Search

We applied synchrotron single-crystal X-ray diffraction in a diamond anvil cell at 48-51 GPa and first-principles theoretical calculations to study the crystal structure of solid atomic iodine at high pressure. We report the synthesis of two phases of atomic iodine at 48-51 GPa via laser heating of I-N2 mixtures. Unlike the familiar monatomic I4/mmm structure, which consists of crystallographically equivalent atoms, a new Pm-3n structure is of inclusion type, featuring two distinct kinds of atoms: a central detached one and peripheral ones forming the linear chains. Moreover, we observe crystallization of the familiar high-pressure face centered cubic (fcc) structure, albeit at much lower pressures compared to cold compressed iodine. The discovery of Pm-3n structure in iodine marks an important step in understanding of the pressure induced phase transition sequence in halogens., Comment: 9 pages, 6 figures in the main text. 5 supplementary figures and 1 supplementary table

Quantization has established itself as the primary approach for decreasing the computational and storage expenses associated with Large Language Models (LLMs) inference. The majority of current research emphasizes quantizing weights and activations to enable low-bit general-matrix-multiply (GEMM) operations, with the remaining non-linear operations executed at higher precision. In our study, we discovered that following the application of these techniques, the primary bottleneck in LLMs inference lies in the softmax layer. The softmax operation comprises three phases: exponent calculation, accumulation, and normalization, Our work focuses on optimizing the first two phases. We propose an analytical approach to determine the optimal clipping value for the input to the softmax function, enabling sub-4-bit quantization for LLMs inference. This method accelerates the calculations of both $e^x$ and $\sum(e^x)$ with minimal to no accuracy degradation. For example, in LLaMA1-30B, we achieve baseline performance with 2-bit quantization on the well-known "Physical Interaction: Question Answering" (PIQA) dataset evaluation. This ultra-low bit quantization allows, for the first time, an acceleration of approximately 4x in the accumulation phase. The combination of accelerating both $e^x$ and $\sum(e^x)$ results in a 36.9% acceleration in the softmax operation.

Diffusion models have recently been successfully applied to a wide range of robotics applications for learning complex multi-modal behaviors from data. However, prior works have mostly been confined to single-robot and small-scale environments due to the high sample complexity of learning multi-robot diffusion models. In this paper, we propose a method for generating collision-free multi-robot trajectories that conform to underlying data distributions while using only single-robot data. Our algorithm, Multi-robot Multi-model planning Diffusion (MMD), does so by combining learned diffusion models with classical search-based techniques -- generating data-driven motions under collision constraints. Scaling further, we show how to compose multiple diffusion models to plan in large environments where a single diffusion model fails to generalize well. We demonstrate the effectiveness of our approach in planning for dozens of robots in a variety of simulated scenarios motivated by logistics environments. View video demonstrations in our supplementary material, and our code at: https://github.com/yoraish/mmd., Comment: The first three authors contributed equally to this work. Under review for ICLR 2025

We have developed a Bayesian optimization (BO) workflow that integrates intra-step noise optimization into automated experimental cycles. Traditional BO approaches in automated experiments focus on optimizing experimental trajectories but often overlook the impact of measurement noise on data quality and cost. Our proposed framework simultaneously optimizes both the target property and the associated measurement noise by introducing time as an additional input parameter, thereby balancing the signal-to-noise ratio and experimental duration. Two approaches are explored: a reward-driven noise optimization and a double-optimization acquisition function, both enhancing the efficiency of automated workflows by considering noise and cost within the optimization process. We validate our method through simulations and real-world experiments using Piezoresponse Force Microscopy (PFM), demonstrating the successful optimization of measurement duration and property exploration. Our approach offers a scalable solution for optimizing multiple variables in automated experimental workflows, improving data quality, and reducing resource expenditure in materials science and beyond., Comment: 22 pages, 9 figures

Photon entanglement is indispensable for optical quantum technologies. Measurement-based optical quantum computing and all-optical quantum networks rely on multiphoton cluster states consisting of indistinguishable entangled photons. A promising method for creating such cluster states on demand is spin-photon entanglement using the spin of a resident charge carrier in a quantum dot, precessing in a weak external magnetic field. In this work, we show theoretically and experimentally that spin-photon entanglement is strongly affected by the hidden anisotropy of quantum dots, which can arise from mechanical stress, shape anisotropy and even specific crystal structure. In the measurements of time-resolved photoluminescence and cross-polarized second-order photon correlation function in a magnetic field, the anisotropy manifests itself in the spin dynamics and, as a consequence, in the spin-photon concurrence. The measured time-filtered spin-photon Bell state fidelity depends strongly on the excitation polarization and reaches an extremely high value of 94% at maximum. We specify the magnetic field and excitation polarization directions that maximize spin-photon entanglement and thereby enhance the fidelity of multiphoton entangled states.

A cylindrical microwave cavity resonator operating in the TM$_{0,1,0}$ mode was excited using an interferometric configuration that allowed manipulation of the electric field and potential within the resonator by adjusting the phase and amplitude in the two interferometer arms driving the resonator. With precise tuning of the phase and amplitude, 25 dB suppression of the electric field at the resonance frequency was achieved while simultaneously resonantly enhancing the amplitude of the time-varying electric-scalar potential. Under these balanced conditions, the system demonstrated electromagnetically induced absorption in the cavity response due to the annulment of the electric field at the resonance frequency. This phenomena can be regarded as a form of extreme dispersion, and led to a measured increase in the cavity phase response by an order of magnitude. This work presents an experimental setup with the features that will allow the electric-scalar Aharonov-Bohm effect to be tested under conditions involving a time-varying electric-scalar potential, without the presence of an electric field or magnetic vector potential, an experiment that has not yet been realised.

Traditional multi-agent path finding (MAPF) methods try to compute entire start-goal paths which are collision free. However, computing an entire path can take too long for MAPF systems where agents need to replan fast. Methods that address this typically employ a "windowed" approach and only try to find collision free paths for a small windowed timestep horizon. This adaptation comes at the cost of incompleteness; all current windowed approaches can become stuck in deadlock or livelock. Our main contribution is to introduce our framework, WinC-MAPF, for Windowed MAPF that enables completeness. Our framework uses heuristic update insights from single-agent real-time heuristic search algorithms as well as agent independence ideas from MAPF algorithms. We also develop Single-Step CBS (SS-CBS), an instantiation of this framework using a novel modification to CBS. We show how SS-CBS, which only plans a single step and updates heuristics, can effectively solve tough scenarios where existing windowed approaches fail.

In this work, we present an AutoTM 2.0 framework for optimizing additively regularized topic models. Comparing to the previous version, this version includes such valuable improvements as novel optimization pipeline, LLM-based quality metrics and distributed mode. AutoTM 2.0 is a comfort tool for specialists as well as non-specialists to work with text documents to conduct exploratory data analysis or to perform clustering task on interpretable set of features. Quality evaluation is based on specially developed metrics such as coherence and gpt-4-based approaches. Researchers and practitioners can easily integrate new optimization algorithms and adapt novel metrics to enhance modeling quality and extend their experiments. We show that AutoTM 2.0 achieves better performance compared to the previous AutoTM by providing results on 5 datasets with different features and in two different languages.

We discuss the exclusive photoproduction of a heavy vector quarkonium, namely $J/\psi$ and $\Upsilon$ at 1-loop in $\alpha_s$. In collinear factorisation (CF), the amplitude for such a process is obtained by the convolution of a hard partonic sub-amplitude, with a universal generalised parton distribution (GPD). For the first time, we perform a complete calculation at 1-loop including full leading-log (LL) GPD evolution. We first demonstrate the huge instability of the cross section at high energies when the factorisation scale $\mu_F$ is varied. This instability has been reported previously in the literature, and occurs due to the large logarithms generated by the huge difference between the hard scale of the process, which is the mass of the heavy quarkonium here, and the centre-of-mass energy of the process. This problem was also reported in inclusive heavy vector quarkonium photoproduction. We show that this issue can be resolved by resumming these large logarithms using high-energy factorisation (HEF), by performing a matching with the result in CF using the doubly logarithmic approximation (DLA) in order to be consistent with the fixed order evolution of GPDs. Finally, we show that the cross sections obtained from such a matching, besides being free from the previously mentioned factorisation scale variation instabilities, are consistent with the H1 data for $J/\psi$ production and with the ZEUS data for $\Upsilon$ production., Comment: 7 pages, 3 figures, Presented by S. Nabeebaccus at the 31st International Workshop on Deep Inelastic Scattering (DIS2024)

We introduce a single-mode Cr:ZnS crystalline waveguide ultrafast amplifier that provides a high gain of 5.5 dB/cm and 2.35 W of average output power. The depressed-cladding buried waveguide is produced by an ultrafast laser writing procedure, which allows a high degree of flexibility in fabrication when the geometry, size, and even effective index can be modified along the waveguide. An analytical model that includes both, pump and pulse propagation, allows calculation and optimization of the waveguide design. In a CPA arrangement with a volume Bragg grating-based stretcher/compressor, we demonstrate a broadband 34-mm long amplifier in a polycrystalline Cr:ZnS with a single-pass gain factor of 75 (5.5 dB/cm) and a high average output power up to 2.35 W., Comment: 14 pages, 13 figures

Robots often face challenges in domestic environments where visual feedback is ineffective, such as retrieving objects obstructed by occlusions or finding a light switch in the dark. In these cases, utilizing contacts to localize the target object can be effective. We propose an online planning framework using binary contact signals for manipulation tasks with pose uncertainty, formulated as a Partially Observable Markov Decision Process (POMDP). Naively representing the belief as a particle set makes planning infeasible due to the large uncertainties in domestic settings, as identifying the best sequence of actions requires rolling out thousands of actions across millions of particles, taking significant compute time. To address this, we propose a hierarchical belief representation. Initially, we represent the uncertainty coarsely in a 3D volumetric space. Policies that refine uncertainty in this space are computed and executed, and once uncertainty is sufficiently reduced, the problem is translated back into the particle space for further refinement before task completion. We utilize a closed-loop planning and execution framework with a heuristic-search-based anytime solver that computes partial policies within a limited time budget. The performance of the framework is demonstrated both in real world and in simulation on the high-precision task of inserting a plug into a port using a UR10e manipulator, resolving positional uncertainties up to 50 centimeters and angular uncertainties close to $2\pi$. Experimental results highlight the framework's effectiveness, achieving a 93\% success rate in the real world and over 50\% improvement in solution quality compared to greedy baselines, significantly accelerating planning and enabling real-time solutions for complex problems., Comment: Under review (2025 IEEE International Conference on Robotics & Automation)

We consider a multilayered metamaterial composed of anisotropic layers with in-plane anisotropy axes periodically rotated with respect to each other. Due to the breaking of inversion symmetry, the structure develops an effective chirality. We derive the effective chirality tensor of the metamaterial describing both normal and oblique incidence scenarios. We also identify the arrangement of the layers maximizing chiral response for the fixed lattice period, wavelength and anisotropy of the layers., Comment: 10 pages, 17 pages Supplementary Materials

This preliminary white paper proposes a novel 8-bit floating-point data format HiFloat8 (abbreviated as HiF8) for deep learning. HiF8 features tapered precision. For normal value encoding, it provides 7 exponent values with 3-bit mantissa, 8 exponent values with 2-bit mantissa, and 16 exponent values with 1-bit mantissa. For denormal value encoding, it extends the dynamic range by 7 extra powers of 2, from 31 to 38 binades (notice that FP16 covers 40 binades). Meanwhile, HiF8 encodes all the special values except that positive zero and negative zero are represented by only one bit-pattern. Thanks to the better balance between precision and dynamic range, HiF8 can be simultaneously used in both forward and backward passes of AI training. In this paper, we will describe the definition and rounding methods of HiF8, as well as the tentative training and inference solutions. To demonstrate the efficacy of HiF8, massive simulation results on various neural networks, including traditional neural networks and large language models (LLMs), will also be presented., Comment: 13 Pages, 4 Figures, 9 Tables

We consider families $A(t)$ of self-adjoint operators with symmetry that causes the spectral flow of the family to vanish. We study the secondary $\mathbb{Z}_2$-valued spectral flow of such families. We prove an analog of the Atiyah-Singer-Robbin-Salamon theorem, showing that this secondary spectral flow of $A(t)$ is equal to the secondary $\mathbb{Z}_2$-valued index of the suspension operator $\frac{d}{dt}+A(t)$. Applying this result, we show that the graded secondary spectral flow of a symmetric family of Toeplitz operators on a complete Riemannian manifold equals the secondary index of a certain Callias-type operator. In the case of a pseudo-convex domain, this leads to an odd version of the secondary Boutet de Monvel's index theorem for Toeplitz operators. When this domain is simply a unit disc in the complex plane, we recover the bulk-edge correspondence for the Graf-Porta module for 2D topological insulators of type AII., Comment: 27 pages. Minor corrections, references are updated

The galaxy cluster Abell 496 has been extensively studied in the past for the clear sloshing motion of the hot gas on large scales, but the interplay between the central radio galaxy and the surrounding cluster atmosphere is mostly unexplored. We present a dedicated radio, X-ray, and optical study of Abell 496 aimed at investigating this connection. We use deep radio images obtained with the Giant Metrewave Radio Telescope at 150, 330 and 617 MHz, Very Large Array at 1.4 and 4.8 GHz, and VLA Low Band Ionosphere and Transient Experiment at 340 MHz, with angular resolutions ranging from 0.''5 to 25''. Additionally, we use archival Chandra and Very Large Telescope MUSE observations. The radio images reveal three distinct periods of jet activity: an ongoing episode on sub-kpc scales with an inverted radio spectrum; an older episode that produced lobes on scales $\sim$20 kpc which now have a steep spectral index ($\alpha = 2.0 \pm 0.1$); and an oldest episode that produced lobes on scales of $\sim$50 - 100 kpc with an ultra-steep spectrum ($\alpha = 2.7 \pm 0.2$). Archival Chandra X-ray observations show that the older and oldest episodes have excavated two generations of cavities in the hot gas of the cluster. The outermost X-ray cavity has a clear mushroom-head shape, likely caused by its buoyant rise in the cluster's potential. Cooling of the hot gas is ongoing in the innermost 20 kpc, where H$\alpha$-bright warm filaments are visible in VLT-MUSE data. The H$\alpha$-filaments are stretched towards the mushroom-head cavity, which may have stimulated ICM cooling in its wake. We conclude by commenting on the non-detection of a radio mini-halo in this vigorously sloshing, but low-mass, galaxy cluster., Comment: Accepted for publication in Astronomy & Astrophysics

Machine learning (ML) models trained on datasets owned by different organizations and physically located in remote databases offer benefits in many real-world use cases. State regulations or business requirements often prevent data transfer to a central location, making it difficult to utilize standard machine learning algorithms. Federated Learning (FL) is a technique that enables models to learn from distributed datasets without revealing the original data. Vertical Federated learning (VFL) is a type of FL where data samples are divided by features across several data owners. For instance, in a recommendation task, a user can interact with various sets of items, and the logs of these interactions are stored by different organizations. In this demo paper, we present \emph{Stalactite} - an open-source framework for VFL that provides the necessary functionality for building prototypes of VFL systems. It has several advantages over the existing frameworks. In particular, it allows researchers to focus on the algorithmic side rather than engineering and to easily deploy learning in a distributed environment. It implements several VFL algorithms and has a built-in homomorphic encryption layer. We demonstrate its use on a real-world recommendation datasets.

Multi-Agent Path Finding (MAPF) is the problem of effectively finding efficient collision-free paths for a group of agents in a shared workspace. The MAPF community has largely focused on developing high-performance heuristic search methods. Recently, several works have applied various machine learning (ML) techniques to solve MAPF, usually involving sophisticated architectures, reinforcement learning techniques, and set-ups, but none using large amounts of high-quality supervised data. Our initial objective in this work was to show how simple large scale imitation learning of high-quality heuristic search methods can lead to state-of-the-art ML MAPF performance. However, we find that, at least with our model architecture, simple large scale (700k examples with hundreds of agents per example) imitation learning does \textit{not} produce impressive results. Instead, we find that by using prior work that post-processes MAPF model predictions to resolve 1-step collisions (CS-PIBT), we can train a simple ML MAPF model in minutes that dramatically outperforms existing ML MAPF policies. This has serious implications for all future ML MAPF policies (with local communication) which currently struggle to scale. In particular, this finding implies that future learnt policies should (1) always use smart 1-step collision shields (e.g. CS-PIBT), (2) always include the collision shield with greedy actions as a baseline (e.g. PIBT) and (3) motivates future models to focus on longer horizon / more complex planning as 1-step collisions can be efficiently resolved.

We propose a programming technology that bridges cross-platform compatibility and hardware acceleration in ray tracing applications. Our methodology enables developers to define algorithms while our translator manages implementation specifics for different hardware or APIs. Features include: generating hardware-accelerated code from hardware-agnostic, object-oriented C++ algorithm descriptions; enabling users to define software fallbacks for non-hardware-accelerated CPUs and GPUs; producing GPU programming API-based algorithm implementations resembling manually ported C++ versions. The generated code is editable and readable, allowing for additional hardware acceleration. Our translator supports single megakernel and multiple kernel path tracing implementations without altering the programming model or input source code. Wavefront mode is crucial for NeRF and SDF, ensuring efficient evaluation with multiple kernels. Validation on tasks such as BVH tree build/traversal, ray-surface intersection for SDF, ray-volume intersection for 3D Gaussian Splatting, and complex Path Tracing models showed comparable performance levels to expert-written implementations for GPUs. Our technology outperformed existing Path Tracing implementations.

We train, for the first time, large language models using FP8 precision on datasets up to 2 trillion tokens -- a 20-fold increase over previous limits. Through these extended training runs, we uncover critical instabilities in FP8 training that were not observable in earlier works with shorter durations. We trace these instabilities to outlier amplification by the SwiGLU activation function. Interestingly, we show, both analytically and empirically, that this amplification happens only over prolonged training periods, and link it to a SwiGLU weight alignment process. To address this newly identified issue, we introduce Smooth-SwiGLU, a novel modification that ensures stable FP8 training without altering function behavior. We also demonstrate, for the first time, FP8 quantization of both Adam optimizer moments. Combining these innovations, we successfully train a 7B parameter model using FP8 precision on 256 Intel Gaudi2 accelerators, achieving on-par results with the BF16 baseline while delivering up to a $\sim 34 \%$ throughput improvement.

Balanced double-null configurations are of general interest for boundary plasma physics, and they have been proposed for some future designs. Experimental observations demonstrate absence of plasma fluctuations in tokamak high-field side scrape-off layer in a balanced double-null configuration [Smick et al 2013 Nucl. Fusion 53 023001], and it is commonly assumed that plasma instabilities are suppressed on high-field side in the edge plasma due to the stabilizing effect of magnetic curvature. At the same time, the experimental evidence points to extremely steep plasma density profiles on high-field side, which should provide a strong instability drive. In the present study, the drift-resistive-ballooning mode instability model is investigated analytically and numerically to determine the characteristics of plasma instabilities, turbulence, and transport in tokamak scrape-off layer on high-field side.

Controlling the spectral norm of the Jacobian matrix, which is related to the convolution operation, has been shown to improve generalization, training stability and robustness in CNNs. Existing methods for computing the norm either tend to overestimate it or their performance may deteriorate quickly with increasing the input and kernel sizes. In this paper, we demonstrate that the tensor version of the spectral norm of a four-dimensional convolution kernel, up to a constant factor, serves as an upper bound for the spectral norm of the Jacobian matrix associated with the convolution operation. This new upper bound is independent of the input image resolution, differentiable and can be efficiently calculated during training. Through experiments, we demonstrate how this new bound can be used to improve the performance of convolutional architectures., Comment: ECCV 2024

The monography considers the problem of constructing a Hamiltonian cycle in a complete graph. A rule for constructing a Hamiltonian cycle based on isometric cycles of a graph is established. An algorithm for constructing a Hamiltonian cycle based on ring summation of isometric cycles of a graph is presented. Based on the matrix of distances between vertices, the weight of each cycle is determined as an additive sum of the weights of its edges. To construct an optimal route of a graph, the basic idea of finding an optimal route between four vertices is used. Further successive constructions are aimed at joining an adjacent isometric cycle with an increase in the number of vertices by one unit. The recursive process continues until all vertices of the graph are connected. Based on the introduced mathematical apparatus, the monography presents a new algorithm for solving the symmetric Traveling salesman problem. Some examples of solving the problem are provided., Comment: 46 pages, 32 figures, a preprint of monography, in Ukrainian language

We examine the color-kinematics duality within the BV formalism, highlighting its emergence as a feature of specific gauge-fixed actions. Our goal is to establish a general framework for studying the duality while investigating straightforward examples of off-shell color-kinematics duality. In this context, we revisit Chern-Simons theory as well as introduce new examples, including BF theory and 2D Yang-Mills theory, which are shown to exhibit the duality off-shell. We emphasize that the geometric structures responsible for flat-space color-kinematics duality appear for general curved spaces as well.

Empirically, the prevailing market prices for liquidity tokens of the constant product market maker (CPMM) -- as offered in practice by companies such as Uniswap -- readily permit arbitrage opportunities by delta hedging the risk of the position. Herein, we investigate this arbitrage opportunity by treating the liquidity token as a derivative position in the prices of the underlying assets for the CPMM. In doing so, not dissimilar to the Black-Scholes result, we deduce risk-neutral pricing and hedging formulas for these liquidity tokens. Furthermore, with our novel pricing formula, we construct a method to calibrate a volatility to data which provides an updated (non-market) price which would not permit arbitrage if quoted by the CPMM. We conclude with a discussion of novel AMM designs which would bring the pricing of liquidity tokens into the modern financial era.

Point cloud completion aims to recover the complete 3D shape of an object from partial observations. While approaches relying on synthetic shape priors achieved promising results in this domain, their applicability and generalizability to real-world data are still limited. To tackle this problem, we propose a self-supervised framework, namely RealDiff, that formulates point cloud completion as a conditional generation problem directly on real-world measurements. To better deal with noisy observations without resorting to training on synthetic data, we leverage additional geometric cues. Specifically, RealDiff simulates a diffusion process at the missing object parts while conditioning the generation on the partial input to address the multimodal nature of the task. We further regularize the training by matching object silhouettes and depth maps, predicted by our method, with the externally estimated ones. Experimental results show that our method consistently outperforms state-of-the-art methods in real-world point cloud completion.

This paper describes the 2nd edition of the ICML Topological Deep Learning Challenge that was hosted within the ICML 2024 ELLIS Workshop on Geometry-grounded Representation Learning and Generative Modeling (GRaM). The challenge focused on the problem of representing data in different discrete topological domains in order to bridge the gap between Topological Deep Learning (TDL) and other types of structured datasets (e.g. point clouds, graphs). Specifically, participants were asked to design and implement topological liftings, i.e. mappings between different data structures and topological domains --like hypergraphs, or simplicial/cell/combinatorial complexes. The challenge received 52 submissions satisfying all the requirements. This paper introduces the main scope of the challenge, and summarizes the main results and findings., Comment: Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM) at ICML 2024

The physics case for quarkonium-production studies accessible at the US Electron Ion Collider is described., Comment: Latex, 84 pages. Review prepared for Progress in Particle and Nuclear Physics

The stochastic background of gravitational wave signals arising from the core-collapse supernovae is produced through various complex mechanisms that need detailed and careful investigation. We proposed a simplified multi-peak waveform of the amplitude spectrum. The corresponding energy spectra of our model fit the energy spectra obtained from different numerical simulations of various types of core-collapse events such as non-rotating, slow and fast rotating massive progenitors resulting in neutron stars. The maximum dimensionless energy density $\Omega_{gw}$ corresponding to our model is of $\mathcal{O}(10 {^{-12})} $ around 650 Hz. Assuming some degree of uncertainty, we estimated the parameters for the core collapse waveform using BILBY. We studied the detectability of the signal of our model against gravitational-wave detectors like the Einstein Telescope, advanced LIGO and Virgo. Our study indicates that these detectors have to gain more sensitivity to pick up the gravitational wave signals of stochastic background., Comment: Accepted for publication in Phys. Rev. D

Large language models (LLMs) have triggered a new stream of research focusing on compressing the context length to reduce the computational cost while ensuring the retention of helpful information for LLMs to answer the given question. Token-based removal methods are one of the most prominent approaches in this direction, but risk losing the semantics of the context caused by intermediate token removal, especially under high compression ratios, while also facing challenges in computational efficiency. In this work, we propose context-aware prompt compression (CPC), a sentence-level prompt compression technique where its key innovation is a novel context-aware sentence encoder that provides a relevance score for each sentence for a given question. To train this encoder, we generate a new dataset consisting of questions, positives, and negative pairs where positives are sentences relevant to the question, while negatives are irrelevant context sentences. We train the encoder in a contrastive setup to learn context-aware sentence representations. Our method considerably outperforms prior works on prompt compression on benchmark datasets and is up to 10.93x faster at inference compared to the best token-level compression method. We also find better improvement for shorter length constraints in most benchmarks, showing the effectiveness of our proposed solution in the compression of relevant information in a shorter context. Finally, we release the code and the dataset for quick reproducibility and further development: https://github.com/Workday/cpc.

Modern artificial intelligence (AI) systems are powered by foundation models. This paper presents a new set of foundation models, called Llama 3. It is a herd of language models that natively support multilinguality, coding, reasoning, and tool usage. Our largest model is a dense Transformer with 405B parameters and a context window of up to 128K tokens. This paper presents an extensive empirical evaluation of Llama 3. We find that Llama 3 delivers comparable quality to leading language models such as GPT-4 on a plethora of tasks. We publicly release Llama 3, including pre-trained and post-trained versions of the 405B parameter language model and our Llama Guard 3 model for input and output safety. The paper also presents the results of experiments in which we integrate image, video, and speech capabilities into Llama 3 via a compositional approach. We observe this approach performs competitively with the state-of-the-art on image, video, and speech recognition tasks. The resulting models are not yet being broadly released as they are still under development.

Effective axion fields in condensed matter and photonics are manifested as $\mathcal{P}$- and $\mathcal{T}$-odd contributions to the electromagnetic response. Here, we show that the phenomena previously attributed to the effective axion fields have two distinct physical origins. One of them corresponds to the standard axion electrodynamics, while another provides its dual-symmetric version having the same symmetry and featuring similar but distinguishable optical properties. We present an example system described by the dual-symmetric modification of axion electrodynamics, derive the key predictions and pinpoint experimentally observable distinctions between the two versions of axion-type response., Comment: 6 pages+23 pages of Supplementary Materials

The development of X-ray free-electron lasers (XFELs) has enabled ultrafast X-ray diffraction (XRD) experiments, which are capable of resolving electronic/vibrational transitions and structural changes in molecules, or capturing molecular movies. While time-resolved XRD has received increasing attention, the extraction of information content from signals is challenging and requires theoretical support. In this work, we combined X-ray scattering theory and trajectory surface hopping approach to resolve dynamical changes in the electronic structure of photo-excited molecules by studying time evolution of electron density changes between electronic excited states and ground state. Using pyrazine molecule as an example, we show that key features of reaction pathways can be identified, enabling the capture of structural changes associated with electronic transitions for a photo-excited molecule.

Parameter Efficient Fine-Tuning (PEFT) methods have gained popularity and democratized the usage of Large Language Models (LLMs). Recent studies have shown that a small subset of weights significantly impacts performance. Based on this observation, we introduce a novel PEFT method, called Gaussian noise Injected Fine Tuning of Salient Weights (GIFT-SW). Our method updates only salient columns, while injecting Gaussian noise into non-salient ones. To identify these columns, we developeda generalized sensitivity metric that extends and unifies metrics from previous studies. Experiments with LLaMA models demonstrate that GIFT-SW outperforms full fine-tuning and modern PEFT methods under the same computational budget. Moreover, GIFT-SW offers practical advantages to recover performance of models subjected to mixed-precision quantization with keeping salient weights in full precision.

The Aharonov-Bohm effect is a quantum mechanical phenomenon that demonstrates how potentials can have observable effects even when the classical fields associated with those potentials are absent. Initially proposed for electromagnetic interactions, this effect has been experimentally confirmed and extensively studied over the years. More recently, the effect has been observed in the context of gravitational interactions using atom interferometry. Additionally, recent predictions suggest that temporal variations in the phase of an electron wave function will induce modulation sidebands in the energy levels of an atomic clock, solely driven by a time-varying scalar gravitational potential [1]. In this study, we consider the atomic clock as a two-level system undergoing continuous Rabi oscillations between the electron's ground and excited state. We assume the photons driving the transition are precisely frequency-stabilized to match the transition, enabling accurate clock comparisons. Our analysis takes into account, that when an atom transitions from its ground state to an excited state, it absorbs energy, increasing its mass according to the mass-energy equivalence principle. Due to the mass difference between the two energy levels, we predict that an atomic clock in an eccentric orbit will exhibit a constant frequency shift relative to a ground clock corresponding to the orbit's average gravitational redshift, with additional modulation sidebands due to the time-varying gravitational potential.

Using the model of four sublattices, the Landau-Ginzburg-Devonshire-Kittel phenomenological approach and the Stephenson-Highland ionic adsorption model for the description of coupled polar and antipolar long-range orders in ferroics, we calculated analytically the phase diagrams and polar properties of Bi$_{1-x}$Sm$_x$FeO$_3$ nanoparticles covered by surface ions in dependence on their size, surface ions density, samarium content "x" and temperature. The size effects and ferro-ionic coupling govern the appearance and stability conditions of the long-range ordered ferroelectric, reentrant ferrielectric and antiferroelectric phases in the Bi1-xSmxFeO3 nanoparticles. Calculated phase diagrams are in a qualitative agreement with the X-ray diffraction phase analysis, electron paramagnetic resonance, infra-red spectroscopy and electrophysical measurements of the Bi$_{1-x}$Sm$_x$FeO$_3$ nanopowders sintered by the solution combustion method. The combined theoretical-experimental approach allows to establish the influence of the ferro-ionic coupling and size effects in Bi$_{1-x}$Sm$_x$FeO$_3$ nanoparticles on their polar properties., Comment: 45 pages, including 13 figures and Appendixes, New TEM images are added

We derive optimal asymptotic and non-asymptotic lower bounds on the Widom factors for weighted Chebyshev and orthogonal polynomials on compact subsets of the real line. In the Chebyshev case we extend the optimal non-asymptotic lower bound previously known only in a handful of examples to regular compact sets and a large class weights. Using the non-asymptotic lower bound, we extend Widom's asymptotic lower bound for weights bounded away from zero to a large class of weights with zeros including weights with strong zeros and infinitely many zeros. As an application of the asymptotic lower bound we extend Bernstein's 1931 asymptotics result for weighted Chebyshev polynomials on an interval to arbitrary Riemann integrable weights with finitely many zeros and to some continuous weights with infinitely many zeros. In the case of orthogonal polynomials, we derive optimal asymptotic and non-asymptotic lower bound on arbitrary regular compact sets for a large class of weights in the non-asymptotic case and for arbitrary Szeg\H{o} class weights in the asymptotic case, extending previously known bounds on finite gap and Parreau--Widom sets.

The PHENIX Collaboration has actively pursued a Data and Analysis Preservation program since 2019, the first such dedicated effort at RHIC. A particularly challenging aspect of this endeavor is preservation of complex physics analyses, selected for their scientific importance and the value of the specific techniques developed as a part of the research. For this, we have chosen one of the most impactful PHENIX results, the joint study of direct photons and neutral pions in high-energy d+Au collisions. To ensure reproducibility of this analysis going forward, we partitioned it into self-contained tasks and used a combination of containerization techniques, code management, and robust documentation. We then leveraged REANA (the platform for reproducible analysis developed at CERN) to run the required software. We present our experience based on this example, and outline our future plans for analysis preservation., Comment: Paper submitted to the Proceedings of the Advanced Computing and Analysis Techniques 2024 (ACAT) conference

Resonant dielectric planar structures can interact selectively with light of particular helicity thus providing an attractive platform for chiral flat optics. The absence of mirror-symmetry planes defines geometric chirality, and it remains the main condition for achieving strong circular dichroism. For planar optical structures such as photonic-crystal membranes and metasurfaces, breaking of out-of-plane mirror symmetry is especially challenging, as it requires to fabricate meta-atoms with a tilt, variable height, or vertically shifted positions. Although transparent substrates formally break out-of-plane mirror symmetries, their optical effect is typically subtle being rarely considered for enhancing optical chirality. Here we reveal that low-refractive-index substrates can induce up to maximum optical chirality in otherwise achiral metastructures so that the transparency to waves of one helicity is combined with resonant blocking of waves of the opposite helicity. This effect originates from engineering twisted photonic eigenstates of different parities. Our perturbation analysis developed in terms of the resonant-state expansion reveals how the eigenstate coupling induced by a substrate gives rise to a pair of chiral resonances of opposite handedness. Our general theory is confirmed by the specific examples of light transmission in the normal and oblique directions by a rotation-symmetric photonic-crystal membrane placed on different transparent substrates., Comment: 18 pages, 6 figures

The transition matrix, frequently abbreviated as T-matrix, contains the complete information in a linear approximation of how a spatially localized object scatters an incident field. The T-matrix is used to study the scattering response of an isolated object and describes the optical response of complex photonic materials made from ensembles of individual objects. T-matrices of certain common structures, potentially, have been repeatedly calculated all over the world again and again. This is not necessary and constitutes a major challenge for various reasons. First, the resources spent on their computation represent an unsustainable financial and ecological burden. Second, with the onset of machine learning, data is the gold of our era, and it should be freely available to everybody to address novel scientific challenges. Finally, the possibility of reproducing simulations could tremendously improve if the considered T-matrices could be shared. To address these challenges, we found it important to agree on a common data format for T-matrices and to enable their collection from different sources and distribution. This document aims to develop the specifications for storing T-matrices and associated metadata. The specifications should allow maximum freedom to accommodate as many use cases as possible without introducing any ambiguity in the stored data. The common format will assist in setting up a public database of T-matrices., Comment: Submitted to the Journal of Quantitative Spectroscopy and Radiative Transfer