arXiv:2605.01224v2 Announce Type: replace
Abstract: This paper argues that contemporary multilingual NLP has converged on a fragile and misleading paradigm of incidental multilingualism. Today's LLMs appear multilingual largely because they are trained on massive, uneven web corpora, not because multilingual or multicultural competence has been treated as a core design objective. We contend that this paradigm systematically produces unequal, brittle, and opaque behavior across languages, with severe consequences in real-world and agentic deployments where models must reason, plan, and act across multiple linguistic contexts. We report a focused empirical study of two practical questions: which languages models self-report as supported and which languages they actually respond in across multilingual prompts. We additionally demonstrate how even a simple language-change attack can surface these failures and expose hidden assumptions about language in LLM-based systems. To address this, we call for a shift toward multilingualism by design: a research agenda that treats equitable multilingual performance, cultural grounding, and cross-lingual behavioral understanding as first-class goals in all aspects of the model pipeline.
Science Journals
arXiv:2605.01395v4 Announce Type: replace
Abstract: In this paper, we design feedback control laws for soft robots modelled using the Cosserat rod theory, which is spatially discretised using the Piecewise Constant Strain (PCS) approach. The PCS approach approximates the nonlinear PDEs describing the Cosserat rod by a finite-dimensional system of nonlinear ODEs. This simplification results in a model describing soft robots which is similar to the serial rigid-link manipulators. We design feedback control laws for the quasi-static PCS model by using external wrenches as control inputs. The control laws are designed based on feedback linearisation in strain and task spaces. An extensive set of numerical results demonstrates the performance of the control laws for end-effector trajectory tracking and shape control of soft robots.
arXiv:2605.01598v3 Announce Type: replace
Abstract: We present a theoretical model of the ionization-recombination balance in the ultracold Rydberg gas-plasma mixture, which is caused by the collective processes rather than by individual interparticle interactions. We consider the electron-ion system where each electron moves for the most part of its time in the effective centrifugal potential formed by the nearest ion and sometimes jumps to the neighboring potential wells due to the perturbations from remote particles. These perturbations are described by a thermal bath with the effective "virial" temperature. Then, the electron with a sufficiently high energy, which can escape from the local potential well, should be considered as free (conducting) one; while the electron with low energy, confined within the well, as forming the Rydberg atom. As follows from our calculations, there is a sharp crossover from the insulating phase (Rydberg gas) to the conducting one (plasma) with increase in the particle density, which somewhat resembles the Mott transition in condensed-matter physics. This model should be well relevant to the steady-state ultracold plasmas obtained in the most recent experiments and represents a promising direction for further research.
arXiv:2605.02566v3 Announce Type: replace
Abstract: Artificial intelligence now produces convincing-looking scientific judgment (reviews, rankings, attributions, verifications) at almost no marginal cost. An influential reading of AI economics holds that prediction becomes cheap while human judgment stays scarce. For science, that reading understates the problem: what has become cheap is a counterfeit of judgment itself. This matters most for institutions whose product is trusted judgment, which is what journals, universities, funders, and learned societies exist to manufacture. They do not merely adapt to the technology; they compete with it for the same functional role. Four things become scarce instead: verified signal, legitimacy, authentic provenance, and integration capacity. Integration capacity means how much AI-delegated judgment a scientific community will accept before it stops trusting the journals, panels, and conferences that admitted it. It is the least developed of the four and the most binding: better tooling cannot buy it. The way forward for AI-augmented science is not acceleration but the redesign of its certifying infrastructure around these new scarcities.
arXiv:2605.03646v5 Announce Type: replace
Abstract: Turbophoresis in inhomogeneous turbulent flows leads to the formation of large-scale nonuniform particle number density distributions of inertial particles. This effect is associated with an effective drift velocity directed toward regions of lower turbulence intensity. It depends on the Stokes and Reynolds numbers, as well as on the gradient of the turbulence intensity. In the present study, turbophoretic transport is experimentally investigated in air turbulent flows generated by one-grid and two-grid oscillating systems. The flow velocity field is measured using Particle Image Velocimetry, and the particle spatial distribution is obtained by applying image processing techniques. To isolate the effect of particle accumulation due to turbophoresis from that associated with mean fluid flow, the measured particle number density of inertial particles is normalized by the corresponding distribution obtained for noninertial tracer particles under identical flow conditions. The measurements show preferential accumulation of inertial particles in regions of minimum mean-square turbulent velocity, consistent with the expected behavior of turbophoretic transport.
arXiv:2605.05061v2 Announce Type: replace
Abstract: Snow slab avalanches are among the most dangerous hazards in mountain regions. Recent numerical, field, and laboratory studies have renewed interest in shear-failure interpretations of avalanche release, particularly in relation to dynamic crack propagation and supershear fracture. Yet most analytical models either idealize the weak layer as perfectly brittle or neglect its pre-peak elasticity, although post-peak dissipation and compliance both control stress redistribution and critical length. Here, we derive an analytical solution for shear-failure propagation beneath an elastic snow slab with finite linear softening. Building on the weak-spot model of Gaume et al. (2013), failure is described by a fully softened residual core, a fracture process zone, and an intact elastic region. The solution recovers the classical brittle length as softening vanishes, distinguishes the residual crack length from the total affected length, and links weak-spot and fracture-energy descriptions through the softening law. Depth-averaged Material Point Method simulations confirm the analytical stress and displacement profiles and the predicted characteristic lengths. We then extend the same compliant-softening framework to collapse-driven anticrack propagation. A simplified Timoshenko anticrack analogue shows that slab bending and transverse shear deformation amplify normal stress at the weak-layer front and introduce a bending-controlled softening length with an approximately fourth-root dependence on softening displacement, supported by three-dimensional MPM simulations. Finally, a mixed-mode Timoshenko formulation couples weak-layer compression, slope-parallel shear, and slab rotation. A compact sharp-front model and a fully coupled finite-softening model reproduce the observed slope-angle dependence of critical cut length using realistic elastic and failure-envelope parameters.
arXiv:2605.07032v3 Announce Type: replace
Abstract: The evolution of generative models from next-token predictors to autonomous engines of complex systems necessitates rigorous safety hardening. Adversarial jailbreaking, the strategic manipulation of models to elicit harmful output, remains a primary threat to safe deployment. While Reinforcement Learning (RL) frames jailbreaking as a multi-step attack through sequential optimization, a mechanistic understanding of why the framework succeeds remains incomplete. To fill this gap, we present the first systematic decomposition of RL jailbreaking. We deconstruct the framework into problem formalization (reward function, action space, episode length), and algorithmic measures (RL algorithm, training data, reward-shaping) to identify the structural determinants of adversarial success. Our results reveal that the RL-jailbreaker successfully compromised all targeted models and safeguards. Through this first-of-its-kind analysis, we demonstrate that environment formalization, specifically dense rewards and extended episode lengths, is the primary driver of jailbreaking success. This work provides a tool for improving RL-jailbreaker efficiency and, ultimately, harden generative models resistant to RL-based attacks.
arXiv:2605.08601v5 Announce Type: replace
Abstract: Many applications process a stream of tuples over a window duration, and require the results within a specified deadline after the end of the window. For such scenarios, processing tuples intermittently (in batches) instead of eagerly processing tuples as they arrive significantly reduces the overall cost. Earlier work on intermittent query processing has addressed only fixed environments. In this paper, we propose scheduling schemes for batched processing of tuples, in an elastic parallel environment, scaling nodes up or down. Our scheduling schemes ensure to meet the deadlines, while incurring minimum cost. Our schemes also handle multiple concurrent queries, the arrival of new queries, and input rate variations. We have implemented our schemes on top of Apache Spark, in the AWS EMR environment, and evaluated performance with both TPC-H and Yahoo Streaming datasets. Our experimental results show that our scheduling algorithms significantly outperform alternatives, such as using a fixed set of nodes without elasticity, or using Spark streaming.
arXiv:2604.03117v2 Announce Type: replace
Abstract: Infrared vision-language models (IR-VLMs) are becoming important for semantic perception in low-visibility environments, yet their robustness to physical semantic attacks remains underexplored. Existing adversarial patch methods are mainly designed for red-green-blue (RGB) images or closed-set infrared detectors and do not directly address open-ended IR-VLM tasks, where one deployable artifact can affect classification, captioning, and visual question answering (VQA) simultaneously. We propose Universal Curved-Grid Patch, abbreviated UCGP, a universal physical adversarial patch framework tailored to IR-VLMs. UCGP represents the patch as a deployable low-frequency curved grid and optimizes it with a representation-driven objective over subspace departure, topology disruption, and local appearance regularization. Meta Differential Evolution (MetaDE) searches for patch parameters with two physical robustness augmentations: Expectation over Transformation (EOT) capture sampling for imaging variations and thin-plate spline (TPS) patch-deformation modeling for non-rigid local shape changes. Rather than manipulating labels or prompts, UCGP disrupts the clean-category manifold in visual representation space, which later appears as degraded cross-modal outputs. Experiments show that a single shared patch degrades classification, captioning, and VQA across diverse IR-VLM architectures while retaining measurable cross-model transfer, cross-dataset generalization, cross-category extensibility, and real-scene physical effectiveness under the evaluated conditions. These results reveal a robustness blind spot in current infrared multimodal systems. Code and reproducibility assets are available at https://github.com/dyx6663/UCGP.
arXiv:2605.10457v2 Announce Type: replace
Abstract: Real-time Light Detection And Ranging (LiDAR) simulation must find, per emitted ray, the closest intersecting triangle even in dynamic scenes containing large numbers of moving and deformable objects. Dominant acceleration-structure approaches require rebuilding each frame for dynamic geometry -- a cost that compounds directly with scene dynamics and cannot be amortized regardless of how little actually changed.
This paper presents the Gajmer Ray-Casting Algorithm (GRCA), which inverts the question: instead of asking what does each ray hit? it asks which rays can each triangle possibly hit? GRCA geometrically models spinning LiDAR emitters as rotation-traced cones or planes and uses each triangle's emitter-centric apparent area to cull, per triangle, which channels and the rays within those channels can possibly reach it -- without any acceleration structure. GRCA is compute-based and vendor-agnostic by design, targeting highly dynamic, high-resolution simultaneous multi-sensor simulation. At its core, GRCA is a general-purpose ray-casting algorithm: the emitter-centric inversion applies to any setting where rays originate from a known position, not only LiDAR.
Benchmarks evaluate 2-8 simultaneous 128x4096-ray LiDARs (360deg/180deg) over complex dynamic scenes -- with just two sensors casting ~1M rays per frame. With range culling inactive, GRCA reaches up to 7.97x over hardware-accelerated OptiX (GPU) and 14.55x over Embree (CPU).
Two independent extensions further boost performance even in the most complex scene (~22M triangles, ~9M of which are dynamic, 8 LiDARs): range culling at realistic deployment ranges (10-100m) reaches up to 7.02x GPU and 9.33x CPU; a hybrid pipeline -- GRCA for dynamic geometry, OptiX/Embree for static -- reaches up to 10.5x GPU and 19.2x CPU.
arXiv:2605.12111v2 Announce Type: replace
Abstract: We study a sequential resource allocation problem motivated by adaptive network recruitment, in which a limited budget of identical resources must be allocated over multiple rounds to individuals with stochastic referral capacity. Successful referrals endogenously generate future decision opportunities while allocating additional resources to an individual exhibits diminishing returns. We first show that the single-round allocation problem admits an exact greedy solution based on marginal survival probabilities. In the multi-round setting, the resulting Bellman recursion is intractable due to the stochastic, high-dimensional evolution of the frontier. To address this, we introduce a population-level surrogate value function that depends only on the remaining budget and frontier size. This surrogate enables an exact dynamic program via truncated probability generating functions, yielding a planning algorithm with polynomial complexity in the total budget. We further analyze robustness under model misspecification, proving a multi-round error bound that decomposes into a tight single-round frontier error and a population-level transition error. Finally, we evaluate our method on real-world inspired recruitment scenarios.
arXiv:2605.12225v2 Announce Type: replace
Abstract: While deep transformer-based models have advanced rapidly, their internal mechanisms remain largely a mystery. Recent work has prioritized understanding text-based transformer models, leaving ASR systems largely unexplored. In order to address this gap, we examine the internal representations of Whisper's encoder using a sparse autoencoder. We find diverse monosemantic features across linguistic and non-linguistic boundaries, spanning a hierarchy from phonetic to semantic representations, and conduct a causal feature-steering campaign across this hierarchy, including cross-lingual steering. We further find that steering is more reliable for higher-level features than lower-level ones, an asymmetry that may reflect redundant encoding of lower-level information. Altogether, this work demonstrates that Whisper's encoder represents a surprisingly rich hierarchy of linguistic information that extends well beyond what is strictly necessary for transcription.
arXiv:2605.14473v4 Announce Type: replace
Abstract: Retrieval-Augmented Generation (RAG) is usually evaluated by whether the final answer is correct. Under knowledge conflict, this hides a key question: did the model follow retrieved evidence, rely on its parametric prior, or produce a post-hoc rationale? We study this as context compliance, the regime in which retrieved context controls the answer even when it conflicts with the model's prior knowledge. We introduce Context-Driven Decomposition (CDD), an inference-time diagnostic intervention that elicits contextual and prior answers, isolates the conflicting premise, and records a resolution trace that can be perturbed. Across Epi-Scale stress tests, TruthfulQA misconception injection, and cross-model reruns, CDD makes three behaviors visible. First, misleading retrieval can severely degrade accuracy: under a worst-case TruthfulQA misconception-injection probe, Standard RAG reaches only 15.0%. Second, better answers need not share the same mechanism: CDD improves adversarial accuracy on Gemini-2.5-Flash and shows directional gains across Claude variants, yet trace-perturbation sensitivity is high only on Gemini. Third, explicit decomposition improves controlled-conflict robustness over a conflict-aware instruction baseline on localized factual conflicts, with the clearest margins on Entity Swap (88.0% vs 79.3%) and Logical Contradiction (83.2% vs 75.4%). We frame RAG conflict handling as an observability problem.
arXiv:2605.15097v3 Announce Type: replace
Abstract: Frontier LLM agents are increasingly capable of localizing suspicious code, but vulnerability reasoning requires more than access to program artifacts. An agent must carry forward the evidence that actually decides whether a vulnerability exists. Otherwise, safety-relevant facts may be present in the artifact but absent from the reasoning used to justify the claim, creating a semantic gap between available program facts and the evidence used by the agent. This problem is especially acute for stripped binaries, where source-level cues are removed and relevant evidence is fragmented across noisy lifted IR and lossy decompiled views.
We formulate stripped-binary vulnerability reasoning as a semantic grounding problem and present Veritas, a three-stage framework for reliable analysis. First, a static-analysis Slicer recovers witness-backed source-to-sink flows from lifted LLVM IR. Second, an LLM-based Discover stage aligns decompiled code with IR witnesses to construct vulnerability claims. Third, a multi-agent Validator checks these claims through guided debugging and runtime oracles. Together, these stages turn fragmented binary views into checkable claims rather than relying on direct agent inference. We instantiate Veritas for out-of-bounds vulnerabilities and evaluate it on a curated benchmark with flow-level annotations. Veritas achieves 90% recall, outperforms static, dynamic, binary-analysis, and agentic baselines, and reports no false positives among 623 exhaustively validated candidates and only two observed false positives in sampled audits. In a real-world case study, Veritas discovered a previously unknown Apple vulnerability that was confirmed and assigned a CVE, demonstrating that grounded reasoning can produce actionable findings beyond the curated benchmark.
arXiv:2605.16182v2 Announce Type: replace
Abstract: Temporal random walks, which sample causality-preserving paths, are widely used to analyze time-stamped interactions in domains such as microservices, finance, and online platforms. Generating such walks at scale is challenging because real-world graphs evolve as high-volume streams, making continuous ingestion, efficient memory usage, and strict temporal ordering essential for practical deployment. We present Tempest (TEMPoral nEtwork Streaming Traversals), a GPU-accelerated engine for streaming temporal random walks. Tempest combines a GPU-native dual-index organization over a shared edge store with a hierarchical cooperative scheduler that dispatches walks at thread, warp, or block granularity based on per-step node convergence, enabling efficient start-edge selection, hop-by-hop causality enforcement, and window-based eviction without synchronization. It further provides closed-form constant-time samplers for common temporal bias functions. Our evaluation demonstrates sustained real-time processing of billion-edge streams under sliding windows, outperforming prior systems in ingestion and walk generation throughput while preserving causal correctness.
arXiv:2605.16716v5 Announce Type: replace
Abstract: Text-to-video (T2V) generation has rapidly progressed in visual fidelity, yet its ability to faithfully represent multiple cultures within a single prompt remains underexplored. We introduce MAVEN, a multi-agent prompt refinement framework designed to improve cultural fidelity in both mono-cultural and cross-cultural T2V generation. MAVEN decomposes prompts into person, action, and location dimensions, handled by specialized agents operating in parallel or sequentially. To support systematic evaluation, we contribute a new benchmark of 243 culturally grounded prompts and 972 corresponding videos, spanning three cultures (Chinese, American, Romanian), three action categories, and both mono-cultural and cross-cultural scenarios. Evaluations combining CLIP-based metrics, VLM-as-judge assessments, and videoquality measures show that multi-agent refinement, particularly parallel specialization, significantly improves cultural relevance while preserving visual quality and temporal consistency. The dataset and code are available at https://github.com/AIM-SCU/MAVEN
arXiv:2605.17409v2 Announce Type: replace
Abstract: While providing a highly accurate framework for simulating laser-induced many-electron dynamics in atom and molecules, including linear and nonlinear steady-state and transient absorption spectra, time-dependent coupled-cluster theory does not offer a straightforward interpretation in chemical terms. This should be contrasted with conventional time-independent equation-of-motion coupled-cluster or frequency-dependent response models where a simple eigenvector analysis readily reveals the dominant orbital-excitation character of individual excited states. We fill this gap by expanding the left and right coupled-cluster functions in Slater-determinant basis, thus allowing for a time-dependent generalization of configuration weights that can be used to track populations throughout a simulation. The same expansions are used to decompose the time-dependent electric-dipole moment and autocorrelation function, providing a computationally straightforward approach to the assignment of absorption peaks to orbital transitions for single-reference systems. At the time-dependent coupled-cluster singles-and-doubles level of theory, we demonstrate the power of the proposed methodology by assigning valence lines in the linear absorption spectra of four ten-electron molecules (HF, H2O, NH3, and CH4) with different point-group symmetries, validating the assignment by comparison with equation-of-motion coupled-cluster singles-and-doubles theory. In addition, core-level excitations are assigned for HF, H2O, and NH3. Finally, the usefulness of time-dependent configuration weights is illustrated by applications to an impulsive stimulated X-ray Raman scattering process in the Ne atom and to a transient pump-probe spectrum of the HF molecule.
arXiv:2605.18920v2 Announce Type: replace
Abstract: Generative Recommendation (GR) has emerged as a promising paradigm by formulating item recommendation as a sequence-to-sequence generation task over item identifiers. Recent studies have incorporated multimodal signals to provide richer token-level evidence for generation. However, existing approaches largely rely on alignment-centric fusion and underexplore synergistic information across modalities. In practice, synergistic information plays a critical role in capturing emergent item properties that cannot be inferred from any single modality alone. Such properties encode intrinsic item semantics and guide user preferences, enabling models to move beyond surface-level feature matching. To address this limitation, we propose \textbf{SynGR}, a synergistic generative recommendation framework that explicitly encourages the exploitation of cross-modal dependencies during generation. By constraining overreliance on dominant modalities, SynGR enables the model to capture emergent item semantics beyond shared or modality-specific signals. Extensive experiments across three benchmark datasets demonstrate that SynGR achieves superior performance.
arXiv:2605.19650v4 Announce Type: replace
Abstract: The P-cubed Target Insertion Device (P$^3$-TID) is a research instrument dedicated to test novel positron source target configurations inside of the proof-of-principle PSI Positron Production (P-cubed or P$^3$) experiment at the Paul Scherrer Institute. The device allows an easy installation, positioning and replacement of different fixed targets. The present article describes its mechanical design at a detailed level.
arXiv:2604.16197v2 Announce Type: replace
Abstract: Data attribution and valuation are critical for understanding data-model synergy for Large Language Models (LLMs), yet existing gradient-based methods suffer from scalability challenges on LLMs. Inspired by human cognition, where decision making relies on a focused readout of relevant memories rather than replaying all pathways, we introduce RISE (Readout Influence Sketching Estimator). Instead of computing and indexing gradients across the entire LLM, RISE focuses on influence hotspots at the output layer, where influence signals concentrate, and the gradient admits a decomposed outer-product form. This enables a dual-channel representation combining a lexical residual channel (RH) and a semantic projected-error channel (GH). Applying CountSketch projections to these channels achieves strong compression while maintaining accurate attribution. Across the OLMo (1B-32B) and Pythia (14M-6.9B) families, RISE reduces index storage by up to 112$\times$ compared to RapidIn and scales to 32B parameters LLM, where gradient-based baselines such as RapidIn and ZO-Inf become memory-infeasible. We evaluate RISE on two paradigms: (1) retrospective attribution, retrieving influential training examples for specific predictions, and (2) prospective valuation, scoring candidate data utility zero-shot. We validate RISE on three tasks: Howdy backdoor data detection, Finance-Medical domain separation, and Brain Rot high-quality data selection. In a closed-loop Brain Rot study, continued pretraining on RISE-selected data yields consistent downstream improvements. Overall, RISE provides a practical and scalable primitive for influence analysis and training-data selection in modern large language models.
arXiv:2605.20209v2 Announce Type: replace
Abstract: Achieving precise, versatile whole-body character control in physics-based animation remains challenging. Recent diffusion-based policies generate rich and expressive motions but typically rely on gradient-based test-time guidance to satisfy task objectives, which is slow and can reduce robustness. We introduce NaP-Control (Navigating Diffusion Prior for Versatile and Fast Character Control), abbreviated as NaP. Our method uses reinforcement learning to manipulate the latent noise of a task-agnostic diffusion policy prior, steering it toward task-specific behaviors for fast, robust control with high motion fidelity. In contrast to methods that rely solely on offline training, NaP interacts with the environment during training to correct motions and optimize task rewards, improving success rates and enabling adaptation to challenging scenarios. By directly predicting task-optimized diffusion noise, NaP eliminates iterative guidance during denoising and enables efficient inference. Experiments show that NaP attains higher success rates and faster inference while preserving natural motion across diverse tasks.
arXiv:2605.22478v4 Announce Type: replace
Abstract: Composed Image Retrieval (CIR) requires both preserving the visual continuity of the reference image and faithfully executing the semantic variables specified in the modification text, which constitute the core challenge of the task. Existing methods often suffer from Perception Myopia in a single space, or fall into Logic Drift in iterative collaboration due to the perception ceiling of the underlying retriever. To address this issue, we propose a one-stop hierarchical Perception-to-Deliberation Framework (PDF), which, to the best of our knowledge, is the first to introduce Test-Time Scaling (TTS) into CIR. Relying on a hierarchical multi-agent architecture, PDF first utilizes an Intent Routing Manager to dynamically dispatch multi-view Worker perception signals based on modification intents to construct a high-recall candidate pool. Subsequently, the Deliberation Manager combines a training-free reasoning experience distillation mechanism with a tournament-style test-time scaling strategy (T-TTS) to perform experience-guided fine-grained reasoning and produce the final retrieval results. Experimental results demonstrate that PDF achieves SOTA performance on three benchmark datasets: CIRR, CIRCO, and Fashion IQ. This study indicates that experience-driven memory mechanism and TTS represent a highly promising and scalable path for achieving fine-grained multimedia retrieval. The code will be made publicly available upon acceptance.
arXiv:2605.22681v2 Announce Type: replace
Abstract: AI systems are increasingly used to support forward-looking scientific judgment, but it remains unclear whether they can form reliable expectations about future scientific advances. Here we show that strong scientific reasoning does not reliably translate into accurate forecasting of future scientific advances. To study this question, we introduce CUSP, a temporally grounded evaluation suite for event-level scientific forecasting across eight scientific disciplines. Across six frontier AI models, we observe a striking asymmetry in forecasting performance together with systematic error patterns. Models often identify plausible mechanisms underlying future scientific advances, yet perform near chance on feasibility assessment, generate solution strategies that only weakly align with realized advances, and systematically predict scientific advances later than they become publicly observable. Providing additional pre-cutoff scientific knowledge improves performance but does not eliminate these forecasting limitations. These findings suggest that current AI systems possess substantial retrospective scientific competence but limited forward-looking predictive capability. Scientific forecasting should therefore be evaluated as a complementary dimension of AI scientific capability when deploying AI systems for research prioritization and scientific decision-making.
arXiv:2605.22779v2 Announce Type: replace
Abstract: Production systems generate millions of log lines daily, yet most anomaly detectors operate at the session or window-level, flagging groups of lines rather than identifying the specific message responsible. This coarse granularity forces operators to inspect many routine lines per alert. Message-level detection offers finer granularity, but remains challenging. A single event template may correspond to both normal and anomalous messages, failures arise from heterogeneous subsystems, and line-level labeling at scale is impractical. Although large language models (LLMs) can reason over log semantics, applying them to every line is too costly for continuous monitoring. We present FAME (Failure-Aware Mixture-of-Experts), a label-efficient message-level mixture-of-experts framework that uses an LLM only once offline. We annotate at most K labeled lines per template to derive binary normal/anomaly indicators and representative examples. The LLM proposes a partition of templates into failure domains, and a certification step validates the proposal before training. FAME trains a lightweight router and domain experts that run on-premise and output anomaly predictions and failure-domain labels. On BGL, FAME achieves F1 = 98.16 at K = 100 reducing annotation effort by 76x and detects 97.7% of anomalies from unseen EventIDs. On Thunderbird, FAME reaches F1 = 99.95 with perfect recall.
arXiv:2605.24233v2 Announce Type: replace
Abstract: A user faces a list returned by a search system, ordered by a noisy proxy for relevance, and decides whether to pay a fixed cost to inspect another item or stop with the best she has uncovered. She does not enter the page knowing how good its items are, so each inspection both produces a candidate item and refines her belief about the page's underlying quality. We show the optimal policy is a standout rule: the user stops as soon as her best find exceeds her posterior mean of an average item on the page by a depth-dependent threshold. The induced dynamics collapse to a one-dimensional Markov chain, which yields the full distribution of inspection depth through a closed-form recursion. The model uncovers three hidden mechanisms (trust, commit, and cut-losses) on why users stop and yields a rich set of testable implications. Moreover, the Bayesian-rational view delivers a novel learning-to-rank likelihood: an observed depth censors the latent relevance path into a polyhedron of survival inequalities, whose Gaussian probability is a differentiable function of any feature-based relevance prediction model.