arXiv:2511.16478v2 Announce Type: replace
Abstract: Music Recommender Systems (MRSs) have long relied on an information retrieval framing, where progress is measured mainly through accuracy on retrieval-oriented subtasks. While effective, this reductionist paradigm struggles to address the deeper question of what makes a good recommendation. Attempts to broaden evaluation, through user studies or fairness analyses, have had limited impact. The emergence of Large Language Models (LLMs) disrupts this framework: LLMs are generative rather than ranking-based, making standard accuracy metrics questionable. They also introduce challenges such as hallucinations, knowledge cutoffs, non-determinism, and opaque training data, rendering traditional train or test protocols difficult to interpret. At the same time, LLMs create new opportunities, enabling natural language (NL) interaction and even allowing models to act as evaluators. This work argues that the shift toward LLM-driven MRSs requires rethinking evaluation. We first review how LLMs reshape user modeling, item modeling, and NL-based recommendation in music. We then examine evaluation practices from NLP, highlighting methodologies and open challenges relevant to MRSs. Finally, we synthesize insights, focusing on how LLM prompting applies to MRSs, to outline a structured set of success and risk dimensions. Our goal is to provide the MRSs community with an updated, pedagogical, and cross-disciplinary perspective on evaluation.
Science Journals
arXiv:2509.02522v3 Announce Type: replace
Abstract: Recent advances in Reinforcement Learning with Verifiable Rewards (RLVR) have empowered large language models (LLMs) to tackle challenging reasoning tasks such as mathematics and programming, however existing RLVR methods often suffer from sparse reward signals and unstable policy gradient updates inherent to RL-based approaches. To address the challenges, we propose PACS, a novel RLVR framework that achieves imPlicit Actor Critic coupling via a Supervised learning framework. By treating the outcome reward as a predictable label, we reformulate the RLVR problem into a supervised learning task over a score function parameterized by the policy model and optimized using cross-entropy loss. A detailed gradient analysis shows that this supervised formulation inherently recovers the classical policy gradient update while providing more stable and efficient training. Extensive experiments demonstrate that PACS significantly outperforms strong open-source models and RLVR baselines, yielding substantial average gains of +8.26% (4B) and +9.57% (8B) over base models offering a promising avenue for LLMs post-training with verifiable rewards. Our code and data are available as open source at https://github.com/ritzz-ai/PACS.
arXiv:2512.05292v3 Announce Type: replace
Abstract: In commercial robotic systems, it is common to encounter a closed inner-loop torque controller that is not user-modifiable. However, the outer-loop controller, which sends kinematic commands such as position or velocity for the inner-loop controller to track, is typically exposed to users. In this work, we focus on the development of an easily integrated add-on at the outer-loop layer by combining disturbance rejection control and robust control barrier function for high-performance tracking and safe control of the whole dynamic system of an industrial manipulator. This is particularly beneficial when 1) the inner-loop controller is imperfect, unmodifiable, and uncertain; and 2) the dynamic model exhibits significant uncertainty. Stability analysis, formal safety guarantee proof, and hardware experiments with a PUMA robotic manipulator are presented. Our solution demonstrates superior performance in terms of simplicity of implementation, robustness, tracking precision, and safety compared to the state of the art. A demonstration video is available at https://youtu.be/e0palGVU_50.
arXiv:2512.05610v2 Announce Type: replace
Abstract: Laser scanning has proven to be an invaluable tool in assessing the decomposition of forest environments. Mobile laser scanning (MLS) has shown to be highly promising for extremely accurate, tree level inventory. In this study, we present NormalView, a projection-based deep learning method for classifying tree species from point cloud data. NormalView embeds local geometric information into two-dimensional projections, in the form of normal vector estimates, and uses the projections as inputs to an image classification network, YOLOv11. In addition, we inspected the effect of multispectral radiometric intensity information on classification performance. We trained and tested our model on high-density MLS data (7 species, ~5000 pts/m2), as well as high-density airborne laser scanning (ALS) data (9 species, >1000 pts/m2). On the MLS data, NormalView achieves an overall accuracy (macro-average accuracy) of 95.5 % (94.8 %), and 91.8 % (79.1 %) on the ALS data. We found that having intensity information from multiple scanners provides benefits in tree species classification, and the best model on the multispectral ALS dataset was a model using intensity information from all three channels of the multispectral ALS. This study demonstrates that projection-based methods, when enhanced with geometric information and coupled with state-of-the-art image classification backbones, can achieve exceptional results. Crucially, these methods rely only on geometric information, and thus are compatible with most sensors. Additionally, we publically release the MLS dataset used in the study, containing 1915 samples.
Cognitive-YOLO: LLM-Driven Architecture Synthesis from First Principles of Data for Object Detection
arXiv:2512.12281v2 Announce Type: replace
Abstract: General-purpose object detection algorithms face a domain-shift bottleneck when deployed in vertical scenarios, and manually customizing a network architecture relies heavily on expert experience with high trial-and-error cost. To address these problems, we present \textbf{Cognitive-YOLO}, an adaptive object-detection model synthesis system driven by a large language model (LLM) collaborating with an autonomous agent. Cognitive-YOLO builds a fully automated \emph{Analyze--Synthesize--Compile} pipeline. First, a data-profiling module quantitatively extracts dataset features, guiding a ReAct-paradigm agent to autonomously retrieve matching network modules from a curated component library. Second, the LLM performs logical reasoning and topological assembly conditioned on the data features, producing a specification-compliant neural architecture description. Finally, a backend hybrid compiler dynamically instantiates the model, closing the loop with end-to-end training and evaluation scheduling. Cross-validation on datasets including rail-surface defects and rice diseases shows that the lightweight models synthesized by Cognitive-YOLO markedly compress the parameter count (down to 1.9M) while matching or surpassing mainstream baselines such as YOLOv11n in mAP@0.5:0.95, effectively balancing model compactness and feature-representation capability.
arXiv:2512.12923v2 Announce Type: replace
Abstract: The efficacy of unmanned aerial vehicle (UAV) swarm cooperative perception fundamentally depends on three-dimensional (3D) formation geometry, which governs target observability and sensor complementarity. In the literature, the exploitation of formation geometry and its impact on UAV sensing have rarely been studied, which can significantly degrade multimodal cooperative perception in scenarios where heterogeneous payloads (vision cameras and LiDAR) should be geometrically arranged to exploit their complementary strengths while managing communication interference and hardware budgets. To bridge this critical gap, targeting the active tracking phase where a prior target estimate is available, we propose an information-optimal optimization framework that optimizes the allocation of UAVs and multimodal sensors, configures formation geometries, and executes flight control. The UAV-sensor allocation is optimized by the Fisher Information Matrix (FIM) determinant maximization. Under this framework, we introduce an equivalent formation transition strategy that enhances field-of-view (FOV) coverage and reduces dynamic communication interference without compromising perception accuracy. Furthermore, we design a novel Lyapunov-stable flight control scheme with logarithmic potential fields to generate energy-efficient trajectories for formation transitions. Extensive simulations demonstrate that our formation-aware design achieves 25.0\% improvement in FOV coverage, 104.2\% enhancement in communication signal strength, and 84.7\% reduction in energy consumption compared to conventional benchmarks. A high-fidelity Gazebo validation further shows up to 20.2\% gain in perception accuracy over hemispherically sampled formations. These results suggest that task-driven geometric allocation is an important rather than incidental component in next-generation UAV swarm perception systems.
arXiv:2607.17959v1 Announce Type: new
Abstract: Onshore wind is expected to supply much of the growth in renewables, yet its expansion increasingly extends into forests, where wind generation and forest conservation act as competing climate strategies for the same land. Net-zero system assessments typically treat forests homogeneously. Here, we grade forest exclusions by ecological value and couple them with a high-resolution net-zero electricity system model, representing onshore wind on forested and non-forested land as two distinct technologies. Applying a set of ecologically graded scenarios to Norway, we quantify how forest protection by its ecological value reshapes system design, the balance of onshore wind within and outside forests, and the robustness of these outcomes to complementary energy policy levers. At equal cost, the system treats forest as its preferred onshore resource, with forest wind comprising 60 to 90 percent of onshore capacity under permissive scenarios. Excluding forest raises system cost by up to 0.5 percent per GW of foregone forest wind, with diminishing returns as ecological criteria for forest exclusions tighten. Directing wind into forests does not compromise system efficiency, with onshore curtailment remaining largely unchanged at 3 to 8 percent. Land-use and technology decisions that appear independent may, in fact, interact, as solar deployment reshapes the value of each wind resource and determines which is displaced from the generation mix. By quantifying the ecological value forgone per unit of forest onshore wind, this study makes explicit a trade-off that energy system modelling usually leaves implicit, providing a transferable basis for balancing the competing climate strategies of protecting forests and sustaining wind expansion.
arXiv:2607.17931v1 Announce Type: new
Abstract: We extend the algorithmic framework of progressive exploration [Fabia\'nski et al., STACS 2019], which yields simple, yet surprisingly general and efficient parameterized algorithms for Dominating Set, Independent Set, and some of their variants. While they identified stability and the Helly property as necessary for their approach, we show that -- with a simple change -- in the case of Dominating Set, one can get rid of the stability requirement. This yields a fixed-parameter tractable algorithm on exactly those graph classes which do not contain long co-matchings or double-ladders as semi-induced subgraphs. Lifting one of these two restrictions makes Dominating Set W[1]-hard on these classes. Our algorithm generalizes results on weakly $\gamma$-closed graphs, and results from Sparsity theory, e.g., nowhere dense and biclique-free classes. At the same time, we match the time complexity of the previously known algorithms on those classes. We demonstrate that this technique can easily be applied to the Distance-$r$ Dominating Set and the Set Cover problem.
arXiv:2510.23614v2 Announce Type: replace
Abstract: One of the major starting points of discrete optimization is the theorem of Nash-Williams and Tutte on the existence of $k$ disjoint spanning trees of a graph, along with its counterpart on the existence of $k$ forests covering all edges of the graph. These elegant results triggered comprehensive research that gave rise to far-reaching generalizations and found applications in seemingly distant areas. Our first goal is to elucidate some aspects of these developments with the hope that the story finds its way to non-experts. But we hope that experts will also find some novelty in our exposition.
arXiv:2607.17416v1 Announce Type: new
Abstract: We derive residual-based a posteriori error estimates for parabolic surface PDEs on closed evolving surfaces. The main contribution is to prove efficiency and reliability for the proposed error indicator, which bounds the error quantities globally from above and globally in space and locally in time from below. We extend methods for adaptivity for parabolic PDEs on stationary surfaces to allow for non-trivial coarsening on evolving surfaces. Multiple numerical experiments are given, which illustrate the asymptotic behaviour of the error and effectiveness of the refinement and coarsening.
arXiv:2607.17550v1 Announce Type: new
Abstract: Modern machine learning (ML) pipelines depend heavily on third party libraries for graph compilation and hardware acceleration. While current practices audit data and model artifacts or rely on file integrity checks, the execution environment remains implicitly trusted. This blind spot enables active threats where a malicious runtime module interacts directly with live training and inference dynamics: exploiting this interaction allows the Trojan to support complex objectives that are challenging for static code or binary modifications, achieving manipulations impossible for standard data and model level attacks. We expose this vulnerability by presenting (A)iSpy, a parasitic infrastructure Trojan that subverts ML systems through an active observe and execute paradigm. Operating within the computation graph, (A)iSpy monitors transient tensor states to perform targeted, stealthy manipulations with negligible overhead. To violate confidentiality, the Trojan identifies all critical training hyperparameters and covertly exfiltrates them via model weights or output logits. To break integrity, it acts as a gradient amplifier: by observing steganographic triggers, it transforms otherwise weak data poisoning into effective backdoor attacks, increasing success rates from near zero to 100%. We further demonstrate broad extensibility across the machine learning lifecycle by validating auxiliary attacks in the appendix, including subpopulation label flipping, availability disruptions, and inference stage manipulations. Importantly, the (A)iSpy module easily evades standard malware scanners, while the associated poisoned inputs and resulting compromised models bypass typical inspection tools. We demonstrate the practicality of this threat with an implementation in the ONNX Runtime training and inference engines.
arXiv:2512.13247v2 Announce Type: replace
Abstract: This paper presents STARCaster, an identity-aware spatio-temporal video diffusion model that addresses both speech-driven portrait animation and dynamic viewpoint control, given an identity embedding or reference image, within a unified framework. Existing 2D speech-to-video diffusion models depend heavily on reference guidance, leading to limited motion diversity. At the same time, 3D-aware animation typically relies on inversion through pretrained tri-plane generators, which often leads to imperfect reconstructions and identity drift. We rethink reference- and geometry-based paradigms in two ways. First, we deviate from strict reference conditioning at pretraining by introducing softer identity constraints. Second, we address 3D awareness implicitly within the 2D video domain by leveraging the inherent multi-view nature of video data. STARCaster adopts a compositional approach progressing from ID-aware motion modeling, to audio-visual synchronization via lip reading-based supervision, and finally to novel view animation through temporal-to-spatial adaptation. To overcome the scarcity of 4D audio-visual data, we propose a decoupled learning approach in which view consistency and temporal coherence are trained independently. Comprehensive evaluations demonstrate that STARCaster generalizes effectively across tasks and identities, consistently surpassing prior approaches in different benchmarks.
arXiv:2510.04748v3 Announce Type: replace
Abstract: The prevalence of online hate and abuse is a pressing global problem. While tackling such societal harms is a priority for research across the social sciences, it is a difficult task, in part because of the magnitude of the problem. People's engagement with reporting mechanisms ('flagging') online is an increasingly important part of monitoring and addressing harmful content at scale. However, users may not flag content routinely enough, and when users do engage, they may be biased by group identity and political beliefs. Across five well-powered and pre-registered online experiments, we examine the extent of ingroup bias in people's flagging of hate and abuse in four different intergroup contexts: political affiliation, vaccination opinions, beliefs about climate change, and stance on abortion rights. Overall, participants reported abuse reliably, with approximately half of the abusive comments in each study reported. However, a pervasive ingroup bias was present whereby across studies, participants were between 17% and 63% more likely to flag abuse directed at the ingroup than at the outgroup. Our findings offer new insights into the nature of user flagging online, an understanding of which is crucial for enhancing user intervention against online hate and thus ensuring a safer online environment.
arXiv:2607.18088v1 Announce Type: new
Abstract: Standard evaluation of many recognition systems contains distribution shift by construction, since benchmarks place disjoint conditions in the training and test splits. Under such a shift, split conformal prediction keeps marginal coverage near the nominal level while per-class coverage fails silently: on a real cross-subject skeleton benchmark, marginal coverage stays near ninety percent, the worst action class is covered about seventy percent of the time, and ten of the sixty classes fall below eighty percent coverage.
We characterize the cost of restoring per-class validity. First, an impossibility: once the shift acts jointly on the covariates and the labels, the target class-conditional score law is unidentified from source labels and an unlabeled target sample, so no label-free method attains per-class coverage that is at once valid and efficient. Second, we make the cost precise: per-class validity alone needs only a handful of target labels per class, while the label count necessary and sufficient for validity together with per-class efficiency grows as the inverse square of the efficiency tolerance and the logarithm of the number of classes, with matching upper and lower bounds. Third, within the evaluated prediction-powered inference family, even the most favorable use of the classifier's own pseudo-labels on an unbounded unlabeled target pool improves efficiency by at most a small constant factor where coverage collapses.
Skeleton action recognition is our real-data case study. A per-class calibration using source labels alone recovers a substantial share of the per-class gap while the shift preserves marginal coverage, and stops helping exactly when marginal coverage itself breaks. Three real shifts of increasing severity trace this boundary, and the same collapse and recovery appears on a natural-image corruption benchmark, beyond any single modality.
arXiv:2607.18109v1 Announce Type: new
Abstract: Understanding human emotions in spoken conversations is a key challenge in affective computing, with applications in empathetic AI, human computer interaction, and mental health monitoring. However, existing datasets vary in scale, emotion distribution, modality alignment, and data partitioning strategies, which can influence reliable cross-dataset generalization and minority-emotion modeling. We introduce SpEmoC a Speaking segment Emotion for Conversations comprising 306,544 raw clips from 3,100 English language movies and TV series. From these, 30,000 high quality, class balanced clips are curated, featuring synchronized visual, audio, and textual modalities annotated for seven emotions through a hybrid pipeline that integrates pretrained models with human validation. SpEmoC uses strict movie- and series-level splits to prevent content overlap between split sets, allowing more reliable evaluation of model generalization. The dataset also maintains a near-balanced distribution across seven emotions, including minority classes such as Fear and Disgust, which supports more balanced learning across categories. Extensive experiments, including in-domain benchmarking, cross-dataset transfer, low-data training, class-imbalance analysis, and modality transfer show that balanced data and careful splitting lead to more stable performance across emotions when models are evaluated on other datasets. These results highlight the importance of dataset design for robust and transferable multimodal emotion recognition.
Graph-Induced Tensor Liftings for Networked SEIR Models: Dimensional Reduction and Residual Analysis
arXiv:2607.17664v1 Announce Type: new
Abstract: Networked SEIR models describe epidemic spread within and between interacting subpopulations through contact-supported nonlinear transmission. Standard polynomial liftings based on complete ordered Kronecker tensors yield linear higher-dimensional representations, but their dimensions grow rapidly because they retain interactions absent from the transmission graph. This paper develops a graph-induced tensor lifting whose observables are selected from the effective transmission support.
An exact edge-based quadratic representation separates linear compartmental transitions from nonlinear infection terms. A homogeneous hierarchy is then constructed recursively. The quadratic transmission field generates the next degree. The linear compartmental field saturates the resulting dictionary within that degree. The first edge-closure dynamics are linear up to an explicit cubic truncation residual, and higher-order truncations contain only next-degree terms.
The first lifted dimension scales with the numbers of subpopulations and effective transmission channels. At fixed order, graph-induced dictionaries grow linearly with network size under uniformly bounded local connectivity, whereas complete polynomial liftings retain order-dependent polynomial growth. Uniform first edge-closure residual bounds depend on the transmission rate and the maximum weighted incoming transmission intensity. Numerical illustrations compare equal intensity per active channel with equal total incoming intensity. They confirm that dictionary dimensions depend only on graph support, whereas residual trajectories also reflect weight accumulation, weight distribution, and nonlinear propagation. These results provide a structured basis for reduced modeling and subsequent model-specific analysis and control.
arXiv:2607.17769v1 Announce Type: new
Abstract: Recent sign language generation (SLG) systems increasingly output dense 3D body representations, which better preserve full-body kinematics and geometry for downstream embodiment on humanoid robots. However, these generated motions frequently exhibit self-intersections such as hand-hand and hand-torso penetration. While such artifacts may be tolerated in offline rendering, they become critical in humanoid execution as they lead to infeasible inverse-kinematics (IK) solutions, collisions, and unstable retargeted trajectories. We present a system-level framework that bridges SLG outputs to humanoid joint-space execution via two components. First, we introduce a volumetric SMPL-X collision-mitigation module that projects generated signing motions toward physically plausible configurations while minimally deviating from the original trajectory. Second, we propose a vision-language-guided retargeting algorithm built on an IK backbone: a VLM serves as a visual critic over rendered humanoid motion, identifies embodiment-specific failure modes, and triggers targeted task-space corrections. Our results highlight collision handling and perception-guided refinement as key missing components for reliable humanoid signing.
arXiv:2512.15783v4 Announce Type: replace
Abstract: This paper proposes a measurement standardisation framework that compresses expert-AI interactions into structured, comparable fields for prospective risk detection in deployed AI systems, without access to model internals. This concept paper defines the framework's scope, semantically and statistically, and specifies a protocol for its empirical testing. The population-level claims it is designed to support therefore belong to a staged research programme rather than to results claimed here. Measurement standardisation underpins three claims. The first is a reliability claim: under bounded conditions, large language models can produce reliable, standardised assessments of the evidential and policy alignment of expert-AI interactions. The second is a governance claim: alignment scores give experts an immediate signal during deployment and give institutions a basis for monitoring alignment patterns across mission types, models, and domains. The third is an outcome validation claim: once measurement standardisation is established, aggregate alignment scores could be used to study associations with downstream outcomes in regulated professional settings. This introduces the possibility of an "AI epidemiology", a form of risk detection based on correlated variables instead of mechanistic analysis, inspired by epidemiological reasoning. A minimal application of the protocol to a published expert-AI corpus shows that the judge reproduces its policy and evidential alignment scores across two runs under the specified conditions. Judge reliability at scale remains to be validated in future work. The paper sets out a defined grammar of eight interaction fields, together with a statistical protocol based on paired bootstrap inference, DeLong's test for paired AUCs as a sensitivity check, a pre-specified one-sided non-inferiority margin of 0.05, and Holm-Bonferroni correction.
arXiv:2607.17442v1 Announce Type: new
Abstract: Background. Predicting conversion from mild cognitive impairment (MCI) to Alzheimer's disease (AD) is central to trial enrichment and care planning, yet existing models provide no individual-level uncertainty estimates and rarely include transparent leakage audits. We introduce the first application of persistent homology to longitudinal clinical trajectory point clouds for this task, and the first split-conformal individual risk guarantee for any AD-conversion model. Methods. We analysed 741 MCI subjects (240 converters, 32.4%) from ADNI with a uniform 4-year follow-up cap. Five leakage sources were corrected; without them a naive pipeline achieved AUC=0.934, inflated by +0.075. Vietoris-Rips persistent homology and sublevel-set proxies were combined with trajectory slopes and engineered features (76 total) in a stacking ensemble evaluated by 5-fold cross-validation. Results. Cox and Random Survival Forest models with TDA features achieved concordance C=0.799 and C=0.826 versus C=0.753 and C=0.812 without (+0.045 and +0.014). The primary nested AUC is 0.840 (same-fold bound 0.866); external AUC was 0.879 on a zero-overlap ADNI-2/GO/3 cohort. H0 persistence entropy was the top SHAP feature and significantly associated with APOE4 dosage (Spearman r=-0.191, p<0.0001, Bonferroni-corrected). Cross-conformal coverage was 90.4%+-2.2% (target 90%); empirical external coverage 96.9%. Maximum fairness gap in false-negative rate across seven subgroups was 0.092. Conclusions. We propose H0 persistence entropy as a topological biomarker of cognitive decline and demonstrate that a leakage-audited, conformally calibrated pipeline reaches competitive accuracy with individual-level uncertainty quantification not previously available for this task.
arXiv:2607.18114v1 Announce Type: new
Abstract: Modern LLMs are alarmingly susceptible to surprisingly simple immaterial changes of input prompts: a casual hint, an incorrectly labeled few-shot example, or a fake prior assistant turn often flips an originally correct answer. We study where this susceptibility, spanning sycophancy and related cue-induced biases, lives inside the model. Across five model families and seven BCT bias types, we extract a per-bias direction from hidden states and triangulate it through three measures: probing, leave-one-dataset-out transfer, and causal intervention. The susceptibility is largely installed by alignment tuning rather than pretraining: pretrained base models barely cave to these biases, and their activations carry no cue-specific signal beyond question content. Within aligned models, each bias becomes a single coherent direction that we can both decode and steer along, recovering the unbiased answer across every family we test. The biases stay representationally distinct, however: cross-bias entanglement is model-specific rather than a property of the bias category, and even behaviorally similar biases occupy different directions. The same intervention also serves as a modest debiasing tool, recovering a meaningful share of bias-induced errors while preserving most correct answers across all instruct families. Cue-induced bias is therefore best understood not as a single flaw in LLMs but as a family of distinct, causally active directions that alignment tuning installs.
arXiv:2607.18116v1 Announce Type: new
Abstract: Recent work leverages Large Language Models (LLMs) to generate executable code for pedagogical animations using libraries such as Manim. However, ensuring spatial correctness and visual legibility remains challenging, as existing frameworks emphasize pedagogical content while overlooking geometric occlusions. We propose the Symbolic Geometric Agent (SGA), a plug-and-play module for code-centric animation pipelines that intercepts LLM-generated code, performs partial execution to extract symbolic scene graphs, and applies targeted refinement when spatial conflicts are detected. We further introduce the Manim Visual Quality Score (MVQS), a deterministic rendering-free proxy for spatial integrity. Experiments on the MMMC-Code benchmark across four LLM backbones and two agentic pipelines show that SGA achieves a peak MVQS of 73.11 (Code2Video + GPT-5.1), corresponding to a 16.1% relative improvement over the raw baseline, and improves MVQS in 7 of 8 backbone x pipeline configurations.
arXiv:2509.25624v3 Announce Type: replace
Abstract: As LLMs advance into autonomous agents with tool-use capabilities, they introduce security challenges that extend beyond traditional content-based LLM safety concerns. This paper introduces Sequential Tool Attack Chaining (\STAC), a novel multi-turn attack framework that exploits agent tool use. \STAC chains together tool calls that each appear harmless in isolation but, when combined, collectively enable harmful operations that only become apparent at the final execution step. At the core of \STAC is an automated, closed-loop pipeline that synthesizes executable multi-step tool chains, validates them through in-environment execution, and reverse-engineers stealthy multi-turn prompts that reliably induce agents to execute the verified malicious sequence. Using this framework, we generate and systematically evaluate 483 \STAC cases, featuring 1,352 sets of user-agent-environment interactions and spanning diverse domains, tasks, agent types, and 10 failure modes. Our evaluations show that state-of-the-art LLM agents are highly vulnerable to \STAC, with an average final attack success rate (ASR) of 91.2\% -- exceeding 90\% for all but one of the eight agents evaluated. We further perform defense analysis and find that existing prompt-based defenses provide limited protection. To address this gap, we propose a new reasoning-driven defense prompt that achieves the strongest initial-turn protection, cutting ASR by up to 28.8\%; however, this advantage erodes sharply under adaptive attacks, and an experience-based defense (ToolShield) proves more durable over sustained multi-turn interactions. These results highlight a crucial gap: defending tool-enabled agents requires reasoning over entire action sequences and their cumulative effects, rather than evaluating isolated prompts or responses.
arXiv:2512.18953v3 Announce Type: replace
Abstract: Symmetry is a strong prior present in many object categories, yet standard benchmarks for 3D generative models rarely report whether this prior is preserved. We study symmetry preservation in unconditional point cloud generation. We first audit the symmetry of generated shapes by several 3D generative models and compute a normalized symmetry score based on the Chamfer Distance (CD). We show that although current 3D generative models achieve competitive results under standard evaluation, they reveal a persistent symmetry gap when a symmetry-aware evaluation protocol is applied. To test whether this gap is merely inherited from the training data, we evaluate these models over a mirrored-objects dataset derived from ShapeNet and analyze symmetry dynamics during training. Mechanism-inspired diagnostic tests were conducted at the sampling and latent-representation levels to further show that reflection symmetry is not reliably encoded in the learned generative process. Finally, to address this gap, we propose a data-centric symmetry-based intervention: training generative models on a half-objects dataset and reconstructing full objects by reflection during sampling. Across multiple backbones, this intervention substantially improves geometric consistency and visual plausibility while remaining competitive under standard metrics. These findings suggest that symmetry-aware evaluation is needed alongside standard benchmarks, and future 3D generative models should incorporate this prior explicitly, either during training or sampling.
arXiv:2607.17543v1 Announce Type: new
Abstract: In two-player zero-sum games whose Nash equilibria form a convex set, regularized solvers such as Regularized Nash Dynamics (R-NaD) empirically select the maximum-entropy member: the information projection (I-projection) of a uniform reference onto the Nash set. On a panel of small games this match is exact, with one apparent exception: in Kuhn poker R-NaD lands at bluff coordinate 0.180 while the maximum-entropy member sits at 0.201, a coordinate gap of about 0.021, even though R-NaD attains 99.7 percent of the maximum entropy. We ask whether this gap is a genuine selection bias or an artifact, and answer it quantitatively. We show that for selection on a one-dimensional Nash manifold the coordinate gap factorizes as $\mathrm{gap} \approx \sqrt{2\delta/\kappa}$, where $\delta$ is the entropy shortfall of the solver and $\kappa$ is the curvature of the entropy landscape at its peak. Across five games this relation holds to within $2 \times 10^{-4}$ (under 1 percent relative error). The four matrix games have $\delta \approx 0$ (R-NaD reaches the maximum-entropy member exactly) and therefore no gap regardless of curvature; only the sequential game (Kuhn) has $\delta > 0$. A causal sweep of the magnet strength drives $\delta \to 0$ and the gap toward zero along the predicted curve (fitted scaling exponent 0.50, $R^2 > 0.999999$, against the exact prediction of 1/2), until the dynamics destabilize at a stability floor: behavior consistent with a removable shortfall and inconsistent with a fixed bias. We quantify the curvature half of the law from measured curvatures and flag a moving-target pitfall in the natural Tsallis-entropy experiment. The Kuhn gap is thus the curvature shadow of a small, removable entropy shortfall on an unusually flat peak; the I-projection account is upheld up to a flatness-limited residual.
arXiv:2602.01131v2 Announce Type: replace
Abstract: With the rapid expansion of the low-altitude economy, Unmanned Aerial Vehicles (UAVs) serve as pivotal aerial base stations supporting diverse services from users, ranging from latency-sensitive critical missions to bandwidth-intensive data streaming. However, the efficacy of such heterogeneous networks is often compromised by the conflict between limited onboard resources and stringent stability requirements. Moving beyond traditional throughput-centric designs, we propose a Sensing-Communication-Computing-Control closed-loop framework that explicitly models the impact of communication latency on physical control stability. To guarantee mission reliability, we leverage the Lyapunov stability theory to derive an intrinsic mapping between the state evolution of the control system and communication constraints, transforming abstract stability requirements into quantifiable resource boundaries. Then, we formulate the resource allocation problem as a Stackelberg game, where UAVs (as leaders) dynamically price resources to balance load and ensure stability, while users (as followers) optimize requests based on service urgency. Furthermore, addressing the prohibitive computational overhead of standard Deep Reinforcement Learning (DRL) on energy-constrained edge platforms, we propose a novel and lightweight pruning-based Proximal Policy Optimization (PPO) algorithm. By integrating a dynamic structured pruning mechanism, the proposed algorithm significantly compresses the neural network scale during training, enabling the UAV to rapidly approximate the game equilibrium with minimal inference latency. Simulation results demonstrate that the proposed scheme effectively secures control loop stability while maximizing system utility in dynamic low-altitude environments.