arXiv:2607.04189v1 Announce Type: new Abstract: Federated Learning (FL) is fundamentally challenged by statistical heterogeneity, where non-identically distributed (non-IID) data induces client drift that severely hampers global convergence. While existing approaches attempt to mitigate this drift through spatial-domain gradient correction or regularization, they overlook the intrinsic spectral structure of optimization signals. In this work, we revisit client drift from a novel frequency-domain perspective and uncover a critical Spectral Bias of Drift: inter-client gradient divergence is predominantly concentrated in low-frequency components which encode client-specific distributional shifts, while high-frequency components representing fine-grained features remain relatively consistent. Motivated by this, we propose SpecGradFilter, a unified Spectral Gradient Filtering Framework that tames heterogeneity by suppressing discordant low-frequency signals. Crucially, we demonstrate that SpecGradFilter is a generalizable principle, effective not only via precise FFT-based truncation but also through spatial approximations like Gaussian detrending. Extensive experiments on benchmarks such as CIFAR-10/100 and Tiny-ImageNet demonstrate that SpecGradFilter significantly performs better performance in highly Non-IID settings with negligible communication overhead, establishing a new paradigm for robust federated optimization.
Science Journals
arXiv:2403.11199v2 Announce Type: replace Abstract: Unitarity is a useful principle for stabilizing deep neural networks, but in graph neural networks (GNNs) instability is induced not only by learnable parameters but also by the graph propagation operator. Motivated by this distinction, we propose Graph Unitary Message Passing (GUMP), a message-passing framework that uses a unitary propagation operator on a transformed graph to avoid graph-induced exponential decay under repeated propagation. GUMP combines (i) a graph transformation that maps an input graph to an Eulerian line-graph construction admitting unitary adjacency matrices, and (ii) a practical unitary projection procedure based on Newton-Schulz iteration. Theoretical analysis clarifies that, under standard analysis assumptions, unitary propagation keeps the graph-propagation term depth-stable, while vanilla normalized propagation exhibits exponential decay in its non-trivial spectral components. Across synthetic long-range tasks, TUDataset benchmarks, and LRGB datasets, GUMP improves over vanilla message passing and achieves competitive or superior performance against strong baselines. Code is available at https://github.com/ucker/gump_code.
arXiv:2403.20000v3 Announce Type: replace Abstract: In this paper the recoverable robust shortest path problem is investigated. Discrete budgeted interval uncertainty representation is used to model uncertain second-stage arc costs. The known complexity results for this problem are strengthened. It is shown that it is Sigma_3^p-hard for the arc exclusion and the arc symmetric difference neighborhoods. Furthermore, it is also proven that the inner adversarial problem for these neighborhoods is Pi_2^p-hard.
arXiv:2510.06459v5 Announce Type: replace-cross Abstract: This article describes an extension to the effective Average-Atom (AA) method for random alloys to account for local ordering (short-range order) effects by utilizing information from partial radial distribution functions. The new Local Order Average-Atom (LOAA) method is rigorously derived based on statistical mechanics arguments and validated for non-stoichiometric binary 2D hexagonal crystals and 3D FeNiCr and NiAl alloys whose ground state is obtained through Monte Carlo sampling. Material properties for these alloys computed from atomistic simulations using standard interatomic potentials (IPs) exhibit a strong dependence on local ordering that is captured by simulations with effective LOAA IPs, but not the original AA method. The advantage of LOAA is that it requires smaller system sizes to achieve statistically converged results and therefore enables the simulation of complex materials, such as high-entropy alloys, at a fraction of the computational cost of standard IPs.
arXiv:2412.02974v2 Announce Type: replace-cross Abstract: Markov chain Monte Carlo (MCMC) is a powerful tool for sampling from complex probability distributions. Despite its versatility, MCMC often suffers from strong autocorrelation and the negative sign problem, leading to slowing down the convergence of statistical error. We propose a novel MCMC formulation based on tensor network representations to reduce the population variance and mitigate these issues systematically. By introducing stochastic projectors into the tensor network framework and employing Markov chain sampling, our method eliminates the systematic error associated with low-rank approximation in tensor contraction while maintaining the high accuracy of the tensor network method. We demonstrate the effectiveness of the proposed method on the two-dimensional Ising model, achieving an exponential reduction in statistical error with increasing bond dimension cutoff. Furthermore, we address the sign problem in systems with negative weights, showing significant improvements in average signs as bond dimension cutoff increases. The proposed framework provides a robust solution for accurate statistical estimation in complex systems, paving the way for broader applications in computational physics and beyond.
arXiv:2502.16285v2 Announce Type: replace-cross Abstract: Control of magnetic order through optical manipulation of orbital states can be realized in a wide range of materials, including insulators. In the latter case, electrons are excited into different electronic states, whose character may be a complex result of spin-orbit-lattice coupling. Here, we use a combination of polarization-resolved photoluminescence spectroscopy and density functional theory calculations to study excited Cr3+ ions in bulk, insulating antiferromagnetic CrCl3. Our results demonstrate a temperature-dependent preferential polarization direction below ~17 K, which reorients in the presence of an applied in-plane magnetic field. Density functional theory calculations reveal a local distortion of the ligand field and the generation of out-of-plane spin and orbital moments, following Cr3+ ion excitation. These changes link the photoluminescence polarization to the magnetic state of CrCl3. Our findings provide insight into the excited-state electronic structure of Cr3+ ions and suggest that CrCl3 is a promising candidate for opto-orbitronic applications.
arXiv:2503.23650v2 Announce Type: replace Abstract: Reinforcement learning (RL), with its ability to explore and optimize policies in complex, dynamic decision-making tasks, has emerged as a promising approach to addressing motion planning (MoP) challenges in autonomous driving (AD). Despite rapid advancements in RL and AD, a systematic description and interpretation of the RL design process tailored to diverse driving tasks remains underdeveloped. This survey provides a comprehensive review of RL-based MoP for AD, focusing on lessons from task-specific perspectives. We first outline the fundamentals of RL methodologies, and then survey their applications in MoP, analyzing scenario-specific features and task requirements to shed light on their influence on RL design choices. Building on this analysis, we summarize key design experiences, extract insights from various driving task applications, and provide guidance for future implementations. Additionally, we examine the frontier challenges in RL-based MoP, review recent efforts to addresse these challenges, and propose strategies for overcoming unresolved issues.
arXiv:2504.04630v3 Announce Type: replace Abstract: Foundation Models (FMs), particularly Large Language Models (LLMs), are increasingly used to support various software engineering activities (e.g., coding and testing). Their adoption in software engineering of Cyber-Physical Systems (CPSs) is also growing. However, research in this area remains limited. Most existing studies have primarily focused on LLMs, only one type of FM, leaving ample opportunities to explore other FMs, such as vision-language models. We argue that, in addition to LLMs, other FMs utilizing different data modalities (e.g., images, audio) and multimodal models (which integrate multiple modalities) hold great potential for supporting CPS software engineering, given that these systems process diverse data types. To address this, in this first systematic effort, we present a forward-looking research roadmap for integrating FMs into commonly known phases of CPS software engineering, thereby making it accessible to most software engineers. We derive the roadmap from the literature, emerging trends in FMs, and gaps identified from the literature. The roadmap highlights key challenges and actionable research opportunities for the software engineering community to guide future research. Moreover, we discuss the common challenges associated with applying FMs across six dimensions (e.g., technical, economic and resource, and human aspects). This roadmap aims to provide a visionary guide for researchers and practitioners, outlining directions for future work and performing future empirical studies.
arXiv:2607.04772v1 Announce Type: new Abstract: With the increasing threat of malware across various software related domains, malware detection and classification is critical to determine the response actions. Different strategies have been adopted to address the challenge of malware detection. With the advent of deep learning techniques, malware detection using image processing has garnered research attention. In this work, we proposed a novel malware binary to image transformation technique HilEnT based on a combination of Hilbert curve-based transformation of malware binary and the entropy feature comparison of malware file with benign and malware classes. Three grayscale images produced during this process are combined to form a three-channel colored image which is then used for malware detection using machine learning techniques. We performed supervised binary and multiclass classification to evaluate the effectiveness of our proposed HilEnT. We also evaluated a few-shot learning technique to assess the robustness of our proposed HilEnT in a practical setting where the number of available class samples is limited. Furthermore, we investigated the benefits of combination of Histogram of Oriented Gradients and Principal Component Analysis for time performance improvements through feature reduction techniques. We evaluated our proposed methodology on four datasets: Dike, Michael Lester Dataset, Microsoft BIG 2015 and a self-collected dataset, and achieved the state-of-the-art results.
arXiv:2607.04776v1 Announce Type: new Abstract: Pulse experiments in the temporal analysis of products (TAP) reactor are one of the most important methods for studying transient kinetics of gas-solid catalytic reactions. The Y-procedure (Yablonsky et al., Chem. Eng. Sci. 62, 6754, 2007) is a model-free analysis framework for inferring the relationship between the reaction-rate $R$ and the reactant concentration $C$ from measurements of the outlet flux of gas. While elegant in conception, its application is hindered by the amplification of measurement noise that results from having to backtrack diffusive transport from the outlet to the reaction zone. Here, we explicitly recognize the inverse problem inherent in the Y-procedure and treat it using well-developed tools from the field of inverse problems. While previous implementations of the Y-procedure used Fourier-based filtering, we do not pre-process the measurements with an ad hoc noise-filter. Instead, we use a basis of localized square pulses to formulate a discrete inverse problem, whose regularized solution is obtained via the truncated singular value decomposition (TSVD) method. This method requires one to select a cutoff mode number; while we show how the choice of this regularization parameter can be guided by a Picard plot, we also develop an objective selection strategy for state defining experiments, for which $R(C)$ is a single-valued function. We apply our proposed inverse-problem approach to synthetic data corresponding to linear and nonlinear reactions and compare the results with the Fourier-filtration method. The former produces better reconstructions of the $R$ vs $C$ relationship, especially for nonlinear reactions. Our work facilitates the automation of pulse response analyses and enables the application of other discrete inverse-problem techniques, such as Tikhonov regularization or machine-learning methods.
arXiv:2510.02208v3 Announce Type: replace-cross Abstract: Diffusion models have emerged as powerful generative priors for solving inverse imaging problems. However, their practical deployment is hindered by the substantial computational cost of slow, multi-step sampling. Although Consistency Models (CMs) address this limitation by enabling high-quality generation in only one or a few steps, their direct application to inverse problems has remained largely unexplored. This paper introduces a modified consistency sampling framework specifically designed for inverse problems. The proposed approach regulates the sampler's stochasticity through a measurement-consistency mechanism that leverages the degradation operator, thereby enforcing fidelity to the observed data while preserving the computational efficiency of consistency-based generation. Comprehensive experiments on the Fashion-MNIST and LSUN Bedroom datasets demonstrate consistent improvements across both perceptual and pixel-level metrics, including the Fr\'echet Inception Distance (FID), Kernel Inception Distance (KID), peak signal-to-noise ratio (PSNR), and structural similarity index measure (SSIM), compared with baseline consistency and diffusion-based sampling methods. The proposed method achieves competitive or superior reconstruction quality with only a small number of sampling steps.
arXiv:2510.08432v3 Announce Type: replace-cross Abstract: Pebble games, an abstraction from classical reversible computing, have found use in the design of quantum circuits for inherently sequential tasks. Gidney showed that allowing Hadamard basis measurements during pebble games can dramatically improve costs -- an extension termed "spooky pebble games" because the measurements leave temporary phase errors called ghosts. Separately, previous work by Blocki et al. studied the benefits of parallelism in pebble games. In this work we define and study parallel spooky pebble games, showing that parallelism and spookiness can yield impressive gains when used together. First, we show by construction that a line graph of length $\ell$ can be pebbled in depth $2\ell$ (exactly optimal) using space $\leq 2.47\log \ell$. Then, to explore pebbling schemes using even less space, we use a highly optimized $A^*$ search implemented in Julia to find the lowest-depth parallel spooky pebbling possible for a range of concrete line graph lengths $\ell$ given a constant number of pebbles $s$. We then show that these techniques can significantly reduce the cost of the arithmetic in Regev's factoring algorithm. For example, we find that 4096-bit integers $N$ can be factored in multiplication depth 193, which outperforms the 680 required of previous variants of Regev and the 444 reported by Eker{\aa} and G\"artner for Shor's algorithm. While the space required for Shor's algorithm is considerably less than any variant of Regev's algorithm including ours, and thus Shor likely remains the best candidate for the first quantum factorization of large integers, our results show that implementations of Regev's algorithm are far from fully optimized, and Regev's algorithm may have practical importance in the future. We also believe our pebbling techniques are applicable in quantum cryptanalysis beyond integer factorization, and in quantum circuit compilation more broadly.
arXiv:2602.08022v3 Announce Type: replace-cross Abstract: We provide a link between response theory, pullback measures, and optimal fingerprinting method that paves the way for a) predicting the impact of acting forcings on time-dependent systems and b) attributing observed anomalies to acting forcings when the reference state is not time-independent. We derive formulas for linear response theory for time-dependent Markov chains and diffusion processes. We discuss existence, uniqueness, and differentiability of the equivariant measure under general (not necessarily slow or periodic) perturbations of the transition kernels. Our results allow for extending the theory of optimal fingerprinting for detection and attribution of climate change (or change in any complex system) when the background state is time-dependent amd when the optimal solution is sought for multiple time slices at the same time. We provide numerical support for the findings by applying our theory to a modified version of the Ghil-Sellers energy balance model. We verify the precision of response theory - even in a coarse-grained setting - in predicting the impact of increasing CO$_2$ concentration on the temperature field. Additionally, we show that the optimal fingerprinting method developed here is capable to attribute the climate change signal to multiple acting forcings across a vast time horizon.
arXiv:2606.17285v2 Announce Type: replace-cross Abstract: Many nonsmooth, nonconvex objectives in learning and signal recovery are $\rho$-weakly convex. We minimize such a function in deterministic and stochastic settings when the weak-convexity parameter $\rho$ is unknown. The objective is not required to be globally Lipschitz continuous or smooth. We propose the Adaptive Prox-Guided Scheme (APS), a single-trial adaptive proximal framework that adapts the proximal parameter online and bidirectionally through a descent test, allowing it to exploit favorable local structure. The framework covers adaptive proximal-point, prox-linear, and proximal-gradient methods from a model-based viewpoint. In the deterministic proximal setting, APS obtains an $O(\varepsilon^{-2})$ iteration complexity for producing an $\varepsilon$-subgradient stationary point. In the stochastic setting, APS achieves a high-probability $O(\varepsilon^{-2})$ iteration bound for driving the Moreau-envelope gradient below $\varepsilon$. This result holds under deliberately weak oracle assumptions: the function-difference estimates may be biased and heavy-tailed, and the stochastic candidate oracle need only be sufficiently accurate with constant probability when the proximal parameter is small, and can be arbitrary otherwise.
arXiv:2406.07925v2 Announce Type: replace Abstract: Large language models (LLMs) have emerged as important components across various fields, yet their training requires substantial computation resources and abundant labeled data. It poses a challenge to robustly training LLMs for individual users (clients). To tackle this challenge, the intuitive idea is to introduce federated learning (FL), which can collaboratively train models on distributed private data. However, existing methods suffer from the challenges of data heterogeneity, system heterogeneity, and model size, resulting in suboptimal performance and high costs. In this work, we proposed a variant of personalized federated learning (PFL) framework, namely FDLoRA, which allows the client to be a single device or a cluster and adopts low-rank adaptation (LoRA) tuning. FDLoRA sets dual LoRA modules on each client to capture personalized and global knowledge, respectively, and only the global LoRA module uploads parameters to the central server to aggregate cross-client knowledge. Finally, an adaptive fusion approach is employed to combine the parameters of the dual LoRAs. This enables FDLoRA to make effective use of private data distributed across different clients, thereby improving performance on the client without incurring high communication and computing costs. We conducted extensive experiments in two practice scenarios. The results demonstrate that FDLoRA outperforms six baselines in terms of performance, stability, robustness, computation cost, and communication cost.
arXiv:2504.09662v4 Announce Type: replace Abstract: Multi-agent large language model simulations have the potential to model complex human behaviors and interactions. If the mechanics are set up properly, unanticipated and valuable social dynamics can surface. However, it is challenging to consistently enforce simulation mechanics while still allowing for rich and emergent dynamics. We present AgentDynEx, an AI system that helps set up, track, and repair simulations. Specifically, AgentDynEx introduces milestones that act as checkpoints and failure conditions that act as guardrails to ensure dynamics are relevant and mechanics are respected as the simulation progresses. It also introduces a method called nudging, where the system dynamically reflects on simulation progress and gently intervenes if it begins to deviate from intended outcomes. A technical evaluation found that nudging enables simulations to progress further without reducing the presence notable dynamics compared to simulations without nudging. A case study with AgentDynEx documented instances where real users were able to simulate lived experiences. We discuss the importance of nudging as a technique for guiding agents towards desirable behaviors while preserving their freedom of choice.
arXiv:2603.24847v2 Announce Type: replace Abstract: Coronary artery disease, a leading cause of cardiovascular mortality worldwide, can be assessed non-invasively by coronary computed tomography angiography (CCTA). Although deep learning has advanced automated CCTA analysis, clinical translation remains constrained by the scarcity of expert-annotated data and by the spatial sparsity of coronary pathology, which occupies only a small fraction of each scan. Widely used label-free pretraining strategies, such as masked image modeling and contrastive learning, optimize for global anatomical reconstruction and tend to under-represent these tiny localized pathological features. Here we present CORA, an annotation-efficient model for comprehensive coronary artery disease assessment. Rather than reconstructing background anatomy, CORA learns from volumetric CCTA through a synthesis-driven self-supervised strategy: an anatomy-guided engine inserts diverse synthetic calcified and non-calcified lesions into unlabeled scans, reframing pretraining as an abnormality-detection task that biases representation learning toward clinically relevant disease features. We pretrained CORA on 10,138 unlabeled CCTA volumes and evaluated it across datasets from nine independent hospitals. Across plaque characterization, stenosis detection, and coronary artery segmentation, CORA consistently outperformed strong self-supervised pretraining baselines, with the largest gains on external multi-center data, indicating robust generalization under distributional shift. Coupling the imaging encoder with structured clinical variables further enabled near-term major adverse cardiac event (MACE) risk stratification. Our results show that pathology-centric, synthesis-driven pretraining is an effective and scalable strategy for annotation-efficient coronary artery disease assessment from CCTA.
arXiv:2607.05055v1 Announce Type: new Abstract: Healthcare appointment scheduling remains a persistent operational bottleneck, driven by manual coordination, fragmented legacy systems, and high administrative overhead. These inefficiencies constrain provider availability and degrade patient access to care. This paper presents CareConnect, a safety-first conversational agent for healthcare logistics automation that leverages large language model (LLM) function calling, retrieval-augmented generation (RAG), and layered deterministic safety guardrails. The system orchestrates eight domain-specific tools to support appointment booking, modification, cancellation, and facility information retrieval, while enforcing strict scope constraints that prohibit medical advice or diagnosis. Safety-critical situations are handled through deterministic short-circuit mechanisms for emergency detection and medical intent refusal. We evaluate CareConnect on a comprehensive benchmark of 680 task-oriented scenarios spanning end-to-end workflows, multi-turn interactions, and edge cases. Experimental results demonstrate a 91.8% task completion rate with a median per-request latency of 2.2 seconds, 96.0% safety compliance on the dedicated safety-critical evaluation subset, and an average operational cost of $0.0324 per appointment, yielding a significant cost reduction compared to manual human scheduling. These findings show that carefully scoped and rigorously safeguarded LLM-based agents can reliably automate complex healthcare operational workflows while maintaining safety guarantees and achieving substantial cost efficiency. The source code and system implementation are publicly available at https://github.com/Hadi-Hsn/CareConnect.
arXiv:2501.15177v3 Announce Type: replace Abstract: Audio-Language Models (ALMs), trained on paired audio-text data, are designed to process, understand, and reason about audio-centric multimodal content. Unlike traditional supervised approaches that use predefined labels, ALMs leverage natural language supervision to better handle complex real-world audio scenes with multiple overlapping events. While demonstrating impressive zero-shot and task generalization capabilities, there is still a notable lack of systematic surveys that comprehensively organize and analyze developments. In this paper, we present the first systematic review of ALMs with three main contributions: (1) comprehensive coverage of ALM works across speech, music, and sound from a general audio perspective; (2) a unified taxonomy of ALM foundations, including model architectures and training objectives; (3) establishment of a research landscape capturing mutual promotion and constraints among different research aspects, aiding in summarizing evaluations, limitations, concerns and promising directions. Our review contributes to helping researchers understand the development of existing technologies and future trends, while also providing valuable references for implementation in practical applications.
arXiv:2607.04777v1 Announce Type: new Abstract: Graph-structured data is increasingly generated and stored in decentralized environments, such as social platforms, mobile applications, and edge networks, where users maintain control over their local graph data. However, collecting and analyzing such decentralized graph data for downstream learning tasks raises significant privacy concerns, as nodes and their attributes often contain sensitive personal information. Local Differential Privacy (LDP) has emerged as a promising solution for privacy-preserving data collection without relying on trusted servers. Nevertheless, existing LDP-based graph learning methods typically assume uniform privacy requirements across users, ignoring the heterogeneous and personalized privacy preferences commonly observed in real-world systems. This uniform treatment leads to inflexible noise injection at the data collection stage, resulting in substantial distortion of graph data and degraded utility in subsequent analysis. To address this limitation, we propose PPGNN, a personalized differentially private framework for decentralized graph data. PPGNN enables user-specific privacy budgets during local perturbation while preserving analytical utility. To handle heterogeneous privacy levels and noise distortion, we design a two-stage solution consisting of a Personalized Perturbation Mechanism (PPM) and a weighted calibration strategy, FlexProp. Extensive experiments on six real-world graph datasets demonstrate that PPGNN effectively balances personalized privacy protection and data utility in decentralized graph learning scenarios.
arXiv:2601.15485v3 Announce Type: replace Abstract: Federal research funding shapes the direction, diversity, and impact of the US scientific enterprise. Large language models (LLMs) are rapidly diffusing into scientific practice, holding substantial promise while raising widespread concerns. Despite growing attention to AI use in scientific writing and evaluation, little is known about how the rise of LLMs is reshaping the public funding landscape. Here, we examine LLM involvement at key stages of the federal funding pipeline by combining two complementary data sources: confidential National Science Foundation (NSF) and National Institutes of Health (NIH) proposal submissions from two large US R1 universities, including funded, unfunded, and pending proposals, and the full population of publicly released NSF and NIH awards. We find that LLM use rises sharply beginning in 2023 and exhibits a bimodal distribution, indicating a clear split between minimal and substantive use. Across both private submissions and public awards, higher LLM involvement is consistently associated with lower semantic distinctiveness, positioning projects closer to recently funded work within the same agency. The consequences of this shift are agency-dependent. LLM use is positively associated with proposal success and higher early-stage publication output at NIH, whereas no comparable associations are observed at NSF. Notably, the productivity gains at NIH are concentrated in non-hit papers rather than the most highly cited work. Together, these findings provide large-scale evidence that the rise of LLMs is reshaping how scientific ideas are positioned, selected, and translated into publicly funded research, with implications for portfolio governance, research diversity, and the long-run impact of science.
arXiv:2607.04779v1 Announce Type: new Abstract: Reasoning segmentation aims to predict pixel-wise masks for targets given complex language queries. Existing approaches leverage Multimodal Large Language Models (MLLMs) for vision-language reasoning and generate intermediate target cues (e.g., points or boxes) to guide a segmentation model. However, compressing rich reasoning into sparse cues often introduces ambiguity and noise, preventing these cues from accurately preserving the reasoning intent. While multiple complementary cues can enrich target information, existing methods typically feed them jointly into a single segmentation process, allowing ambiguous or erroneous cues to affect the entire prediction. Therefore, we propose DGSeg, a reasoning segmentation framework that learns to fuse predictions guided by semantic and spatial cues. Specifically, the MLLM jointly reasons about both target identity and spatial location, producing complementary semantic and spatial cues that are fed into separate segmentation branches. Their predictions are adaptively integrated by a lightweight dynamic gating module trained with relative branch-quality supervision to suppress noisy or conflicting regions. Extensive experiments demonstrate that DGSeg consistently outperforms strong baselines on multiple benchmarks and achieves 69.6% and 67.3% gIoU on the challenging ReasonSeg validation and test splits. Code is available at https://github.com/RZZeng/DGSeg.
arXiv:2504.08791v3 Announce Type: replace Abstract: On-device inference offers privacy, offline use, and instant response, but consumer hardware restricts large language models (LLMs) to low throughput and capability. To overcome this challenge, we present prima.cpp, a distributed on-device inference system that runs 30-70B LLMs on consumer home clusters with mixed CPUs/GPUs, insufficient RAM/VRAM, slow disks, Wi-Fi links, and heterogeneous OSs. We introduce pipelined-ring parallelism (PRP) to overlap disk I/O with compute and communication, and address the prefetch-release conflict in mmap-based offloading. We further propose Halda, a heterogeneity-aware scheduler that co-optimizes per-device CPU/GPU workloads and device selection under RAM/VRAM constraints. On four consumer home devices, a 70B model reaches 674 ms/token TPOT with <6% memory pressure, and a 32B model with speculative decoding achieves 26 tokens/s. Compared with llama.cpp, exo, and dllama, our proposed prima.cpp achieves 5-17 times lower TPOT, supports fine-grained model sizes from 8B to 70B, ensures broader cross-OS and quantization compatibility, and remains OOM-free, while also being Wi-Fi tolerant, privacy-preserving, and hardware-independent. The code is available at https://github.com/OpenCPIL/prima.cpp.
arXiv:2607.02373v2 Announce Type: replace Abstract: High-brilliance sources of polarized gamma rays are widely sought after to pump and probe matter at subatomic length scales. However, existing accelerator facilities and optical lasers cannot reach a sufficiently high center-of-mass energy to produce polarized, multi-GeV gamma rays from unpolarized electrons via inverse Compton scattering. Here we propose a scheme where the optical laser photons are first "accelerated" to the extreme ultraviolet in a beam-driven plasma wakefield, then reflected by a plasma mirror back onto a trailing electron beam, producing a flash of gamma rays. Numerical simulations demonstrate this light source can achieve a high peak-brilliance (10^25 photons/s mm^2 mrad^2 0.1% BW) and a high degree of circular (95 %) or linear (77 %) polarization at multi-GeV photon energies, paving the way for the production of spin-polarized positrons and tests of light-by-light scattering.
arXiv:2404.06294v2 Announce Type: replace-cross Abstract: Super-Resolution (SR) is a time-hallowed image processing problem that aims to improve the quality of a Low-Resolution (LR) sample up to the standard of its High-Resolution (HR) counterpart. We aim to address this by introducing Super-Resolution Generator (SuRGe), a fully-convolutional Generative Adversarial Network (GAN)-based architecture for SR. We show that distinct convolutional features obtained at increasing depths of a GAN generator can be optimally combined by a set of learnable convex weights to improve the quality of generated SR samples. In the process, we employ the Jensen-Shannon and the Gromov-Wasserstein losses respectively between the SR-HR and LR-SR pairs of distributions to further aid the generator of SuRGe to better exploit the available information in an attempt to improve SR. Moreover, we train the discriminator of SuRGe with the Wasserstein loss with gradient penalty, to primarily prevent mode collapse. The proposed SuRGe, as an end-to-end GAN workflow tailor-made for super-resolution, offers improved performance while maintaining low inference time. The efficacy of SuRGe is substantiated by its superior performance compared to 28 state-of-the-art contenders on 10 benchmark datasets.