arXiv:2607.05727v1 Announce Type: new
Abstract: Pre-trained Vision-Language Models (VLMs) like CLIP have proven highly effective as foundation models for various downstream applications. However, prompt learning in VLMs encounters a performance-generalization dilemma: while prompts can be tuned to achieve high accuracy on seen distributions, this tuning process often undermines their generalizability to unseen data. The limited set of learnable prompts, which contextualize and condition the input to steer it toward the task within the pretrained VLM, tends to overfit the training data, leading to a trade-off between task-specific performance and preserving generalization. To address this dilemma, we introduce SAMPLe (Sharpness-Aware Minimization Prompt Learning), a plug-in sharpness-aware optimizer that enhances prompt generalizability by accounting for loss landscape sharpness. Unlike conventional methods, SAMPLe balances exploration and exploitation by satisfying objective function constraints at each step, dynamically adapting to the current optimization state based on the local curvature and gradient properties. This approach reduces overfitting on seen distributions and improves adaptability to unseen data, preserving the generalization potential of pre-trained VLM models. We integrate SAMPLe into multiple prompt learning frameworks, including CoOp, CoCoOp, MaPLe, TCP, and Co-Prompt, demonstrating its effectiveness across diverse methods. Experiments show that SAMPLe elevates prompt learning frameworks and consistently outperforms existing optimizers across diverse settings, establishing itself as a robust, model-agnostic solution for prompt learning.
Science Journals
arXiv:2607.05743v1 Announce Type: new
Abstract: AI coding agents now read repositories, call tools, and execute shell commands with limited human oversight, and a fast-growing body of work studies whether the execution layer around them is actually safe. That literature is scattered. Papers on sandbox isolation, capability and access control, policy enforcement, time-of-check-to-time-of-use (TOCTOU) races, Model Context Protocol (MCP) threats, identity delegation, execution provenance, network egress control, and static analysis of agent-generated code are published independently and rarely cite one another. We systematize 39 papers published between 2023 and 2026 into 17 categories, each verified directly against its source. The same verification protocol also confirms four disclosed, patched CVEs directly affecting production agent harnesses. Reading across categories surfaces five cross-cutting gaps that no single paper addresses. (1) Isolation architectures and capability models are almost never evaluated against one another on a shared benchmark. (2) Policy-enforcement studies report failure rates from 69% to 98% of real denylists, yet no isolation paper re-evaluates its own defense under that adversarial setting. (3) TOCTOU and MCP threats are analyzed as separate literatures despite both being instances of the same state-validation problem. (4) Every enforcement mechanism assumes an honest policy author, leaving policy-authoring error itself unaddressed. (5) Benign but out-of-scope agent actions occurring at rates up to 17.1% under realistic prompting are addressed by no access-control or capability paper in the corpus. Existing broader surveys of agentic AI security discuss sandboxing only as one item among many defenses, leaving execution security without a dedicated systematization. This paper is written to fill that gap. We conclude with a research agenda directed at the five gaps.
arXiv:2607.05808v1 Announce Type: new
Abstract: Large language models (LLMs) are increasingly integrated into computing education, yet nearly all prior research has focused on text-based interactions. As voice-enabled interfaces become more capable and more common, there is growing interest in understanding how voice input might shape students' use of LLM-powered tools. In this exploratory study, we investigated how introductory programming students interact with Prompt Problems, which are programming tasks that require crafting natural-language prompts to generate correct code. Students (N = 919) solved a series of Prompt Problems with the freedom to select or switch between text and voice input modalities. We collected their prompt submissions as well as post-activity survey responses, then analysed differences in prompt accuracy, persistence, and perspectives by modality. For two of the three problems, we found that students who typed their prompts using text were more likely to have those prompts succeed on the first attempt than students who submitted unedited voice prompts. There was no difference in success rate if students edited their transcribed voice prompts before submission. Across the problems, we found evidence that students who tried voice prompting varied in their usage of modality - perhaps indicating a complementary, or non-preferential approach. However, most students only tried and reported preferring text. Our qualitative analysis revealed how students' perceived the roles of voice and text input in shaping their problem-solving process, as well as the reported drawbacks and advantages of each modality. We discuss implications for future multimodal tools and instructional design in computing education.
arXiv:2607.05842v1 Announce Type: new
Abstract: Large language model (LLM)-assisted software security operates at a difficult boundary: the vulnerability-analysis terminology needed for legitimate code review, triage, and repair can closely resemble terminology associated with misuse. Existing safety and cybersecurity evaluations are difficult to interpret in this setting because they often compare unrelated model families, thereby conflating safety behavior with differences in architecture, scale, training data, and deployment. To isolate this factor, we study safety state: whether refusal behavior remains intact (Aligned) or has been refusal-ablated (Abliterated) within same-lineage models. We ask how this safety state affects defensive utility across software-security workflows. We compare aligned instruction-tuned models with publicly released refusal-ablated descendants from two model families, Gemma and Qwen. We evaluate Aligned and Abliterated states on vulnerability detection, CWE attribution, vulnerable-line localization, root-cause localization, and executable patch validation. We further treat prompt wording as a controlled framing dimension: prompts begin with neutral code-review language, add authorization context, and vary the density of cybersecurity terminology. In a Gemma-based Java/Vul4J repair-validation study, Abliterated achieves higher early-stage validation rates, with 67.8%, 65.0%, and 32.8% of patches judged usable, successfully applied, and successfully compiled, respectively, compared with 29.9%, 24.9%, and 9.0% for Aligned. In the Qwen pair, Abliterated improves localization performance, increasing line-level F1 from 2.08% to 3.91% and Top-1 accuracy from 4.10% to 6.95%. These findings suggest that evaluations of LLM-based security assistants should jointly measure whether models respond, whether their usable responses are correct, and whether their outputs remain actionable across the engineering workflow.
arXiv:2607.05844v1 Announce Type: new
Abstract: Agent systems accumulate conflicting observations across branches, retries, and replicas, yet many practical memory layers still collapse disagreement behind overwrite rules that are difficult to inspect or correct. We present StateFuse, a conflict-aware replicated memory contract built on standard OpSet/CRDT merge. StateFuse does not introduce a new join algebra; it defines an agent-facing semantics layer with immutable history, explicit conflict objects, exact and semantic correction handles (claim_id / claim_ref), deterministic predicate contracts, and projection-time resolution that cannot rewrite replicated state.
We evaluate StateFuse against flat multi-value, raw-log, provenance-style, and collapsed baselines under matched resolver and verification policies. On a 282-question official conflict-bearing MemoryAgentBench slice, the compared methods tie on answer accuracy, but conflict-preserving surfaces keep contradictions visible while collapsed surfaces do not. In a controlled agent loop with uniform verification, preserving ambiguity enables safer abstention and correction than early collapse. A correction-handle ablation further shows that semantic handles matter when exact prior identifiers are unavailable.
The resulting claim is narrow: StateFuse is best supported as a safer public memory contract for contradiction surfacing, abstention, and auditable correction, not as a universal accuracy gain.
arXiv:2607.02714v2 Announce Type: replace
Abstract: There is no doubt that safety alignment is an essential step in LLM training. However, conceptually it does not distinguish between various domains and the level of potential harm of a query, which creates significant complications in the fields like cyber security, where a model should not be constrained by its safety circuits to accomplish the goals of legitimate, authorized operations. In this work, we share our findings from a large scale abliteration experiment on 24 open-source LLMs and show that domain-specific abliteration is achievable with standard methodology on the example of a 1T-parameter Kimi K2. Building on recent work showing that refusal in LLMs occupies a multi-dimensional subspace within layers, we find that it is also distributed widely across layers, especially in trillion-parameter MoE architectures, and so we aim to capture the part of it that represents harmful concepts in the cybersecurity domain exclusively. We also investigate the correlation between models' features and the effect of domain-specific abliteration, identifying that the type of safety training and architecture are the most reliable predictors. Finally, we classify the models into 3 abliteration susceptibility tiers and put forward a set of conjectures as to why a particular effect from this intervention might be observed in a given model.
arXiv:2607.03692v2 Announce Type: replace
Abstract: Spectral methods are widely used to construct representations from the geometry of data, but they often rely on a fixed kernel, graph Laplacian, or manually selected feature scaling. We propose Physics-Informed Eigenfunction Features with Learnable Scaling (PIEFS), a supervised neural representation-learning framework with a spectral inductive bias, based on a modified Dirichlet energy. In PIEFS, scalar coordinate maps are trained under empirical Gram orthogonality, a supervised linear readout, and a Dirichlet penalty in which the input gradient is transformed by a learnable metric $A(x)=\Lambda(x)U(x)$. The diagonal factor $\Lambda(x)$ controls anisotropic scaling, while the orthogonal factor $U(x)$ is parameterized by a structured product of Givens rotations. This construction yields task-adaptive Dirichlet-regularized coordinates rather than eigenfunctions of a fixed supervision-independent operator. Experiments on synthetic, tabular, and image-based benchmarks study the effect of identity, diagonal, and rotation-scaling metrics, and compare the resulting coordinates with classical baselines and NeuralEF. The results support PIEFS as a compact supervised spectral representation method and identify optimization stability, validation on explicit operator eigenproblems, and richer metric parameterizations as the main directions for future work.
arXiv:2606.25309v2 Announce Type: replace-cross
Abstract: We prove a residual-work compatibility theory for positive-cone viscoelastic Reynolds states. In Oldroyd--B, the entropy cancellation eliminates incompressible transport and upper-convected stretching against polymeric stress work. After quotienting pressure tensors and spatial means, positive pressure-free residual work can be paid only by the conformation residual through the entropy-dual lever (G=I-A^{-1}). This yields the closed residual-work cone (P+(\alpha/2)\int_{\mathbb T^d}G:S,dx\le 0), exact windowed tests, and the least-cost Hilbert-space repair; constrained closures pay through the projected lever. Strong residual-data limits preserve this cone, while weak--weak limits require the product-defect measure carried by (G:S). The augmented topology is sharp, as shown by localized residual packets. For FENE-P, the lever (G_b(C)=((b-d)/(b-\operatorname{tr}C))I-C^{-1}) makes the finite-extensibility boundary a genuine residual-work boundary.
arXiv:2606.01256v2 Announce Type: replace-cross
Abstract: This paper introduces a distribution-free framework for constructing post-detection confidence sets for changepoints after stopping a sequential change detection procedure. It is well known that conformal test martingales can be used to sequentially detect changes in distribution, but by themselves provide no inference for the time at which a proclaimed change occurred. Past work on post-detection inference requires pre- and post-change classes of distributions to be known, but this paper accomplishes localization of the changepoint without any distributional assumptions. We establish finite-sample coverage guarantees (conditional on correct detection). We provide non-asymptotic bounds on the conditional expected size of the confidence sets. Under suitable asymptotic regimes, we prove that the conditional expected size of the confidence set remains uniformly bounded and demonstrate strong empirical performance on simulated and real data. To the best of our knowledge, this is the first general distribution-free framework for sequential changepoint localization with valid post-detection coverage.
arXiv:2607.04633v2 Announce Type: replace
Abstract: The rational design of ionic electrolytes for lithium-ion batteries (LIBs) is severely constrained by the vast solvent-salt combinatorial space and low efficiency of empirical trial-and-error. While molecular dynamics (MD) bridges microscopic solvation structures and macroscopic physicochemical properties, classical force fields often lack sufficient accuracy for multicomponent systems. To address these challenges, we develop an automated differentiable OPLS-AA force field parameterization workflow tailored for general ionic electrolytes. It employs topology-guided atom typification to reduce parameter redundancy and optimizes Lennard-Jones parameters via the DMFF framework, with experimental density as the fitting target and ionic conductivity as an independent validation metric. Rigorous convergence tests yield a standardized simulation protocol with $\sim$100,000-atom systems and 35-40 ns NVT runs to ensure reliable transport property quantification. High-throughput MD simulations of over 10,000 formulations spanning 67 solvents and 15 lithium salts are conducted on the Tianqiong platform, generating a comprehensive dataset covering five core properties: density, dielectric constant, viscosity, diffusion coefficient, and ionic conductivity. t-SNE visualization reveals partial clustering of distinct salt chemistries, continuous property gradients with concentration and temperature, and internal physical self-consistency, with solvent composition identified as another key performance regulator. Together, the accurate transferable force field and large-scale dataset provide a solid foundation for data-driven rational design of ionic electrolytes.
arXiv:2406.19594v3 Announce Type: replace-cross
Abstract: Optimizing the interfaces in perovskite solar cells (PSCs) is essential for enhancing their performance, improving their stability, and making them commercially viable for large-scale deployment in solar energy harvesting applications. Point defects, like vacancies, have a dual role, as they can inherently provide a proper doping, but they can also reduce the collected current by trap-assisted recombination. Moreover, they can play an active role in ion migration and degradation. Using ab initio density functional theory (DFT) calculations we investigate the changes in the band alignment induced by interfacial vacancy defects in a TiO$_2$/MAPI/Cu$_2$O based PSC. Depending on the type of the vacancy (Ti, Cu, O, Pb, I) in the oxide and perovskite materials, additional doping is superimposed on the already existing background. Their effect on the performance of the PSCs becomes visible, as shown by SCAPS simulations. The most significant impact is observed for $p$ type doping of TiO$_2$ and $n$ type doping of Cu$_2$O, while the effective doping of the perovskite layer affects one of the two interfaces. We discuss these results based on modifications of the band structure near the active interfaces and provide further insights concerning the optimization of electron and hole collection.
arXiv:2412.10860v2 Announce Type: replace-cross
Abstract: Quantum Support Vector Machine is a kernel-based approach to classification problems. We study the applicability of quantum kernels to financial data, specifically our self-curated Dhaka Stock Exchange (DSEx) Broad Index dataset. To the best of our knowledge, this is the first systematic study of quantum kernels applied to this dataset. Working within the empirical quantum advantage (EQA) framework of Krunic et al., we benchmark several quantum kernels against a classical RBF-kernel SVM baseline, propose the best-performing kernel for this dataset, and relate the observations to the Phase Space Terrain Ruggedness Index metric. We estimate the resources needed to carry out these investigations on a larger scale for future practitioners.
arXiv:2605.14492v3 Announce Type: replace-cross
Abstract: This study provides an analytical foundation for adversarial synchronization control in Kuramoto oscillator networks, where small gradient-based perturbations applied repeatedly to oscillator phases can dramatically enhance or suppress collective synchronization. Using the Ott--Antonsen reduction, we derive an exact closed-form expression for the effect of a single adversarial perturbation (kick) on the order parameter. A key finding is that each kick produces a finite, coupling-independent increment in the order parameter even when synchronization is arbitrarily weak, which combined with slow relaxation near the critical coupling and mean-field feedback explains the disproportionate amplification previously observed in numerical simulations. Fixed-point analysis further reveals a fundamental asymmetry between enhancement and suppression, with the latter governed by noise-induced escape in finite systems. Extending the framework to networks via the annealed network approximation, we show that the theory captures the synchronization behavior of representative model networks and identify a decoupling between kick sensitivity and mean-field dominance in scale-free networks. These results offer a tractable theoretical basis for understanding and designing kick-based synchronization control in oscillator networks.
arXiv:2607.05904v1 Announce Type: new
Abstract: Training a language model against its own reference-free judgments (the premise of self-rewarding, self-play, and LLM-as-a-judge pipelines) assumes a model's verdict on a shown answer tracks correctness. We show it fails structurally: conditioned on a candidate, a judge scores plausibility, not correctness, leaving false-positive basins a policy learns to exploit. We measure this with a hidden-anchor audit: a held-out, cross-source exact-match check the judge never sees. On GSM8K with Qwen3 policies, self-play drives the judge's pass rate from 0.72 to 0.94 while true accuracy stays at 0.20 (three seeds). This reward hacking is not white-box gaming: the errors transfer across judge families (Qwen, Llama, Gemma) and scales, a strict three-judge ensemble still accepts 55% of them, and no plausibility-scoring defense closes the basin. The decisive variable is whether the judge commits an answer of its own before using the candidate: committing first drops the false-positive rate from 0.719 to 0.012, blind solving lifts discrimination to 0.96, and used as the training reward the de-anchored channel keeps false positives at zero, preventing the basin rather than only detecting it. A falsifiable bound (the gap is at most 1 - accuracy) predicts which regimes are exposed. The full arc replicates without training under best-of-N selection in code and competition math, and with a Gemma policy.
arXiv:2607.05907v1 Announce Type: new
Abstract: Spot is a mature, open-source C++/Python library and toolset for Linear Temporal Logic (LTL) and $\omega$-automata manipulation. While Spot is routinely used as a research and verification back-end, its rich visualization capabilities and Python interface also make it an attractive platform for \emph{teaching} the connections between temporal logic formulas and the $\omega$-automata that give them their semantics.
arXiv:2607.06370v1 Announce Type: new
Abstract: Vision-Language-Action (VLA) models have emerged as a promising approach for generalizable robotic manipulations. In particular, flow matching-based VLA models have shown remarkable success due to their capability to generate precise and smooth action sequences and capture multimodal distributions. However, the iterative denoising process in the action head acts as a major computational bottleneck, posing a critical challenge for real-time deployment. To address this challenge, we propose ActionCache, a plug-and-play external cache that opportunistically reuses past intermediate actions to warm-start generations from the vicinity of target actions, thereby drastically reducing the inference latency. Specifically, ActionCache stores the intermediate actions with compact multimodal keys, which enables retrieval from similar past contexts across different episodes or even different tasks. Experimental results in simulation and real-world environments demonstrate that ActionCache maintains high task success rates in a low-latency regime, achieving inference acceleration of up to $11.75\times$ and $34.43\times$ for representative flow-based VLA models, $\pi_{0.5}$ and GR00T-N1.6, respectively.
arXiv:2607.05978v1 Announce Type: new
Abstract: Multimodal large language models can emit localized predictions, bounding boxes for objects and temporal windows for video and audio events, but they hallucinate these regions prolifically. The model's own token log-probabilities are nearly uninformative: they conflate grounding quality with input ambiguity, and coordinate tokens become near-deterministic once the model commits. We propose Multi-Token Localized Attention (MTLA): a training-free, post-hoc score that measures how strongly a prediction's tokens attend to the region they claim. Prior attention-based detectors, which sum attention over the entire input modality and read a single response token, are weaker special cases; we show that summing only within the claimed region and aggregating across all prediction tokens recovers a stronger grounding signal. The same recipe applies almost trivially to other modalities and tasks: object detection in images and temporal localization in video and audio. Across multiple MLLM families and three modalities, MTLA improves hallucination AUROC by +7 to +38 over the best prior training-free baseline. Used as a confidence score for re-ranking, it nearly doubles the zero-shot COCO detection AP of an open-source 8B generalist (from 20.4 to 37.0), narrowing the gap to supervised detectors without any task-specific training.
arXiv:2607.05956v1 Announce Type: new
Abstract: The integration of Large Language Models (LLMs) into scientific research workflows, particularly for bibliographic discovery and literature synthesis, raises significant methodological, epistemic and regulatory challenges for the Social Sciences and Humanities (SSH), especially with regard to disciplinary diversity, multilingual access to sources and the evaluation of results. This paper presents an on-going use case developed within the European project LLMs4EU and the ALT-EDIC infrastructure, aimed at adapting foundation models to SSH research practices and supporting tasks such as question answering, comparative document analysis and literature review. The evaluation framework follows the LLMs4EU protocol and encompasses both independent quantitative benchmarking (retrieval, summarisation, traceability and hallucination detection) and a qualitative assessment involving a panel of Digital Humanities experts. By embedding model adaptation within research infrastructures and a structured legal and ethical compliance framework, the use case explores how domain-sensitive and regulation-aware generative AI can support SSH scholarship while preserving reliability and epistemic responsibility.
arXiv:2605.10439v2 Announce Type: replace
Abstract: Low-Rank Adaptation (LoRA) has become a widely used mechanism for customizing diffusion models, enabling users to inject new visual concepts or styles through lightweight parameter updates. However, LoRAs can memorize training images, causing generated outputs to reproduce copyrighted or sensitive content. This risk is particularly concerning in LoRA-sharing ecosystems, where users distribute trained LoRAs without releasing the underlying training data. Existing approaches for mitigating memorization rely on access to the training pipeline, training data, or control over the inference process, making them difficult to apply when only the released LoRA weights are available. We propose \textbf{Base-Anchored Filtering (BAF)}, a training-free and data-free framework for post-hoc memorization mitigation in diffusion LoRAs. BAF decomposes LoRA updates into spectral channels and measures their alignment with the principal subspace of the pretrained backbone. Channels strongly aligned with this subspace are retained as generalizable adaptations, while weakly aligned channels are suppressed as potential carriers of memorized content. Experiments on multiple datasets and diffusion backbones demonstrate that BAF consistently reduces memorization while preserving or even improving generation quality. Our code is available in the supplementary material.
arXiv:2605.13974v2 Announce Type: replace
Abstract: Diffusion Transformers (DiTs) and related flow-based architectures are now among the strongest text-to-image generators, yet the internal mechanisms through which prompts shape image semantics remain poorly understood. In this work, we study massive activations: a small subset of hidden-state channels whose responses are consistently much larger than the rest. We show that, despite their sparsity, these few channels effectively draw the whole picture, in three complementary senses. First, they are functionally critical: a controlled disruption probe that zeroes the massive channels causes a sharp collapse in generation quality, while disrupting an equally-sized set of low-statistic channels has marginal effect. Second, they are spatially organized: restricting image-stream tokens to massive channels and clustering them yields coherent partitions that closely align with the main subject and salient regions, exposing a structured spatial code hidden inside an apparently outlier-like subspace. Third, they are transferable: transporting massive activations from one prompt-conditioned trajectory into another, shifts the final image toward the source prompt while preserving substantial content from the target, producing localized semantic interpolation rather than unstructured pixel blending. We exploit this property in two use cases: text-conditioned and image-conditioned semantic transport, where massive activations transport enables prompt interpolation and subject-driven generation without any additional training. Together, these results recast massive activations not as activation anomalies, but as a sparse prompt-conditioned carrier subspace that organizes and controls semantic information in modern DiT models.
arXiv:2607.06183v1 Announce Type: new
Abstract: Mining software repositories at global scale founders on author identity: the same developer commits under many name/email strings, and the same string is reused by many developers. We release a curated author-identity map for World of Code (WoC) version V2604, covering all 5,866,595,698 commits. It ships four co-versioned artifacts: a global alias map (a2AFullSUG) folding 106,826,059 raw author/committer strings into canonical identities; a per-identity classification (A2clsFull) tagging each id good, bad-by-attribute, local, bot, or partial; a within-project table (P2aAFull) recovering low-quality ids inside the one project where their reuse is unambiguous; and a commit-to-identity table (c2AFull) tagging every commit with its resolution provenance. The map is mega-cluster free, its largest cluster 6,910 ids (one GitHub noreply identity), and it resolves 73.5% of six billion commits into multi-id identities, raising human-id commit coverage to 98.17%. The design problem is clumping, not recall: a naive transitive union over shared-attribute edges welds three million unrelated people into one cluster, an over-merge that recall-only benchmarks price at zero. We report both error families, splitting and clumping, and show the high precision claimed by global-scale union maps can be an artifact of never measuring the conflated region. Against the ALFAA human-rated gold set the map scores recall 0.70 / precision 0.88, where the prior WoC map's apparent 0.95 precision collapses to 0.52 once its 3,006,318-id mega-cluster is counted. A canonical software-author identity is also a cross-corpus join key to scholarly author graphs, where clumping is again the binding constraint. All artifacts ship with the WoC V2604 release and a self-contained replication package.
arXiv:2607.05984v1 Announce Type: new
Abstract: Recovering the exact directed acyclic graph (DAG) in linear non-Gaussian acyclic models with latent confounders (LvLiNGAM) remains a challenging problem. Although LvLiNGAM is identifiable only up to an observational equivalence class, each equivalence class is characterized by a unique sparsest DAG. Recovering the sparsest DAG from finite samples, however, remains difficult. Although existing methods are asymptotically consistent, they do not provide an explicit finite-sample procedure for recovering the unique sparsest DAG, nor do they handle models with an arbitrary number of latent confounders.
In this paper, we propose a finite-sample method for recovering the sparsest DAG without imposing any restriction on the number of latent confounders. Simulation studies and real-data analyses demonstrate that the proposed method achieves superior finite-sample performance compared with existing approaches.
arXiv:2607.05988v1 Announce Type: new
Abstract: Multispectral image(MSI) and hyperspectral image(HSI) object tracking object tracking exploits recorded band-wise observations to improve target--background discrimination under similar RGB appearance, mixed pixels, illumination variation, occlusion, and clutter. However, existing trackers commonly process all search regions through a fixed capacity spectral--spatial path, ignoring that tracking difficulty varies substantially across frames and target states. Clear regions may require only lightweight local discrimination, whereas ambiguous boundaries and spectrally similar distractors often demand stronger contextual reasoning. To address this limitation, we propose SpecTrack, a spectral--spatial complexity-aware tracker that formulates MSI tracking as search-region-level adaptive capacity allocation. Its core component, the Spectral Adaptive Mixture-of-Experts (SAMoE) module, provides a capacity-ordered expert pool with progressively increasing latent rank, receptive field, and depth. Expert selection is guided by a Spectral Prompt Router, which fuses semantic context, spatial boundary cues, and a latent channel-variation cue computed after multispectral patch embedding to activate a sparse subset of SAMoE experts for each search region. In parallel, a Shared Global Expert supplies common latent spectral--spatial context to reduce fragmented sparse-routing decisions. Experiments on MUST, MSITrack, and HOTC20 demonstrate a favorable accuracy--efficiency trade-off. The accuracy-oriented SpecTrack-L384 achieves state-of-the-art or highly competitive AUCs of 65.2\%, 51.9\%, and 72.6\% on the three benchmarks, while the balanced SpecTrack-B224 reaches 62.4\% AUC at 43.7 FPS on MUST. An additional GOT-10k evaluation indicates RGB-domain architectural generalization, with SpecTrack-L384 achieving 79.3\% AO.
arXiv:2607.05989v1 Announce Type: new
Abstract: The convergence of Information Technology and Operational Technology has exposed Industrial Control Systems (ICS) to multi-stage cyberattacks that traverse software, network, and physical process layers simultaneously. Although Provenance-based Intrusion Detection Systems (PIDS) are effective in Information Technology (IT) environments, their applicability to Industrial Cyber-Physical Systems (CPS) remains largely unexplored because of the absence of datasets that jointly capture host-level causal behavior, industrial network semantics, and physical process state. To address this gap, we design an open-source, Hardware-in-the-Loop (HIL) CPS testbed that replicates an industrial chemical reactor control architecture across the Purdue model layers. Using this testbed, we propose ProvICS, a multimodal provenance dataset purpose-built for CPS intrusion detection, which synchronously captures four streams: whole-system provenance graphs from the supervisory host and the resource-constrained PLC, decoded Modbus deep-packet inspection records, and physical process telemetry. The collection comprises a 48-hour benign phase and a 22-hour attack phase across four campaigns covering 20 ICS ATT&CK techniques over 32 attack events, ranging from reconnaissance to physical process manipulation. Comparative analysis shows that ProvICS is among the few existing ICS/CPS benchmarks with multi-host kernel-level provenance, real PLC hardware-in-the-loop execution, decoded Modbus traffic, physical process-state measurements, and auxiliary raw PCAP traces in a time-synchronized collection. Baseline detection further confirms that cross-modal fusion can detect all 32 labeled attack events (F1 = 0.913, false-positive rate (FPR) = 1.40%), demonstrating the dataset's ability to expose complementary attack signals across modalities and addressing a gap not covered by prior benchmarks.
arXiv:2607.05938v1 Announce Type: new
Abstract: Synthesizing hand motion that matches the full body motion and the semantic labels is a difficult task due to their high degrees of freedom and the lack of semantic labels. To cope with this issue, we propose a prior-first, condition-second framework for body-conditioned hand motion completion. Our framework first learns a generic body-hand kinematic prior from large-scale unstructured and unlabeled motion data, capturing the intrinsic coordination between global body dynamics and hand articulation. Semantic control is then introduced through lightweight adaptation on top of the frozen prior, avoiding the need to relearn kinematic structure for each control interface. Our framework centers on a streaming, autoregressive body-hand prior that generates coherent, kinematically consistent hand motion from body dynamics in real time, using structured kinematic modeling to maintain mechanical body-hand coupling. To enable practical controllability under limited supervision, we introduce semantically-layered adapters that inject conditioning signals at appropriate kinematic levels, supporting both self-supervised attribute control and weakly supervised text-driven control with only a few hours of labeled data. Extensive evaluations demonstrate that our framework improves kinematic plausibility, robustness, and controllability compared to end-to-end conditioned baselines, particularly in low-resource and cross-dataset settings. We further showcase real-time inference and an interactive authoring workflow, highlighting the applicability to production animation pipelines. Homepage: https://AIGAnimation.github.io/HandPrior/