, by collecting crowdsourced annotations of granular consistency errors, while achieving a substantial inter-annotator agreement ($\kappa > 0.7)$. Then, we implement several methods for automatically detecting localized factual inconsistencies, with both supervised entailment models and open-source LLMs.

We prove a product theorem for sublinear bilipschitz equivalences which generalizes the classical work of Kapovich, Kleiner and Leeb on quasiisometries between product spaces. Independently we develop a new tool, based on a theorem of Cornulier on sublinear bilipschitz equivalences between solvable Lie groups, to evaluate the distortion of certain subgroups in central extensions. This is useful to provide lower bounds on Dehn functions. Building on work of Cornulier and Tessera, we compute the Dehn functions of all simply connected solvable Lie groups of exponential growth up to dimension $5$. We employ our product theorem to distinguish up to quasiisometry certain families among these groups which share the same dimension, cone-dimension and Dehn function. Finally, using a theorem of Peng together with our computations of cone-dimensions and Dehn functions, we establish the quasiisometric rigidity of the rank-two, five-dimensional solvable Lie group acting simply transitively by isometries on the horosphere in general position inside the product of three real hyperbolic planes.

Diacritical marks in the Hebrew language give words their vocalized form. The task of adding diacritical marks to plain Hebrew text is still dominated by a system that relies heavily on human-curated resources. Recent models trained on diacritized Hebrew texts still present a gap in performance. We use a recently developed char-based PLM to narrowly bridge this gap. Presenting MenakBERT, a character level transformer pretrained on Hebrew text and fine-tuned to produce diacritical marks for Hebrew sentences. We continue to show how finetuning a model for diacritizing transfers to a task such as part of speech tagging., Comment: Published at ISCOL2022 as a poster

We study which machine learning algorithms have tight generalization bounds. First, we present conditions that preclude the existence of tight generalization bounds. Specifically, we show that algorithms that have certain inductive biases that cause them to be unstable do not admit tight generalization bounds. Next, we show that algorithms that are sufficiently stable do have tight generalization bounds. We conclude with a simple characterization that relates the existence of tight generalization bounds to the conditional variance of the algorithm's loss.

Semiconductor crosshatch patterns in thin film heterostructures form as a result of strain relaxation processes and dislocation pile-ups during growth of lattice mismatched materials. Due to their connection with the internal misfit dislocation network, these crosshatch patterns are a complex fingerprint of internal strain relaxation and growth anisotropy. Therefore, this mesoscopic fingerprint not only describes the residual strain state of a near-surface quantum well, but also could provide an indicator of the quality of electron transport through the material. Here, we present a method utilizing computer vision and machine learning to analyze AFM crosshatch patterns that exhibits this correlation. Our analysis reveals optimized electron transport for moderate values of $\lambda$ (crosshatch wavelength) and $\epsilon$ (crosshatch height), roughly 1 $\mu$m and 4 nm, respectively, that define the average waveform of the pattern. Simulated 2D AFM crosshatch patterns are used to train a machine learning model to correlate the crosshatch patterns to dislocation density. Furthermore, this model is used to evaluate the experimental AFM images and predict a dislocation density based on the crosshatch waveform. Predicted dislocation density, experimental AFM crosshatch data, and experimental transport characterization are used to train a final model to predict 2D electron gas mean free path. This model shows electron scattering is strongly correlated with elastic effects (e.g. dislocation scattering) below 200 nm $\lambda_{MFP}$.

We scrutinize the reported lensing anomaly of the CMB by considering several phenomenological modifications of the lensing consistency parameter, $A_{\rm L}$. Considering Planck spectra alone, we find statisically significant evidence for scale dependence (`running') of $A_{\rm L}$. We then demonstrate that the anomaly is entirely driven by Planck's low multipoles, $\ell \leq 30$. When these data points are excluded a joint analysis with several other datasets clearly favors $\Lambda$CDM over the extended $\Lambda \rm CDM+A_L$ model. Not only that the lensing anomaly and low $\ell$ anomaly of the CMB go away in this case, but also the $S_8$ tension is ameliorated, and only the Hubble tension persists., Comment: 33 pages, 9 figures, and 8 tables

We study the conditions for finding an island in an anisotropic universe - Bianchi Type-I filled with radiation. We verify that the existence of islands does not depend on their shape. We then find that islands may form at certain times, near the turnaround point - where the universe turns from contraction to expansion in one of the directions. This is in line with previous analyses regarding cosmological space-times where islands form if one has two energy scales in the problem, such as the typical temperature of the universe and, on top of that, CC, curvature, anisotropy, or some mass scale., Comment: 19 Pages, 3 figures

Here, we report on a new class active particles capable of dynamically programmable motion powered by electricity. We have implemented physical principles that separate the propulsion and steering mechanisms of active motion using optically activated, patterned, photoresponsive semiconductor coatings on intricate microstructures. Our engineered microswimmer robots employ an induced-charge electro-phoresis (ICEP) mechanism to achieve linear motion and optically modulated electrokinetic propulsion (OMEP) for steering. Optical modulation is achieved by manipulating the polarizability of patterned ZnO semiconductor coating through exposure to light with wavelengths above its bandgap, exploiting the semiconductor's photoconductive properties. Unlike previous methods that rely on changing the direction of optical illumination or spatially controlling narrow optical beams, our approach achieves optical steering under uniform ambient illumination conditions, thereby greatly reducing the complexity of the optical system. The decoupling of propulsion and steering allows for the programming of micromotor trajectories in both open and closed-loop control modes. We anticipate that our findings will pave the way for efficient optically gated control of the trajectory of photoresponsive active particles. Furthermore, they will enable the selective manipulation of specific subgroups of engineered active microparticles with various semiconducting coatings having different band gaps.

In recent years, workplaces and educational institutes have widely adopted virtual meeting platforms. This has led to a growing interest in analyzing and extracting insights from these meetings, which requires effective detection and tracking of unique individuals. In practice, there is no standardization in video meetings recording layout, and how they are captured across the different platforms and services. This, in turn, creates a challenge in acquiring this data stream and analyzing it in a uniform fashion. Our approach provides a solution to the most general form of video recording, usually consisting of a grid of participants (\cref{fig:videomeeting}) from a single video source with no metadata on participant locations, while using the least amount of constraints and assumptions as to how the data was acquired. Conventional approaches often use YOLO models coupled with tracking algorithms, assuming linear motion trajectories akin to that observed in CCTV footage. However, such assumptions fall short in virtual meetings, where participant video feed window can abruptly change location across the grid. In an organic video meeting setting, participants frequently join and leave, leading to sudden, non-linear movements on the video grid. This disrupts optical flow-based tracking methods that depend on linear motion. Consequently, standard object detection and tracking methods might mistakenly assign multiple participants to the same tracker. In this paper, we introduce a novel approach to track and re-identify participants in remote video meetings, by utilizing the spatio-temporal priors arising from the data in our domain. This, in turn, increases tracking capabilities compared to the use of general object tracking. Our approach reduces the error rate by 95% on average compared to YOLO-based tracking methods as a baseline., Comment: Accepted to ECCV 2024 workshop

Bird's Eye View (BEV) map prediction is essential for downstream autonomous driving tasks like trajectory prediction. In the past, this was accomplished through the use of a sophisticated sensor configuration that captured a surround view from multiple cameras. However, in large-scale production, cost efficiency is an optimization goal, so that using fewer cameras becomes more relevant. But the consequence of fewer input images correlates with a performance drop. This raises the problem of developing a BEV perception model that provides a sufficient performance on a low-cost sensor setup. Although, primarily relevant for inference time on production cars, this cost restriction is less problematic on a test vehicle during training. Therefore, the objective of our approach is to reduce the aforementioned performance drop as much as possible using a modern multi-camera surround view model reduced for single-camera inference. The approach includes three features, a modern masking technique, a cyclic Learning Rate (LR) schedule, and a feature reconstruction loss for supervising the transition from six-camera inputs to one-camera input during training. Our method outperforms versions trained strictly with one camera or strictly with six-camera surround view for single-camera inference resulting in reduced hallucination and better quality of the BEV map., Comment: This Paper has been accepted to the 27th IEEE International Conference on Intelligent Transportation Systems (ITSC 2024)

The cores of stars are the cosmic furnaces where light elements are fused into heavier nuclei. The fusion of hydrogen to helium initially powers all stars. The ashes of the fusion reactions are then predicted to serve as fuel in a series of stages, eventually transforming massive stars into a structure of concentric shells. These are composed of natal hydrogen on the outside, and consecutively heavier compositions inside, predicted to be dominated by helium, carbon/oxygen, oxygen/neon/magnesium, and oxygen/silicon/sulphur. Silicon and sulphur are fused into inert iron, leading to the collapse of the core and either a supernova explosion or the direct formation of a black hole. Stripped stars, where the outer hydrogen layer has been removed and the internal He-rich layer (in Wolf-Rayet WN stars) or even the C/O layer below it (in Wolf-Rayet WC/WO stars) are exposed, provide evidence for this shell structure, and the cosmic element production mechanism it reflects. The types of supernova explosions that arise from stripped stars embedded in shells of circumstellar material (most notably Type Ibn supernovae from stars with outer He layers, and Type Icn supernovae from stars with outer C/O layers) confirm this scenario. However, direct evidence for the most interior shells, which are responsible for the production of elements heavier than oxygen, is lacking. Here, we report the discovery of the first-of-its-kind supernova arising from a star peculiarly stripped all the way to the silicon and sulphur-rich internal layer. Whereas the concentric shell structure of massive stars is not under debate, it is the first time that such a thick, massive silicon and sulphur-rich shell, expelled by the progenitor shortly before the SN explosion, has been directly revealed., Comment: 48 pages, 12 figures and 10 tables. Submitted to a high-impact journal. The reduced spectra and photometry will be made available via the journal webpage and the WISeREP archive after the acceptance of the paper

General web-based agents are increasingly essential for interacting with complex web environments, yet their performance in real-world web applications remains poor, yielding extremely low accuracy even with state-of-the-art frontier models. We observe that these agents can be decomposed into two primary components: Planning and Grounding. Yet, most existing research treats these agents as black boxes, focusing on end-to-end evaluations which hinder meaningful improvements. We sharpen the distinction between the planning and grounding components and conduct a novel analysis by refining experiments on the Mind2Web dataset. Our work proposes a new benchmark for each of the components separately, identifying the bottlenecks and pain points that limit agent performance. Contrary to prevalent assumptions, our findings suggest that grounding is not a significant bottleneck and can be effectively addressed with current techniques. Instead, the primary challenge lies in the planning component, which is the main source of performance degradation. Through this analysis, we offer new insights and demonstrate practical suggestions for improving the capabilities of web agents, paving the way for more reliable agents.

In multitask learning, conflicts between task gradients are a frequent issue degrading a model's training performance. This is commonly addressed by using the Gradient Projection algorithm PCGrad that often leads to faster convergence and improved performance metrics. In this work, we present a method to adapt this algorithm to simultaneously also perform task prioritization. Our approach differs from traditional task weighting performed by scaling task losses in that our weighting scheme applies only in cases where tasks are in conflict, but lets the training proceed unhindered otherwise. We replace task weighting factors by a probability distribution that determines which task gradients get projected in conflict cases. Our experiments on the nuScenes, CIFAR-100, and CelebA datasets confirm that our approach is a practical method for task weighting. Paired with multiple different task weighting schemes, we observe a significant improvement in the performance metrics of most tasks compared to Gradient Projection with uniform projection probabilities.

We present the updated design of HighSpec, a high-resolution $\mathcal{R} \sim 20,000$ spectrograph designed for the Multi Aperture Spectroscopic Telescope (MAST). HighSpec offers three observing modes centered at the Ca II H&K, Mg b triplet, and H$\alpha$ lines. Each mode is supported by a highly optimized ion-etched grating, contributing to an exceptional instrument peak efficiency of $\gtrsim85\%$ for the two latter bands ($\gtrsim55\%$ for the Ca II H&K band). Optimizing throughput over wavelength coverage ($\Delta \lambda=10-17$ nm), HighSpec enables the precise measurement of spectral lines from faint targets. This approach is especially relevant for stellar object studies, specifically of WDs, which are intrinsically faint and have few spectroscopic lines. Each observing mode was tailored to target spectral features essential for WD research. Its integration with MAST, an array of 20 custom-designed telescopes that can function as a single large telescope (equivalent to a $2.7$ m telescope in collecting area) or multiplexing over the entire sky, provides unique adaptability for extensive and effective spectroscopic campaigns. Currently in its final assembly and testing stages, HighSpec's on-sky commissioning is scheduled for 2025., Comment: SPIE Conference "Space Telescopes and Instrumentation 2024: Ground-based and Airborne Instrumentation for Astronomy X", Yokahama, Japan

In this paper, we investigate a variant of the classical stochastic Multi-armed Bandit (MAB) problem, where the payoff received by an agent (either cost or reward) is both delayed, and directly corresponds to the magnitude of the delay. This setting models faithfully many real world scenarios such as the time it takes for a data packet to traverse a network given a choice of route (where delay serves as the agent's cost); or a user's time spent on a web page given a choice of content (where delay serves as the agent's reward). Our main contributions are tight upper and lower bounds for both the cost and reward settings. For the case that delays serve as costs, which we are the first to consider, we prove optimal regret that scales as $\sum_{i:\Delta_i > 0}\frac{\log T}{\Delta_i} + d^*$, where $T$ is the maximal number of steps, $\Delta_i$ are the sub-optimality gaps and $d^*$ is the minimal expected delay amongst arms. For the case that delays serves as rewards, we show optimal regret of $\sum_{i:\Delta_i > 0}\frac{\log T}{\Delta_i} + \bar{d}$, where $\bar d$ is the second maximal expected delay. These improve over the regret in the general delay-dependent payoff setting, which scales as $\sum_{i:\Delta_i > 0}\frac{\log T}{\Delta_i} + D$, where $D$ is the maximum possible delay. Our regret bounds highlight the difference between the cost and reward scenarios, showing that the improvement in the cost scenario is more significant than for the reward. Finally, we accompany our theoretical results with an empirical evaluation.

We present Jamba-1.5, new instruction-tuned large language models based on our Jamba architecture. Jamba is a hybrid Transformer-Mamba mixture of experts architecture, providing high throughput and low memory usage across context lengths, while retaining the same or better quality as Transformer models. We release two model sizes: Jamba-1.5-Large, with 94B active parameters, and Jamba-1.5-Mini, with 12B active parameters. Both models are fine-tuned for a variety of conversational and instruction-following capabilties, and have an effective context length of 256K tokens, the largest amongst open-weight models. To support cost-effective inference, we introduce ExpertsInt8, a novel quantization technique that allows fitting Jamba-1.5-Large on a machine with 8 80GB GPUs when processing 256K-token contexts without loss of quality. When evaluated on a battery of academic and chatbot benchmarks, Jamba-1.5 models achieve excellent results while providing high throughput and outperforming other open-weight models on long-context benchmarks. The model weights for both sizes are publicly available under the Jamba Open Model License and we release ExpertsInt8 as open source., Comment: Webpage: https://www.ai21.com/jamba

Out-of-distribution (OOD) object detection is a critical task focused on detecting objects that originate from a data distribution different from that of the training data. In this study, we investigate to what extent state-of-the-art open-vocabulary object detectors can detect unusual objects in street scenes, which are considered as OOD or rare scenarios with respect to common street scene datasets. Specifically, we evaluate their performance on the OoDIS Benchmark, which extends RoadAnomaly21 and RoadObstacle21 from SegmentMeIfYouCan, as well as LostAndFound, which was recently extended to object level annotations. The objective of our study is to uncover short-comings of contemporary object detectors in challenging real-world, and particularly in open-world scenarios. Our experiments reveal that open vocabulary models are promising for OOD object detection scenarios, however far from perfect. Substantial improvements are required before they can be reliably deployed in real-world applications. We benchmark four state-of-the-art open-vocabulary object detection models on three different datasets. Noteworthily, Grounding DINO achieves the best results on RoadObstacle21 and LostAndFound in our study with an AP of 48.3% and 25.4% respectively. YOLO-World excels on RoadAnomaly21 with an AP of 21.2%.

Quantum transport is often characterized not just by mean observables like the particle or energy current, but by their fluctuations and higher moments, which can act as detailed probes of the physical mechanisms at play. However, relatively few theoretical methods are able to access the full counting statistics (FCS) of transport processes through electronic junctions in strongly correlated regimes. While most experiments are concerned with the steady state properties, most accurate theoretical methods rely on computationally expensive propagation from a tractable initial state. Here, we propose a simple approach for computing the FCS through a junction directly at the steady state, utilizing the propagator noncrossing approximation (NCA). Compared to time propagation, our method offers reduced computational cost at the same level of approximation; but the idea can also be used within other approximations or as a basis for numerically exact techniques. We demonstrate the method's capabilities by investigating the impact of lead dimensionality on electronic transport in the nonequilibrium Anderson impurity model at the onset of Kondo physics. Our results reveal a distinct signature of one dimensional leads in the noise and Fano factor not present for other dimensionalities, showing the potential of FCS measurements as a probe of the environment surrounding a quantum dot.

By coupling a semiconductor-based planar Josephson junction to a superconducting resonator, we investigate the Andreev bound states in the junction using dispersive readout techniques. Using electrostatic gating to create a narrow constriction in the junction, our measurements unveil a strong coupling interaction between the resonator and the Andreev bound states. This enables the mapping of isolated tunable Andreev bound states, with an observed transparency of up to 99.94\% along with an average induced superconducting gap of $\sim 150 \mu$eV. Exploring the gate parameter space further elucidates a non-monotonic evolution of multiple Andreev bound states with varying gate voltage. Complimentary tight-binding calculations of an Al-InAs planar Josephson junction with strong Rashba spin-orbit coupling provide insight into possible mechanisms responsible for such behavior. Our findings highlight the subtleties of the Andreev spectrum of Josephson junctions fabricated on superconductor-semiconductor heterostructures and offering potential applications in probing topological states in these hybrid platforms.

Detecting semantic arguments of a predicate word has been conventionally modeled as a sentence-level task. The typical reader, however, perfectly interprets predicate-argument relations in a much wider context than just the sentence where the predicate was evoked. In this work, we reformulate the problem of argument detection through textual entailment to capture semantic relations across sentence boundaries. We propose a method that tests whether some semantic relation can be inferred from a full passage by first encoding it into a simple and standalone proposition and then testing for entailment against the passage. Our method does not require direct supervision, which is generally absent due to dataset scarcity, but instead builds on existing NLI and sentence-level SRL resources. Such a method can potentially explicate pragmatically understood relations into a set of explicit sentences. We demonstrate it on a recent document-level benchmark, outperforming some supervised methods and contemporary language models., Comment: 9 pages, ACL 2024

Justin Moore's weak club-guessing principle $\mho$ admits various possible generalizations to the second uncountable cardinal. One of them was shown to hold in ZFC by Shelah. A stronger one was shown to follow from several consequences of the continuum hypothesis by Inamdar and Rinot. Here we prove that the stronger one may consistently fail. Specifically, starting with a supercompact cardinal and an inaccessible cardinal above it, we devise a notion of forcing consisting of finite working parts and finitely many two types of models as side conditions, to violate this analog of $\mho$ at the second uncountable cardinal.

Symmetries play a pivotal role in our understanding of the properties of quantum many-body systems. While there are theorems and a well-established toolbox for systems in thermal equilibrium, much less is known about the role of symmetries and their connection to dynamics out of equilibrium. This arises due to the direct link between a system's thermal state and its Hamiltonian, which is generally not the case for nonequilibrium dynamics. Here we present a pathway to identify the effective symmetries and to extract them from data in nonequilibrium quantum many-body systems. Our approach is based on exact relations between correlation functions involving different numbers of spatial points, which can be viewed as nonequilibrium versions of (equal-time) Ward identities encoding the symmetries of the system. We derive symmetry witnesses, which are particularly suitable for the analysis of measured or simulated data at different snapshots in time. To demonstrate the potential of the approach, we apply our method to numerical and experimental data for a spinor Bose gas. We investigate the important question of a dynamical restoration of an explicitly broken symmetry of the Hamiltonian by the initial state. Remarkably, it is found that effective symmetry restoration can occur long before the system equilibrates. We also use the approach to define and identify spontaneous symmetry breaking far from equilibrium, which is of great relevance for applications to nonequilibrium phase transitions. Our work opens new avenues for the classification and analysis of quantum as well as classical many-body dynamics in a large variety of systems, ranging from ultracold quantum gases to cosmology.

Active fluids display spontaneous turbulent-like flows known as active turbulence. Recent work revealed that these flows have universal features, independent of the material properties and of the presence of topological defects. However, the differences between defect-laden and defect-free active turbulence remain largely unexplored. Here, by means of large-scale numerical simulations, we show that defect-free active nematic turbulence can undergo dynamical arrest. We find that flow alignment -- the tendency of nematics to reorient under shear -- enhances large-scale jets in contractile rodlike systems while promoting arrested flow patterns in extensile systems. Our results reveal a mechanism of labyrinthine pattern formation produced by an emergent topology of nematic domain walls that partially suppresses chaotic flows. Taken together, our findings call for the experimental realization of defect-free active nematics, and suggest that topological defects enable turbulence by preventing dynamical arrest.

Unlike human intestines, which are long, hollow tubes, the intestines of sharks and rays contain interior helical structures surrounding a cylindrical hole. One function of these structures may be to create asymmetric flow, favoring passage of fluid down the digestive tract, from anterior to posterior. Here, we design and 3D print biomimetic models of shark intestines, in both rigid and deformable materials. We use the rigid models to test which physical parameters of the interior helices (the pitch, the hole radius, the tilt angle, and the number of turns) yield the largest flow asymmetries. These asymmetries exceed those of traditional Tesla valves, structures specifically designed to create flow asymmetry without any moving parts. When we print the biomimetic models in elastomeric materials so that flow can couple to the structure's shape, flow asymmetry is significantly amplified; it is 7-fold larger in deformable structures than in rigid structures. Last, we 3D-print deformable versions of the intestine of a dogfish shark, based on a tomogram of a biological sample. This biomimic produces flow asymmetry comparable to traditional Tesla valves. The ability to influence the direction of a flow through a structure has applications in biological tissues and artificial devices across many scales, from large industrial pipelines to small microfluidic devices., Comment: 7 pages, 4 figures, SI (10 S. figures)

AI Infrastructure plays a key role in the speed and cost-competitiveness of developing and deploying advanced AI models. The current demand for powerful AI infrastructure for model training is driven by the emergence of generative AI and foundational models, where on occasion thousands of GPUs must cooperate on a single training job for the model to be trained in a reasonable time. Delivering efficient and high-performing AI training requires an end-to-end solution that combines hardware, software and holistic telemetry to cater for multiple types of AI workloads. In this report, we describe IBM's hybrid cloud infrastructure that powers our generative AI model development. This infrastructure includes (1) Vela: an AI-optimized supercomputing capability directly integrated into the IBM Cloud, delivering scalable, dynamic, multi-tenant and geographically distributed infrastructure for large-scale model training and other AI workflow steps and (2) Blue Vela: a large-scale, purpose-built, on-premises hosting environment that is optimized to support our largest and most ambitious AI model training tasks. Vela provides IBM with the dual benefit of high performance for internal use along with the flexibility to adapt to an evolving commercial landscape. Blue Vela provides us with the benefits of rapid development of our largest and most ambitious models, as well as future-proofing against the evolving model landscape in the industry. Taken together, they provide IBM with the ability to rapidly innovate in the development of both AI models and commercial offerings., Comment: Corresponding Authors: Talia Gershon, Seetharami Seelam,Brian Belgodere, Milton Bonilla

Improvements in language models' capabilities have pushed their applications towards longer contexts, making long-context evaluation and development an active research area. However, many disparate use-cases are grouped together under the umbrella term of "long-context", defined simply by the total length of the model's input, including - for example - Needle-in-a-Haystack tasks, book summarization, and information aggregation. Given their varied difficulty, in this position paper we argue that conflating different tasks by their context length is unproductive. As a community, we require a more precise vocabulary to understand what makes long-context tasks similar or different. We propose to unpack the taxonomy of long-context based on the properties that make them more difficult with longer contexts. We propose two orthogonal axes of difficulty: (I) Diffusion: How hard is it to find the necessary information in the context? (II) Scope: How much necessary information is there to find? We survey the literature on long-context, provide justification for this taxonomy as an informative descriptor, and situate the literature with respect to it. We conclude that the most difficult and interesting settings, whose necessary information is very long and highly diffused within the input, is severely under-explored. By using a descriptive vocabulary and discussing the relevant properties of difficulty in long-context, we can implement more informed research in this area. We call for a careful design of tasks and benchmarks with distinctly long context, taking into account the characteristics that make it qualitatively different from shorter context., Comment: EMNLP 2024

This study focuses on advancing metascintillators to break the 100 ps barrier and approach the 10 ps target. We exploit nanophotonic features, specifically the Purcell effect, to shape and enhance the scintillation properties of the first-generation metascintillator. We demonstrate that a faster emission is achievable along with a more efficient conversion efficiency. This results in a coincidence time resolution improved by a factor of 1.6, crucial for TOF-PET applications.