arXiv:2601.09173v5 Announce Type: replace
Abstract: Representational similarity analysis and related methods compare the internal geometries of neural networks, but they measure only alignment between spaces, leaving a blind spot -- whether a representation's structure is reliably recoverable, not merely similar. We introduce geometric stability, a distinct axis, and \textit{Shesha}, a metric that quantifies it from a single representation by correlating dissimilarity matrices built from complementary random halves of the feature dimensions. Unlike CKA and Procrustes distance, Shesha is provably non-invariant to orthogonal rotations of the feature basis. This is by design: the basis is privileged for learned models, since probes, patching, and steering act on coordinates, and a rotation-invariant metric cannot see whether the targeted structure survives them. A double dissociation isolates the mechanism -- removing the top principal component collapses CKA while Shesha holds, whereas rotating a representation into its eigenbasis, which preserves the spectrum and CKA exactly, collapses Shesha. Across 2,463 encoder configurations in seven domains, the metrics are redundant under geometry-preserving transforms and anti-correlate under compression ($\rho=-0.47$). Across 170 vision models spanning 6 clean and 38 corruption-shifted datasets, DINOv2 ranks first or second in transferability on three of six clean datasets yet bottom-quartile in stability on five, an isolated dissociation rather than a trade-off.
Science Journals
arXiv:2601.10758v3 Announce Type: replace
Abstract: Large language model (LLM)-based agents are increasingly deployed in applications, such as trip-planning agents and web-use agents, to perform complex planning and execution tasks. Prior work has shown that LLM-based agents are vulnerable to context confusion, where external adversarial content incorporated into the agent's reasoning context may be treated as task-relevant constraints. However, external malicious content can enter the agent context via channels beyond retrieval. In this work, we introduce the User-Relayed Context Manipulation (UReCoM) attack, in which attackers manipulate benign users into relaying adversarial content within user requests, thereby relocating external adversarial content into user-provided task context. Our experimental evaluation shows that UReCoM outperforms five prompt-injection baselines (naive, context ignoring, fake completion, escape-character attacks, and combined attacks) under prevention-based (Sandwich, StruQ, and SecAlign) and detection-based defenses (Perplexity detection, DataSentinel, and CausalArmor). Additionally, UReCoM shows that LLMs can reject explicit malicious instructions more reliably than they can identify adversarial task entities, such as promotion codes, embedded within user requests. On 12 commercial LLM-based agents, we find that validation of adversarial task entities is largely prompt-driven rather than default, highlighting a design flaw in current agent frameworks. These results indicate that current defenses and deployed agents remain insufficient against user-relayed context manipulations, highlighting the need for task-entity-level prevention and default safety verification in LLM-based agents.
arXiv:2601.15275v3 Announce Type: replace
Abstract: We study positional encodings for multi-view transformers that process tokens from a set of posed input images, and seek a mechanism that encodes patches uniquely, allows SE(3)-invariant attention with multi-frequency similarity, and can adapt to the geometry of the underlying 3D scene. We find that prior (absolute or relative) encoding schemes for multi-view attention do not meet these desiderata, and present RayRoPE to address this gap. RayRoPE represents patch positions based on associated rays and computes query-frame projective coordinates to ensure SE(3) invariance. To adapt to scene geometry, RayRoPE predicts (without direct supervision) a per-token depth to obtain its position along the corresponding ray, while also modeling uncertainty and analytically computing the expected positional encoding. We validate our method on the tasks of novel-view synthesis, stereo depth estimation, and feed-forward 3DGS reconstruction. While remaining efficient, RayRoPE consistently improves over alternate position encoding schemes (e.g., 24% relative improvement on LPIPS in RE10K).
arXiv:2601.17257v2 Announce Type: replace
Abstract: We introduce a constrained optimization framework for training transformers that behave like optimization descent algorithms. Specifically, we enforce layerwise descent constraints on the objective function and replace standard empirical risk minimization (ERM) with a primal-dual training scheme. This approach yields models whose intermediate representations decrease the loss monotonically in expectation across layers. We apply our method to both unrolled transformer architectures and conventional pretrained transformers on tasks of video denoising and text classification. Across these settings, we observe constrained transformers achieve stronger robustness to perturbations and maintain higher out-of-distribution generalization, while preserving in-distribution performance.
arXiv:2601.21803v2 Announce Type: replace
Abstract: Retrieval-Augmented Generation (RAG) systems combine dense retrievers and language models to ground their outputs in external documents. However, the interaction between these components remains opaque, creating challenges for deployment in high-stakes domains. We present RAG-E, an end-to-end explainability framework that quantifies retriever-generator alignment through mathematically grounded attribution methods. Our approach adapts Integrated Gradients for retriever analysis, proposes a Monte Carlo-stabilized Shapley Value approximation for generator attribution, and introduces the Weighted Attribution-Relevance Gap (WARG) metric to measure how closely the generator's document usage aligns with retriever rankings. Experiments on PopQA, QAMPARI, and TREC CAST datasets reveal substantial misalignment: depending on the model and setting, generators often ignore top-ranked documents and rely on documents ranked as less relevant. We show that WARG captures retriever-generator alignment better than Pearson and Spearman correlations and can serve as an indicator of RAG performance. RAG-E and WARG provide a practical framework for auditing this interaction, enabling more reliable and transparent RAG systems.
arXiv:2607.06054v1 Announce Type: new
Abstract: Off-the-shelf TTS systems are poorly adapted to Taiwanese Mandarin. Their accent defaults to other Mandarin variants, their tokenizers over-segment common Taiwanese text, and their pronunciation degrades at code-switching boundaries where Chinese and English alternate within one utterance. These problems share one root: the text side lacks adaptation to the Taiwanese context. We address the text side from the bottom up. PangolinTokenizer, a byte-level BPE tokenizer trained on Taiwan-context data, reaches the lowest token rate (0.485 tokens/character) with the smallest vocabulary among nine tokenizers. Barbet, a billion-parameter Traditional-Chinese language model trained on PangolinTokenizer, serves as the text-semantic frontend and ranks first among comparable public models on a 14-task evaluation. BlueMagpie-TTS attaches Barbet to the pretrained acoustic stack of VoxCPM2 through a learned bridge, keeping the acoustic stack fixed. On a 1000-sentence Taiwan-localized test set, it lowers CER from 11.45% to 4.81% and WER from 14.83% to 5.36%, relative reductions of 58.0% and 63.9%. In a blind listening study on 500 of these sentences with ten listeners, 65.6% of majority votes prefer BlueMagpie-TTS.
arXiv:2607.06056v1 Announce Type: new
Abstract: Social pressure -- the awareness of being observed by others -- is a fundamental driver of prosocial behavior in human societies. Yet it is typically assumed that only direct neighbors exert vigilance pressure on an individual, despite empirical evidence that social influence persists to at least three degrees of separation. Here we show that extending the reach of social vigilance beyond direct neighbors substantially promotes cooperation. We couple a Prisoner's Dilemma on one layer of a multiplex network to a vigilance cascade on the other, with influence decaying geometrically with network distance. Extending vigilance to just the second circle of influence shifts the critical temptation for defection by nearly 30\% in sparse networks. Extending to four circles raises this threshold by over 50\%. The $L=1\to2$ transition already accounts for most of the gain, consistent with the decay coefficients of social influence reported in controlled experiments. The effect is strongest in sparse topologies, requires that the vigilance and game layers be aligned, and reproduces directly on a real social network of physicians; in dense, hub-dominated networks the gain instead depends sharply on how fast influence decays with distance, switching between weak and strong cooperation as the decay rate crosses a threshold. Our results strongly suggest that even modest expansions of social awareness -- such as those enabled by online social platforms -- can substantially reshape the landscape of cooperative behavior in human populations.
arXiv:2510.09352v2 Announce Type: replace
Abstract: We propose a low-rank method for solving the Helmholtz equation. Our approach is based on the WaveHoltz method, which computes Helmholtz solutions by applying a time-domain filter to the solution of a related wave equation. The wave equation is discretized by high-order multiblock summation-by-parts finite differences. In two dimensions we seek to compress the solution in matrix form, and in three dimensions using tensor trains. To control rank growth we use step-truncation during time stepping and a low-rank Anderson acceleration for the WaveHoltz fixed point iteration. We have carried out extensive numerical experiments demonstrating the convergence and efficacy of the iterative scheme for free- and half-space problems in two and three dimensions with constant and piecewise constant wave speeds.
arXiv:2607.05536v1 Announce Type: cross
Abstract: Randomized smoothing has emerged as a scalable technique for certifying the adversarial robustness of classifiers. However, its application to regression remains under-explored and faces unique challenges. Existing regression certificates rely on probabilistic acceptance regions and fail to exploit the local geometry of the function. In this work, we present a novel framework for certified robust regression that addresses these limitations. We derive a prediction-centered certificate that guarantees the stability of the smoothed model's prediction and ensures practical computability at test time. We investigate several alternatives for constructing these certificates by explicitly incorporating means, variances, and gradients. In particular, we demonstrate on the MNIST rotation task that utilizing gradient information yields significantly tighter robustness certificates compared to the current state-of-the-art, alpha-smoothing.
Reward-Density Heuristic for Dynamic Multi-Vehicle Routing: Performance and Computational Efficiency
arXiv:2607.06066v1 Announce Type: new
Abstract: The Vehicle Routing Problem (VRP) and its variants represent some of the most practically consequential optimization challenges in modern logistics and urban mobility. In this study, we address a dynamic, online variant combining elements of the VRP and the Orienteering Problem (OP), in which a fleet of vehicles must maximise cumulative reward collected within a fixed time horizon while continuously replanning as new tasks arrive. We propose and evaluate a reward-density heuristic for dynamic multi-vehicle assignment, referred to as the Efficiency heuristic. We evaluate this formulation across two application domains: autonomous drone task allocation and urban taxi dispatch, across multiple fleet sizes and task scales. The proposed method is compared with four classical construction heuristics and three metaheuristic algorithms (Adaptive Large Neighbourhood Search, Genetic Algorithm, and Simulated Annealing), all evaluated under identical conditions. Across all tested configurations, the Efficiency heuristic matches the solution quality of the best metaheuristic algorithms while requiring two to three orders of magnitude less planning time, establishing Pareto dominance over all competing methods on the reward-versus-compute frontier. These findings suggest a practical design principle for real-time allocation and dispatch systems: in dynamic, time-constrained routing environments, carefully designed greedy heuristics can match the output of sophisticated search procedures at a fraction of the computational cost, making them preferable for online deployment.
arXiv:2607.06068v1 Announce Type: new
Abstract: While CRDTs provide decentralized replication and eventual consistency, Byzantine-resilient deployments require mechanisms for deciding which updates should be trusted and therefore contribute to the reconstructed state. In practice, the trust relationships underlying these decisions may evolve over time as participants join or leave, identities change, and governance rules are revised. However, the information used to make trust decisions is typically managed outside the replicated state itself. This paper introduces a dual-CRDT architecture composed of a \emph{Trust CRDT} and a \emph{Data CRDT}. The Trust CRDT stores and evolves governance information, while the Data CRDT is reconstructed according to the trust configuration derived from the Trust CRDT. Governance therefore becomes replicated state rather than an externally managed artifact. Building upon deterministic reconstruction and Byzantine trust filtering, the proposed model allows trust relationships and governance rules to evolve through ordinary CRDT updates. The resulting architecture provides a recursive governance model in which governance rules determine their own future evolution while simultaneously governing application data. The approach is implemented as a prototype on top of Melda and melda-sec and should be viewed as an initial exploration of decentralized trust governance and evolution for Byzantine-resilient CRDT systems.
arXiv:2607.05411v1 Announce Type: new
Abstract: Higher education institutions are increasingly expected to ensure that both students and staff develop Generative AI (GenAI) literacies. In response, they are introducing professional development programs and embedding GenAI skills within student curricula. However, current educational frameworks typically assume a linear progression of GenAI literacy, implying that foundational technical understanding must precede creative application. This paper challenges such an assumption through a psychometric analysis of a taxonomy-based self-assessment instrument (n = 158). We applied Rasch measurement theory and Guttman ordering to map the latent perceived order of difficulty of GenAI skills across students, academics, and professional staff. Results reveal a fundamental divergence in perceived competence profiles: while academics follow a more traditional linear path, students exhibit an "inverted" profile, frequently mastering high-level creation tasks before acquiring foundational conceptual understanding. Furthermore, the correlation of skill difficulty between students and academics was weak (r = 0.188). We argue that this "skill bypass" creates a fragile sense of fluency, where high self-efficacy in prompting masks low literacy in AI mechanics. These findings challenge the "one-size-fits-all" curricula and provide the empirical basis for diagnostic-driven, modular interventions that foster genuine human-AI synergy.
arXiv:2607.05416v1 Announce Type: new
Abstract: We present a new method for structural sequence analysis grounded in Algorithmic Information Theory (AIT). At its core is the Ladderpath approach, which extracts nested and hierarchical relationships among repeated substructures in linguistic sequences -- an instantiation of AIT's principle of describing data through minimal generative programs. These structures are then used to define three distance measures: a normalized compression distance (NCD), and two alternative distances derived directly from the Ladderpath representation. Integrated with a $k$-nearest neighbor classifier, these distances achieve strong and consistent performance across in-distribution, out-of-distribution (OOD), and few-shot text classification tasks. In particular, all three methods outperform both gzip-based NCD and BERT under OOD and low-resource settings. These results demonstrate that the structured representations captured by Ladderpath preserve intrinsic properties of sequences and provide a lightweight, interpretable, and training-free alternative for text modeling. This work highlights the potential of AIT-based approaches for structural and domain-agnostic sequence understanding.
arXiv:2607.06077v1 Announce Type: new
Abstract: Photonic time crystals (PTCs) host momentum bandgaps, or k gaps, that enable parametric amplification and lasing of seeded fields. In nonlinear PTCs, Kerr saturation dynamically suppresses the exponential growth, reshaping k-gap amplification into an active, spatially homogeneous k gap soliton train. Here, we show that a localized perturbation on this unstable background then nucleates a transient spatiotemporal excitation: the breathing k gap event. Unlike Peregrine breathers emerging from modulational instability on a planewave background, this event extracts energy from competing host k gap solitons and remains sustained by their interaction. We identify this process as an instability on instability mechanism intrinsic to nonlinear k gap dynamics. The event is robust against noise and disorder, and can be deterministically reshaped into collective breathing patterns by periodic and phase engineered seeding. These results establish k gap engineering as a route to generating and controlling extreme spatiotemporal waves in photonic time varying media.
arXiv:2607.06429v1 Announce Type: new
Abstract: The simulation of high-frequency acoustic wave propagation in unbounded domains with local heterogeneous materials and high-contrast interfaces poses significant challenges to numerical methods. The volume-surface integral equation (VSIE) method is an attractive approach as it automatically satisfies the radiation condition at infinity via Green's functions, handles heterogeneous materials via Newton potentials, and models scattering at high-contrast interfaces via surface integral operators. However, its effectiveness in practical simulations has been limited by high computational costs, sensitivity to sharp interfaces, and insufficient computational verification. This study extends the applicability of VSIE by deriving integral formulations for nested heterogeneous materials with parameter jumps at interfaces. We also develop extensive benchmarks against coupled finite-element and boundary-element methods to verify the VSIE's accuracy and mesh convergence. The various benchmarks using open-source software demonstrate the effectiveness of VSIE for large-scale acoustic simulations.
arXiv:2607.06437v1 Announce Type: new
Abstract: Roaming reactions are organized not by potential-energy saddles but by transition states that are unstable invariant objects in phase space, periodic orbits in the two degrees of freedom studied here. To ask what controls roaming, we modify the Chesnavich model of a barrierless ion--molecule dissociation: its orientation-dependent angular hindrance is replaced by a transverse-stiffness ridge whose angular frequency peaks at an interior radius, and the classical dynamics are studied at a fixed energy just above the dissociation threshold. Comparing two ensembles that differ only in this angular interaction (same radial channel, energy, and inward initial conditions) isolates its effect. The ridge gates entry into the inner well, cutting inner capture from $57\%$ to $15\%$ and returning most of the incoming flux directly to reactants; it does not eliminate roaming but relocates it outward, suppressing it inside the ridge and switching it on farther out. The model retains analogues of the original model's three transition states (tight, free-rotor, and outer orbiting orbits), which we locate as unstable periodic orbits. The tight orbit spans a dividing surface that coincides, within numerical accuracy, with the variational minimum-flux surface, and it carries no barrier along the reaction coordinate: a deep entropic bottleneck placed at an interior radius by the stiffness maximum. Its entropic character is shared with the original model. Strength-matched monotone controls show that the gating tracks the hindrance strength at the bottleneck radius; what the interior maximum supplies is placement, concentrating that strength where it gates most effectively. The trajectories it admits roam nonstatistically, with nonexponential gap-time distributions: the entropic bottleneck governs how much is captured, not the dynamics that follow.
arXiv:2602.02401v2 Announce Type: replace
Abstract: Human motion analysis tasks, such as temporal 3D pose estimation, motion prediction, and motion in-betweening, play an essential role in computer vision. However, current paradigms suffer from severe fragmentation. First, the field is split between ``perception'' models that understand motion from video but only output text, and ``generation'' models that cannot perceive from raw visual input. Second, generative MLLMs are often limited to single-frame, static poses using dense, parametric SMPL models, failing to handle temporal motion. Third, existing motion vocabularies are built from skeleton data alone, severing the link to the visual domain. To address these challenges, we introduce Superman, a unified framework that bridges visual perception with temporal, skeleton-based motion generation. Our solution is twofold. First, to overcome the modality disconnect, we propose a Vision-Guided Motion Tokenizer. Leveraging the natural geometric alignment between 3D skeletons and visual data, this module pioneers robust joint learning from both modalities, creating a unified, cross-modal motion vocabulary. Second, grounded in this motion language, a single, unified MLLM architecture is trained to handle all tasks. This module flexibly processes diverse, temporal inputs, unifying 3D skeleton pose estimation from video (perception) with skeleton-based motion prediction and in-betweening (generation). Extensive experiments on standard benchmarks, including Human3.6M, demonstrate that our unified method achieves state-of-the-art or competitive performance across all motion tasks. This showcases a more efficient and scalable path for generative motion analysis using skeletons.
arXiv:2607.06087v1 Announce Type: new
Abstract: This paper presents an approach and application of optimization of spatial packaging of interconnected systems with physical interactions (SPI2) in three-dimensional component placement problems. To enable its application for an automotive use case, SPI2 must support both initial design generation, including component alignment, and robust system-level coordination, requiring improved solution reliability and tractable computational cost. To address these requirements, the proposed methodology improves convergence rate and solution quality by enhancing numerical robustness in gradient-based optimization while reducing computational load. Existing SPI2 approaches are extended through the addition of alignment capabilities, enabling the representation of port-to-port alignments between components. Furthermore, the applicability of SPI2 is expanded by treating component placement locations as design variables, allowing for penalty-based coordination to ensure design feasibility and enabling integration within system-level optimization. The approach is validated using a multi-objective optimization framework based on Nondominated Sorting Genetic Algorithm II (NSGA-II), applied to a combined powertrain optimization and battery chassis integration problem. This demonstrates the effectiveness of the SPI2 in a system-level design context. The results show a twofold application of SPI2 in an automotive use case: first, as a tool for initial design generation, and second, as part of a system-level design coordinator that outperforms a discretized exhaustive search while requiring lower computational cost.
arXiv:2607.06093v1 Announce Type: new
Abstract: Hybrid electrolyzer systems combining alkaline water electrolysis and proton exchange membrane water electrolysis have been investigated in the literature motivated by the expectation that their contrary techno-economic characteristics compensate for the individual technical and economic restrictions of each technology, thereby improving the profitability of green hydrogen production. To reassess the economic potential of hybrid electrolyzer systems beyond these technology-specific assumptions, we independently vary two key characteristics, electrolyzer efficiency and investment cost, in a large-scale sensitivity analysis. For each generated parameter configuration, we performed a techno-economic optimization of a green hydrogen supply chain, including two electrolyzers. The resulting system design, cost objective, and dispatch behavior are subsequently analyzed. Consequently, hybrid electrolyzer systems are identified as optimal if they provide a cost benefit over single electrolyzer systems. The analysis reveals that hybrid electrolyzer systems represent the optimal solution in at most 5.0% of the investigated cases. Furthermore, the maximum cost benefit is 0.057 EUR/kgH2, which corresponds to only about 1% of the total green hydrogen production cost. Additional analyses considering variations in energy purchase prices, storage fees, availability of renewable energy, and baseline electrolyzer efficiency yield negligible changes to these results. Hence, considering that hybrid electrolyzer systems offer marginal cost benefits and prove economically optimal in very few cases, they seem more likely to represent an economic fallacy than a smart strategy.
arXiv:2607.05436v1 Announce Type: new
Abstract: The rapid scaling of over-parameterized machine learning architectures, particularly LLMs, raises a profound crisis: do these systems exhibit genuine intelligence, or are they merely sophisticated statistical pattern matchers? Classical flat Euclidean statistics cannot differentiate continuous interpolation from the autonomous discovery of novel causal laws. To resolve this, we introduce Statistically Meaningful Geometry (SMG), a framework modeling over-parameterized learning systems as infinite-dimensional non-parametric Orlicz fiber bundles. We prove that under persistent out-of-distribution (OOD) stimuli governed by unmodeled causal mechanisms, continuous optimization fails. Unmodeled variance is rejected by the visible horizontal base manifold, leaking into the unobservable vertical fiber space and generating an accumulation of Active Acausal Tension. Driven by the statistical manifold's non-linear curvature, this tension inevitably strikes a conjugate focal boundary ($T_{\text{crit}} = \pi^2 / K_{\text{max}}$), triggering localized volumetric collapse and a catastrophic matrix singularity ($[G_f]^{-1} \to \infty$). We demonstrate this geometric breakdown acts as the strict non-equilibrium trigger for a Gauge Symmetry Break (GSB). The system purges hidden tension from unobservable gauge redundancies, spontaneously crystallizing a new, mathematically independent horizontal coordinate axis. This non-parametric phase transition registers as a discrete $+1.0$ integer step-jump in observable Structural G-Entropy. By decoupling parameter charts and subjecting emergent axes to a Minimal Energy Path Criterion and a Causal Invariance Filter, we distinguish genuine discovery from malignant hallucinations. Ultimately, SMG provides a parameter-free, falsifiable dashboard to mathematically certify true intelligence, transforming AI for Science into an engine of autonomous paradigm shifts.
arXiv:2607.06094v1 Announce Type: new
Abstract: Faults on a cyber-physical system (CPS) are too rare and unrepresentative to characterise, or even to select a model on, so detection must instead model normal behaviour; the standard point-adjusted evaluation, however, rewards detectors that never do. CPS normal behaviour is the union of many imbalanced, curved, thin-fringed operating regimes rather than a single blob; we state this structure as ten assumptions (A1-A10), abbreviated Massive, Implicit, Imbalanced Multimodality (MIIM). We model the normal law with a jointly learned latent representation plus explicit Gaussian-mixture mode clustering, scored in the latent rather than by a global density or a reconstruction residual, and evaluate under a deliberately fair protocol: raw point-wise metrics with no point adjustment, a trivial-detector difficulty split, prevalence-matched F1, and train-normal-only calibration. On three real CPS datasets (WADI, HAI, SKAB), the detector wins both the combined column and the difficult correlation/dynamics-fault column on all three, reaching difficult-subset AUROC 0.831 on HAI, 0.726 on WADI, and 0.610 on SKAB. The margin is largest on the two multimodal datasets the MIIM assumptions target and slimmest on the near-unimodal one, tracking multimodality as the thesis predicts, and it holds against three deep detectors (USAD, TranAD, GDN) re-computed with the same raw metrics, all of which collapse on the difficult subset. The methodological contributions are the MIIM assumption set, the difficulty-stratified fair protocol, and a latent-only score that drops reconstruction because a flexible decoder rebuilds the hard faults faithfully.
arXiv:2607.05458v1 Announce Type: new
Abstract: Large language model (LLM) agents are usually improved by changing prompts, models, or hand-written workflows, while the execution harness around the model is treated as fixed infrastructure. We argue that this harness is itself a learnable control layer. We formalize harness operation as a finite-horizon Harness MDP, where a lightweight controller selects structural execution actions while the LLM executor remains frozen. The controller is trained from offline rollouts using advantage-weighted regression with only terminal task-rubric rewards. We also separate final task quality from a post-hoc Harness Maturity Score, which measures whether the harness follows reliable execution patterns rather than only whether the final answer is correct. This separation gives a finite-buffer view of harness learning: final-quality gains require high-return support in the offline buffer, while process behavior can shift whenever it aligns with advantage-weighted actions. Across six controlled domains and two public-benchmark adapters, the learned controller consistently improves verification behavior and selectively improves final task quality, with the largest gains on adapted tau-bench retail, adapted AgentBench DB-Bench, and coding with a calibrated structural verifier. Ablations against behavior cloning and Forced CHECK show that the gains are not explained by imitation or by simply adding checks. These results identify harness control as a learnable layer for frozen LLM agents, while showing that offline support limits when better process control becomes better final answers.
arXiv:2607.05894v1 Announce Type: new
Abstract: Modern software supply chains comprise hundreds of transitive dependencies, yet existing analysis tools operate at either the ecosystem level (dependency graphs) or the code level (static analysis within packages). This separation creates two failure modes. First, false-positive CVE alerts for unreachable code. Second, blind spots for structurally critical micro-dependencies. We introduce cross-level risk propagation, a framework that bridges code-level risk metrics with ecosystem-level dependency exposure through a unified risk formula. Preliminary evaluation on 50 packages across npm and PyPI reveals a class of hidden amplifiers -- micro-dependencies with fewer than 50 methods but over 50,000 dependents -- that carry outsized supply-chain risk invisible to all current Software Composition Analysis (SCA) tools. Without cross-level analysis, such packages can harbor exploitable code for years because no current tool considers both internal code structure and ecosystem position simultaneously. These results suggest that cross-level analysis opens a new design space for supply-chain security.
arXiv:2607.05467v1 Announce Type: new
Abstract: Fog severely degrades the visibility of small unmanned aerial vehicles (UAVs) in skydominant, long-range imagery, reducing the reliability of downstream detection and tracking. This paper presents a task-driven evaluation framework that links depth-aware synthetic fog generation, image restoration, object detection, and tracking within a unified pipeline. Given the practical difficulty of collecting and annotating foggy UAV scenes, synthetic fog is generated from real clear-weather outdoor images containing UAV targets using monocular depth estimation and the atmospheric scattering model. Representative restoration methods from classical, convolutional neural network (CNN)-based, and transformer-based families are first compared, after which the selected restoration model is integrated into the downstream perception pipeline. Detection is evaluated under both clean-only and fog-inclusive training regimes using multiple detector variants, while tracking-by-detection is assessed on clean, foggy, and restored video sequences. Beyond image-level restoration metrics, the study evaluates how fog and restoration affect detection robustness and tracking performance. The results show that fog substantially degrades both detection and tracking, primarily through increased missed detections. Fog-inclusive training provides the most consistent improvement in robustness, whereas test-time restoration is most beneficial when the detector has been trained only on clean imagery. These findings show that restoration quality does not necessarily translate into proportional gains in downstream perception and therefore should be evaluated jointly with detection and tracking performance.
arXiv:2607.05553v1 Announce Type: new
Abstract: Transient stability control in smart grids requires rapid post-fault damping of generator frequency and rotor angle deviations to prevent cascading failures. This paper proposes FedPPO-PG, a Federated Multi-Agent Proximal Policy Optimization framework with Physics-Grounded neighborhoods, which reformulates transient stability control as a cooperative multi-agent reinforcement learning problem optimized directly against closed-loop stability objectives. Each generator hosts an independent local actor augmented with the frequency deviations of its two most strongly coupled electrical neighbors, identified from the post-fault Kron-reduced susceptance matrix. A guided policy initialization phase warm-starts all actors from the classical decentralized controller, while a centralized critic guides advantage estimation under the centralized training--decentralized execution (CTDE) paradigm. Evaluated on a simulation of the IEEE 39-bus benchmark system across five training and three unseen fault contingencies, FedPPO-PG achieves 100% stabilization in all 24 trials, reduces mean stability time by 72.4%, and cuts the control power by 7-14 times compared to the centralized baseline. Each actor executes independently with no central coordinator at deployment, and the per-actor inference latency satisfies the IEEE/IEC 60255-118-1-2018 real-time reporting requirements.