arXiv:2607.17287v1 Announce Type: new
Abstract: The suffix array ($\SA$) and inverse suffix array ($\ISA$) are fundamental data structures in string algorithms. Given a text of length $n$ over an integer alphabet of size $\sigma$, explicitly storing these structures requires $\Theta(n\log n)$ bits, motivating compressed representations that approach the information-theoretic minimum of $n\log\sigma$ bits. Two landmark solutions are the FM-index, which achieves succinct space and supports $\SA$ and $\ISA$ queries in roughly logarithmic time, and the Compressed Suffix Array (CSA), which uses $O(n\log\sigma)$ bits and answers both queries in $O(\log_\sigma^\epsilon n)$ time. Nearly all subsequent compressed indexes, including entropy-compressed and repetitiveness-aware variants, support $\SA$ and $\ISA$ using essentially the same techniques, leading to similar space--time trade-offs. This raises a fundamental question: under the same asymptotic space bound, do $\SA$ and $\ISA$ have the same inherent query complexity? We provide evidence that this symmetry is not intrinsic. Under an $O(n\log\sigma)$-bit space bound, we show that $\SA$ queries are unlikely to admit $\log^{o(1)} n$ query time, suggesting that improving the classical CSA bound would require a major breakthrough. In contrast, we present a new encoding supporting $\ISA$ queries in $O(\tau+\log\log n/\log\log\sigma)$ time using $(1+1/\tau+o(1))n\log\sigma$ bits, for any $\tau\ge1$. Our main insight is that the $\LF$-mapping, traditionally implemented using the Burrows--Wheeler Transform, can be adapted to support $\ISA$ queries directly from the packed text, eliminating the need to store the BWT while achieving substantially faster query times.
Science Journals
arXiv:2607.17673v1 Announce Type: new
Abstract: Contrastive learning is increasingly moving toward settings with three or more modalities instead of image-text pairs. Yet, extending models from pairwise to higher-order multimodal alignment can introduce optimization and representation challenges. We identify encoder Jacobian conditioning as a key factor in trimodal contrastive learning: poorly conditioned encoders exhibit collapsing or amplified singular-value spectra, leading to exploding Jacobian condition numbers and degraded multimodal alignment. We introduce geometry-preserving encoders (GPEs) by directly conditioning the Jacobian through regularization and demonstrating that simple modifications like LeakyReLU activations and residual paths recover these geometric benefits. Across a synthetic benchmark and four real-world datasets including missing modalities, improving Jacobian conditioning boosts retrieval and linear probe performance across multiple contrastive objectives, whereas expressive objectives yield little benefit in linear probes. More broadly, our results show that multimodal contrastive learning depends not only on objective expressivity, but also on the geometric and optimization properties of the underlying encoders.
arXiv:2607.17674v1 Announce Type: new
Abstract: A language model $p_\theta(y \mid x)$ trained on reasoning tasks learns to solve problems via multiple distinct strategies, yet these strategies are implicit and entangled within the model's response distribution. We study the problem of decomposing the response distribution of a given pretrained language model into a structured, strategy-conditioned representation. Specifically, we learn a latent-variable factorization $p_\theta(y \mid x) \leadsto (r_\phi(z \mid x), g_\phi(y \mid x,z))$, where a router $r$ maps each input to a distribution over latent strategies $z$ and a generator $g$ produces the response conditioned on that strategy. A key challenge is that the generator, initialized from the base model, already represents $p_\theta(y \mid x)$ without using $z$. Standard variational inference therefore gives the model no incentive to route information through $z$ and can yield a severe form of posterior collapse. To address this, we propose a variational objective that measures fractional information gain relative to the base model's response loss and concentrates reconstruction pressure on tokens with high base model surprisal, encouraging $z$ to encode strategy-relevant response variation. We introduce a benchmark of multi-strategy algorithmic tasks and show that this objective recovers latent codes aligned with distinct reference strategies while preserving the base model's response distribution.
arXiv:2607.17675v1 Announce Type: new
Abstract: Current video generation models achieve impressive results in single-shot generation, yet remain limited in cinematic video generation, where coherent narratives and effective multi-shot composition require explicit shot planning. To address this challenge, we propose ShotPlan, a framework for explicit multi-shot cinematic video generation built upon a video diffusion foundation model. Our method introduces learnable planning tokens that capture shot-level transition cues and can be seamlessly integrated with the original video generation tokens to control transition timestamps. Unlike standard video generation tokens, the proposed planning tokens are equipped with Fractional Temporal Rotary Position Embedding (FRoPE), enabling shot transitions to be modeled at the frame level. Experiments demonstrate that ShotPlan significantly outperforms existing cinematic video generation methods, offering more flexible shot management and stronger inter-shot consistency.
arXiv:2607.17686v1 Announce Type: new
Abstract: Modern software teams have mature tools for low-level testing, such as pytest, JUnit, and Jest, which make it inexpensive to write unit tests and run them on every commit. Systems engineering, in parallel, has developed rigorous principles for design verification and validation (V&V), which has worked very well across engineering discipline to align user expecations and requirements with developers' deliverables. In practice, however, the two rarely connect, and the link between users' high-level requirements and the low-level tests that machines actually run is maintained by hand, if at all. This gap is increasingly costly for AI-enabled and cyber-physical systems, for which regulators now ask for traceable evidence that high-level requirements are met, while raw test results provide little of the structure such evidence requires. We introduce VNVSpec, an open-source framework that makes V&V specifications machine-readable and executable. With this framework, users state high-level requirements directly or import them from catalogs derived from published standards. Then, the framework checks requirement quality, supports decomposition into module-level requirements with explicit metrics and acceptance criteria, links these requirements to test results through a traceability graph, and compiles the collected evidence into verdicts and audit-ready reports. We evaluate the framework by self-application, in which it is continuously assessed in CI against its own specification of 36 requirements verified by 449 tests, completed within limited time which scales linearly and thus can handle up to 10,000 requirements. We also discuss how the framework extends to testing black-box AI models and AI coding agents. The framework, its full test suite, the catalogs, and the benchmark scripts are available at https://github.com/ai-vnv/vnvspec.
arXiv:2607.17689v1 Announce Type: new
Abstract: Multilingual language models rely on shared subword vocabularies to represent multiple languages within a limited number of token units. While such sharing is often useful, it can also create cases in which identical surface forms are treated too uniformly across languages, even when their meanings or usage differ. We investigate this limitation through cross-lingual homographs and false friends, and examine whether introducing language information earlier in the tokenization process can improve their treatment. We propose a simple tokenizer-level intervention based on language cues: language-specific characters replacing initial characters of shared-vocabulary words, reducing common identity during vocabulary construction. In intrinsic analysis, we find through tokenizer-level statistics that BPE and UnigramLM often treat cross-lingual homographs in a largely language agnostic way, whereas the context-sensitive SaGe tokenizer diverges more strongly; our intervention removes this gap. In downstream English-to-X machine translation, our cues yield modest improvements in several settings, especially under BPE, although the effect is not consistent across all languages and evaluation sets. Overall, the findings suggest that adding lightweight language information at the tokenizer level is a promising direction for further exploration.
arXiv:2607.17693v1 Announce Type: new
Abstract: Test-time adaptation (TTA) aims to mitigate distribution shifts by adapting models with unlabeled target data at inference time. While TTA with vision-language models (VLMs) has shown promising results in classification, extending it to medical image segmentation remains challenging. In this setting, the adaptation gains from optimizing on VLM-generated predictions are often outweighed by the degradation to the VLM's strong pretrained features caused by noisy, update-driven learning, resulting in limited and unstable improvements. We therefore propose Memory-Supported Synergistic Adaptation (MSSA), a novel training-free TTA framework for medical image segmentation. Without updating model parameters, MSSA dynamically selects reliable image-text predictions to construct an online memory, uses them as text-guided semantic priors, and couples them with cross-image structural alignment for robust adaptation. Specifically, MSSA consists of (i) a noise-aware memory construction module that filters and stabilizes cross-modal predictions, and (ii) a relevance-driven prototype alignment module that aligns the target sample with structurally consistent memory samples and their reliable predictions to improve adaptation. Extensive experiments on multiple medical segmentation benchmarks demonstrate that MSSA consistently improves VLM-based segmentation models and outperforms existing fine-tuning-based TTA methods by a clear margin, with gains of up to 12.2% DSC and 11.7% mIoU. Project page: https://lingrayy.github.io/MSSA/ .
arXiv:2607.17698v1 Announce Type: new
Abstract: Optical phased arrays (OPAs) are key components for solid-state beam steering in emerging photonic technologies such as LiDAR, optical communication, and adaptive optics. However, conventional integrated OPA designs face trade-offs between bandwidth, steering range, and fabrication complexity. Here we designed and numerically analyzed a novel three-dimensional end-firing antenna array compatible with fabrication using two-photon polymerization (2PP) directly on a thin-film lithium niobate (TFLN) platform. By elevating the polymer antennas above the chip surface, the design enables two-dimensional beam steering while maintaining the broadband advantages of end-fire emission. Full-wave electromagnetic simulations demonstrate transmission efficiencies up to 89.5\% over the \SIrange{1.4}{1.6}{\micro\meter} wavelength range, achieving a field of view of \ang{24.9} $\times$ \ang{22.8} with beamwidths of approximately \ang{1.7}. The architecture's compatibility with electro-optic phase control and advanced array configurations suggests significant potential for high-speed, low-loss beam steering systems. This work establishes a foundation for scalable 3D photonic phased arrays that bridge integrated optics with free-space beam manipulation.
arXiv:2607.17700v1 Announce Type: new
Abstract: Threshold, aggregate, and multi-signatures -- which we collectively call quorum signatures -- certify that a quorum of nodes endorsed a statement, with a certificate as small as a single signature. No constant-size post-quantum quorum signature is known: all candidates grow with the number of signers and are slow to aggregate, making quorum signatures the hardest obstacle to migrating byzantine fault-tolerant systems to post-quantum security.
In this paper, we sidestep this open cryptographic problem by changing how the protocol communicates. We introduce a primitive we call Distributed Quorum Signature (DQS), built solely from ordinary digital signatures and a Bracha-style approval broadcast. DQS turns certificates from network messages into local events. Two event types divide the roles certificates play: weak certificates capture safety, strong certificates capture liveness. In DQS every message is constant size, fitting a single datagram regardless of the number of nodes. The total communication is quadratic, and no security assumptions change. In a large distributed system, the overhead of post-quantum DQS is competitive with the canonical pre-quantum BLS scheme.
arXiv:2607.17701v1 Announce Type: new
Abstract: Proactive agents are expected to anticipate user needs and provide autonomous assistance by perceiving environmental context without explicit instructions. A fundamental capability of such agents is to identify and track users' upcoming events, enabling continuous and event-specific assistance. For example, by recording the time and location of a planned hike, an agent can deliver weather reminders in advance or provide navigation support before departure. However, existing works on proactive agents largely overlook event-centric assistance, and the open-ended nature of proactive assistance poses challenges for reliable evaluation.
To bridge these gaps, we introduce ProEvent, the first event-centric benchmark designed to assess an agent's ability to proactively maintain a user's timetable based on ongoing instant messaging chats. ProEvent provides synthesized yet realistic chats that consider the dynamic interaction among users, concurrent chat threads, and noise in the real world, and evaluates proactive agents on response timing, single-step response correctness, and multi-step response correctness. Experiments on eight LLMs and pipelines reveal that current agents frequently overact and struggle with event cancellation. Notably, even GPT-5.1 only reacts correctly in 26.7% of scenarios. Further qualitative analysis reveals fundamental limitations of current LLMs as proactive agents, particularly in detecting implicit events and reasoning from the user's first-person perspective.
arXiv:2607.16294v1 Announce Type: new
Abstract: Paired image-to-image translation underpins a wide range of computer vision tasks, including image editing, sensor translation, and domain adaptation. Bridge matching and flow matching have recently emerged as powerful frameworks, extending diffusion models to arbitrary source and target distributions. However, their standard formulations assume perfectly aligned training pairs, treating all source-target correspondences as equally reliable. In practice, real-world applications often involve weakly aligned pairs due to changes of acquisition conditions, including e.g. asynchronous captures, different illuminations, or misregistration. In this work, we introduce Alignment-Aware Bridge Matching (A${}^2$BM), a bridge matching method that leverages image pairs alignment during training. By incorporating alignment scores, the model learns to disentangle true semantic correspondences from misalignment artifacts. At inference time, we use the alignment score as a control variable over translation fidelity, with strongly aligned outputs obtained when prompting the model with the highest alignment score. We validate A${}^2$BM on both controlled synthetic experiments and on challenging real-world tasks, including cross-sensor super-resolution and pixel-space unsupervised domain adaptation. In all settings, A${}^2$BM consistently improves translation fidelity over strong GAN-, diffusion-, and Schr{\"o}dinger bridge-based baselines, establishing alignment conditioning as a principled solution for image translation models with weakly aligned data.
arXiv:2607.16998v1 Announce Type: new
Abstract: This paper presents a method for user-driven robot Learning from Demonstration (LfD) that reduces user effort while ensuring compliant and precise reproduction. The method eliminates repeated teaching for the same task and enables real-time learning from a single demonstration. Demonstrated motions are reproduced with high precision, while impedance variations are learned in real time to provide both compliance and robustness against perturbations. This mitigates potential safety issues in Human-Robot Interaction (HRI) that arise from conventional time-indexed trajectories lacking compliance. The proposed approach integrates a three-dimensional (3D) Fast Diffeomorphic Matching (FDM) algorithm with a Dynamical System (DS)-based motion generator to achieve real-time single-shot demonstration learning and reproduction. An Extended Kalman Filter (EKF) framework compensates for reproduction errors and recovers from external interactions. Furthermore, an impedance parameterization function is incorporated to learn impedance variations from demonstrations and maintain surface contact for specific applications. The proposed approach is validated through comprehensive experiments on a 7 Degree-of-Freedom (DOF) KUKA LWR IV+ robot.
arXiv:2607.17702v1 Announce Type: new
Abstract: The Reynolds analogy between velocity and temperature fields is a central problem in the statistical theory of compressible wall turbulence. The generalized Reynolds analogy (GRA) previously established by the author accurately describes the mean velocity-temperature relation, but a self-consistent closure for the fluctuating field has remained elusive. In this paper, the instantaneous similarity relation -- that the generalized total enthalpy (minus its wall value) is proportional to the local velocity -- is shown to close both fields at once. The mean field reproduces the GRA solution, while the fluctuating field yields the closed relation $T'_{rms}/u'_{rms} = |a_u^{-1/2}Pr^{-1/4}Pr_m^{-1/2}\,\partial\bar{T}/\partial\bar{u}|$, derived from three universalities: the universal structure constant of the streamwise energy fraction $a_u$ ($\approx1.0$-$1.1$)}, the quasi-equilibrium of production and dissipation, and the Obukhov-Corrsin cutoff-scale universality -- with no free parameters. It recovers exactly the refined strong Reynolds analogy (RSRA) of Huang et al.~(2025) -- the most accurate benchmark to date -- recasting it from empirical relation to a consequence of universality principles and explaining its fitted coefficient $1.09$ as $a_u^{-1/2}Pr^{-1/4}$ (for $Pr=0.71$, $a_u=1.0$), and is validated by boundary-layer direct numerical simulation (DNS) data across Mach numbers and wall thermal conditions, collapsing to unity.
arXiv:2607.17704v1 Announce Type: new
Abstract: Much effort is put into helping students at different educational levels develop Computational Thinking (CT) skills. Self-efficacy is important for skill development. It can predict perseverance, engagement and success on educational tasks. We created an instrument to measure self-efficacy of students in higher education for the CT skills abstraction, algorithmic thinking, decomposition, evaluation and generalization. First, 91 candidate items were created by including, adapting and extending items found in the literature. These items were evaluated by experts in the field of CT and education. 54 items remained and to reduce the number of items further, data was collected from 270 students in higher education recruited both through Prolific and a university setting in Costa Rica. Through principle component analysis (PCA) using a subset of 200 responses, the number of items was reduced to 27. Confirmatory factor analysis (CFA) using the remaining responses in the dataset showed the items can be divided into two categories: (1) creating the solution and (2) evaluating the solution. The created instrument can be valuable when assessing CT self-efficacy of students in higher education. With additional validation (e.g. examination of test-retest validity), we believe the scale could be used to evaluate the effectiveness of interventions, or decide what interventions should be provided to foster CT skill development.
arXiv:2607.17706v1 Announce Type: new
Abstract: Half-harmonic generation in microresonators with mixed $\chi^{(2)}$ and $\chi^{(3)}$ nonlinearities provides access to a rich landscape of two-colour soliton states and frequency combs. We numerically demonstrate the formation of both bright and topological two-colour solitons at relatively modest pump powers and phase-mismatch parameters. At higher powers and larger phase mismatches, we identify a regime where parametric down-conversion assists the formation of dark solitons in the normally dispersive pump field. While the dark-soliton core is primarily shaped by the Kerr nonlinearity, nonlinear coupling between the pump and its half-harmonic modes enables soliton excitation using a standard laser-frequency scanning technique, a capability not normally associated with dark solitons. The simulations are performed using parameters representative of lithium-niobate microresonators pumped near 775 nm and generating a half-harmonic near 1550 nm. These results reveal how the interplay of $\chi^{(2)}$ and $\chi^{(3)}$ nonlinearities can substantially expand the range and types of experimentally accessible soliton states in a single integrated frequency-comb platform.
arXiv:2607.17707v1 Announce Type: new
Abstract: Since the first conference in Marseille in 1982, the International Conference on Logic Programming (ICLP) has been the premier international event for presenting research in logic programming. These proceedings include the Technical Communications of the 42nd ICLP, held 20-23 July 2026 as part of FLoC 2026 in Lisbon, Portugal. The papers and extended abstracts in this volume address the following areas and topics: theoretical foundations, language design and programming methodologies, program analysis and optimization, applications and implementation methodologies. This volume features contributions to two submission tracks of ICLP 2026: the Main track and the Recently Published Research track.
arXiv:2607.17709v1 Announce Type: new
Abstract: In this work, we propose a novel framework for the functional controllability of the ant swarm model, a well-known and relevant model of collective behaviour. Our approach introduces a population of controlling stigmergic agents, trained via Reinforcement Learning (RL), that act on the environment to influence the system dynamics and promote the emergence of ordered behaviour. Stigmergic agents are optimized in a centralized-training decentralized-execution setting, interacting with ants only through the shared pheromone field. The reward design promotes trail pheromone structures and alignment of ant positions with high-pheromone paths, without requiring control of specific microscopic configurations. Our results demonstrate that the learned policies effectively shift the phase transition line that characterizes the global behaviour of the system, enabling the emergence of trails scenarios in regimes that are typically dominated by randomness. This study provides insights into the potential of RL based control strategies for complex systems, contributing to the general understanding of functional controllability in this field.
arXiv:2607.17715v1 Announce Type: new
Abstract: Long-context inference is central to modern large language model (LLM) applications such as retrieval-augmented generation and multi-document reasoning. To mitigate the growing inference cost, recent work has explored key-value (KV) cache reuse to reduce redundant prefill computation. However, existing reuse methods primarily focus on computation savings and overlook a critical bottleneck in long-context LLM serving: the cost of storing and accessing large KV caches. While KV compression appears to be a natural complement, naively combining compression with non-prefix KV reuse often leads to severe accuracy degradation. In this work, we propose C$^2$KV, a unified framework for non-prefix KV reuse that jointly optimizes KV extraction and inference-time concatenation. C$^2$KV learns a composable and compressed KV cache manifold that is explicitly designed to be position-agnostic. Our approach introduces a lightweight sidecar Extractor with learnable compression tokens and a structured attention flow, enabling modular KV representations that can be flexibly reused and concatenated without modifying the frozen base model. We further employ a compression-concatenation co-training strategy to align extraction-time representations with their downstream reuse behavior. Extensive experiments across multiple long-context benchmarks and model families demonstrate that C$^2$KV significantly reduces KV cache storage and transfer costs, achieving up to 17$\times$ inference speedup under long contexts, while preserving generation quality.
arXiv:2607.17722v1 Announce Type: new
Abstract: The Linux kernel is one of the most complex software systems, where automated fuzzing continuously exposes thousands of crashes, yet root-cause diagnosis remains a manual and time-consuming bottleneck. Existing LLM-based root cause analysis (RCA) techniques, effective for distributed systems, do not readily generalize to kernel debugging due to sparse low-level artifacts, heterogeneous diagnostic evidence (e.g., syscalls, logs, and crash reports), and complex non-linear fault propagation that demands fine-grained method-level reasoning. To address these challenges, we propose KernelDiag, an agent-based framework for kernel root-cause diagnosis via structured causal reasoning. KernelDiag first aligns heterogeneous diagnostic artifacts through log-to-code mapping, and then employs artifact-specialized agents to iteratively reason over source-level program semantics and crash-specific kernel configurations. The inferred causal dependencies are incrementally organized into structured Evidence Graphs, enabling accurate faulty-method localization and causal explanations. We evaluate KernelDiag on the real-world KGYM benchmark. KernelDiag consistently outperforms state-of-the-art localization approaches at both file and method levels, achieving significant improvements in Top@k accuracy, including up to 4x and 2x gains in challenging settings without explicit hints. Furthermore, both human and LLM-assisted evaluations show that KernelDiag generates accurate, coherent, and actionable diagnostic explanations. Overall, this work lays the foundation for automated kernel root-cause diagnosis by bridging low-level diagnostic evidence with source-level causal reasoning.
arXiv:2607.17723v1 Announce Type: new
Abstract: Nonlinearly preconditioned inexact Newton methods form an effective class of solvers for large-scale nonlinear algebraic systems arising from the discretization of partial differential equations. A central challenge in nonlinear elimination (NE) preconditioning is the reliable identification of the slowly converging components to be eliminated. Existing selection strategies often rely on problem-specific physical information or user-tuned thresholds applied directly to the raw nonlinear residual, which may contain irregular oscillatory structures near stagnation regions, making the selected bad subset highly sensitive to threshold parameters. In this work, we propose an online-learning-enhanced NE preconditioner that identifies the bad subset from the dominant structure of the nonlinear residual rather than from the raw residual itself. Residual snapshots are collected online during the stagnation phase of the current Newton solve, and an unsupervised extraction model is trained to capture the principal nonlinear imbalance. We consider both a linear extractor based on principal component analysis and nonlinear extractors based on autoencoder neural networks. Moreover, we integrate the approach into a parallel domain decomposition framework, which trains a local extraction model independently on each subdomain. The learned residual reconstruction is then used to define the bad subset and guide the nonlinear elimination process. Numerical experiments on lid-driven cavity flows at Reynolds numbers up to 10,000 show that the proposed method produces more reliable and coherent bad subsets, is robust with respect to both NE and learning parameters, and outperforms the baseline NE preconditioner in terms of the convergence.
arXiv:2607.17724v1 Announce Type: new
Abstract: Coordinating payload transfers between subsystems is a critical challenge in lifelong Multi-Agent Pickup and Delivery (MAPD). We study systems where agents are confined to separate regions and must exchange payloads through shared handover stations. These stations, equipped with single docks and finite buffers, are inherently vulnerable to blocking and starvation. We formalize this problem as Multi-Subsystem MAPD with Buffer-limited Handover Stations (MS-MAPD-BHS). We then propose Handover-Aware Reservation and Routing (HARR), an online controller that couples per-subsystem planners. HARR uses a shared dock reservation calendar and a deterministic rolling-horizon projection of buffer occupancy to coordinate actions. A candidate route is accepted only if its dock interval is free and the resulting buffer occupancy projection remains within capacity. Under perfect execution, these checks ensure collision-free dock use and buffer-safe committed operations within the reservation horizon. In simulation, HARR achieves up to 77% higher throughput and 92% lower backlog than a fixed-dock ablation at moderate load, while also reducing planning time relative to a coupled station-aware Token Passing baseline. These results show that explicit interface coordination substantially improves stability in modular multi-subsystem transport.
arXiv:2607.17725v1 Announce Type: new
Abstract: The intricate motion arising from fluid--boundary interactions is visually compelling, yet notoriously difficult and computationally expensive to simulate in the presence of complex boundaries. Accurately resolving boundary geometry requires body-fitted grids constructed via cut-cell methods, which often leads to poorly conditioned linear systems in the pressure projection stage and, consequently, prohibitive computational cost. We present FastVEM, an efficient boundary-conforming fluid simulation framework that enables high-fidelity flow--boundary interaction at substantially reduced cost. Computational efficiency is achieved through a coordinated, top-down design spanning numerical discretization, grid construction, and linear solvers. FastVEM adopts a Virtual Element Method (VEM) discretization to robustly enforce incompressibility and boundary conditions on irregular body-fitted grids, and employs a VEM polynomial-space Particle-in-Cell scheme for advection. Complementing this discretization, a convexity-preserving cut-cell strategy is introduced to construct simulation-friendly body-fitted grids. To accelerate pressure projection, we develop a Galerkin geometric multigrid solver featuring a diffusion-free prolongation operator that prevents coarse-level matrix densification, along with a nested, boundary-aware grid hierarchy that ensures well-posed placement of coarse-level degrees of freedom. Compared to prior cut-cell--based fluid simulators, FastVEM speeds up the computationally dominant pressure projection stage by up to 100x, while robustly handling even more challenging boundary geometries.
arXiv:2607.17733v1 Announce Type: new
Abstract: 4-bit quantization enables efficient LLM inference, but suffers from significant accuracy degradation due to outliers. Prior work addresses this problem via data rotation or mixed-precision integer quantization, but often relies on software-managed scaling and frequent dequantization, incurring substantial overhead. Microscaling formats, such as MXINT, eliminate these inefficiencies by encoding scales in hardware, yet remain incompatible with rotation-based methods. Our analysis reveals that outliers vary in severity, from rare extremes to frequent mild deviations, and that quantization sensitivity is unevenly distributed across layers and columns. These insights motivate a fine-grained, sensitivity-guided approach. We introduce MXSens, a training-free method that assigns mixed mantissa bitwidths (4/6/8) based on column- and layer-wise sensitivity, naturally leveraging the block-wise structure of MXINT. MXSens outperforms state-of-the-art quantization methods across a range of models and tasks. Under the W4A4KV4 setting, MXSens achieves perplexities of 3.77 and 7.63 on LLaMA-2-70B and LLaMA-3-8B, respectively, substantially improving over existing baselines on WikiText-2. Our work establishes a new balance between accuracy and resource efficiency for LLM quantization.
arXiv:2607.17735v1 Announce Type: new
Abstract: The skew-symmetric discretisation has recently been proposed as a new robust simulation method for weakly approximating stochastic differential equations (SDEs) with non-globally Lipschitz drift. This work develops a pathwise version of the scheme by representing the noise increment as a skew-normal distribution and coupling it with the driving Brownian increments, thereby enabling its use in the multilevel Monte Carlo (MLMC) framework. Under suitable conditions, we establish strong convergence of order 1/2 in $L^2$. Subsequently, the associated MLMC estimator is shown to have computational complexity $\mathcal{O} \bigl(\varepsilon^{-2} (\log (1/\varepsilon))^2 \bigr)$ to achieve a mean-squared error $\varepsilon^2$. We then analytically compare the proposed scheme with the tamed Euler scheme, another benchmark for robust discretisation. Under a strong inward-drift regime with the current state being far from the stable region, we show that the probability of moving in the wrong direction tends to vanish in the skew-symmetric scheme, whereas the tamed Euler scheme makes such moves with a non-trivial probability. Furthermore, in the MLMC setting, employing a one-dimensional stochastic Ginzburg-Landau model, we specify the range of step sizes for which the asymptotic variance of the coupled level difference obtained via the pathwise skew-symmetric scheme is lower than that obtained via the tamed Euler scheme. Numerical experiments on several model examples support the theoretical rate of strong convergence and demonstrate the stability and effectiveness of the resulting MLMC in the superlinear drift setting.
arXiv:2607.16244v1 Announce Type: new
Abstract: Training multi-turn evidence-reading agents with outcome-only reinforcement learning is unstable because intermediate turns receive little direct credit. In HotpotQA experiments with Qwen2.5-3B-Instruct, GRPO initially improves (standard F1 0.430) but subsequently collapses to 100% format-violating outputs. Training-log diagnosis reveals a zero-advantage lock-in mechanism: all sampled trajectories receive the minimum format penalty (-2.0), group-relative advantages vanish, and the policy-gradient loss becomes zero--an optimization deadlock. We propose a variance-injection strategy: by assigning per-turn rewards to intermediate evidence-reading turns, we prevent the group reward distribution from collapsing to a single value--preserving the variation that GRPO's group-relative advantage requires. Contextual Information-Gain Policy Optimization (CIGPO) implements this strategy using the marginal increase in the frozen reference model's log-likelihood of the ground-truth answer as the per-turn signal. With separate normalization of IG and F1 rewards and an IG-weight curriculum, CIGPO reaches a standard F1 of 0.518 on HotpotQA at the 3B scale (from 0.252 base; +105%), compared with 0.430 for the best GRPO checkpoint and 0.000 for the final GRPO checkpoint. CIGPO maintains meaningful reward variance and avoids zero-advantage lock-in throughout training. These results identify reward-variance collapse as a concrete failure mode of outcome-only GRPO and show that turn-level IG rewards can prevent it in this HotpotQA setting.