arXiv:2606.07244v1 Announce Type: new
Abstract: Vision-Language Navigation in Continuous Environments (VLN-CE) requires agents to follow natural-language instructions while navigating in real-world-like environments. Most VLN-CE approach\-es adopt a three-stage framework: a waypoint predictor proposes navigable waypoints, and a navigator selects the best waypoint, with a low-level controller executing the movement to it. However, this decoupled paradigm often leads to unreachable waypoints or inconsistencies between planning and control. In this work, instead of predicting isolated waypoints, we introduce a novel paradigm called Trajectory Waypoint, which grounds each candidate waypoint in an executable trajectory. To realize this, we design a Trajectory Waypoint Predictor formulated as a TSDF-guided diffusion policy, which steers trajectory generation away from obstacles, inherently ensuring the reachability of the predicted waypoints. We further propose a trajectory-enhanced navigator that injects the associated trajectory as additional information for planning, enabling strict consistency between high-level semantic decisions and low-level execution. Extensive experiments on the VLN-CE benchmark show that our Trajectory Waypoint paradigm achieves superior performance over the baselines.
Science Journals
arXiv:2606.04003v2 Announce Type: replace-cross
Abstract: Root-MUSIC is a spectral estimation algorithm that approximates the unknown signal frequencies by constructing a high-degree polynomial and finding a subset of roots which are closest to the complex unit circle. Previous works found asymptotic expectation formulas for the performance of Root-MUSIC under the implicit assumption that the aforementioned root selection criterion does not select extraneous roots -- those which are unrelated to the correct parameters. This paper removes the need for this assumption by showing all extraneous roots lie outside an annulus of a certain thickness and therefore are not selected by the algorithm. This paper also provides sharp, non-asymptotic, and explicit error bounds for the correct roots in terms of fundamental model parameters. All results hold under a natural separation condition on the correct signal frequencies and are applicable in both the single- and multi-snapshot models. More specifically, in the multi-snapshot model, we prove that Root-MUSIC estimates the frequencies with error at most $O(\sigma /(m \sqrt n))$, where $\sigma^2$ is the noise variance, $m$ is the number of sensors, and $n$ is the number of snapshots. A novelty of this non-asymptotic bound is the explicit $1/m$ decay, which indicates that there is a significant advantage in utilizing additional sensors. Numerical simulations confirm our theory. The main mathematical insight of this paper is a geometric property of the Root-MUSIC polynomial: its correct roots are highly stable to noise while its extraneous roots must lie outside of an annulus.
arXiv:2606.07208v1 Announce Type: new
Abstract: Disturbance rejection is a central objective in process control, particularly when measurable disturbances can be exploited through feedforward action. Although Model Predictive Control (MPC) naturally incorporates disturbance models and prediction capabilities, standard formulations cannot achieve complete disturbance rejection since the cost function penalises control effort. This limitation prevents MPC from reproducing the behaviour of classical feedforward compensators. This work proposes a novel framework to embed true feedforward capabilities within MPC without removing the control effort penalty. The approach introduces a dual-control structure in which two control actions are computed simultaneously: a tracking-oriented action addressing set-point tracking and robustness, and a feedforward-oriented action dedicated to disturbance rejection. Both contributions are combined into a single control signal on which the process constraints are explicitly enforced. The feedforward-oriented action is formulated without penalising control effort, enabling full compensation of measurable disturbances. The methodology is developed for Dynamic Matrix Control (DMC), Generalised Predictive Control (GPC), and state-space MPC. Its effectiveness is demonstrated through simulation studies, including comparisons with standard MPC and classical feedforward schemes. A case study based on a reverse osmosis process shows that the proposed approach improves disturbance rejection while preserving constraint handling and overall control performance.
arXiv:2606.07246v1 Announce Type: new
Abstract: High-Level Synthesis (HLS) enables rapid development of FPGA accelerators, yet achieving high-quality results (QoR) remains challenging due to the large and irregular design space induced by compiler directives (a.k.a pragmas). Selecting effective configurations requires reasoning over complex interactions between program structure, memory behavior, and often conflicting objectives such as latency and resource utilization. Prior model-driven approaches exhibit limited generalization across kernels and fail to capture higher-level optimization intent. Recently, Large Language Models (LLMs) capture code semantics and high-level intent, but their sequential representations hinder modeling of structural dependencies and global trade-offs, leading to suboptimal HLS designs.
We present MailoHLS, a hybrid framework that combines LLM-based semantic reasoning with GNN-based structural modeling for objective-aware directive optimization. By integrating structural embeddings via cross-attention and leveraging PEFT with objective-conditioned LoRA adapters and Pareto-driven optimization, MailoHLS enables joint reasoning over code semantics, structure, and design trade-offs. Across seen and unseen kernels, MailoHLS achieves up to 12.42x and 8.4x speedup (9.48x and 4.97x geometric mean) for latency optimization, consistently producing near-Pareto-optimal designs. On fully unseen applications, it reaches up to 10.2x speedup (6.58x geometric mean), outperforming high-end LLMs and prior approaches while narrowing the gap to the Pareto frontier.
arXiv:2606.07416v1 Announce Type: new
Abstract: Atmospheric plasma spraying (APS) is a widely used coating process in which in-flight particle temperature and velocity strongly influence coating quality. However, these particle characteristics are difficult to monitor continuously during operation, motivating the development of non-invasive data-driven diagnostic methods. In this work, we investigate the predictive potential of high-speed video observations of the plasma plume for estimating in-flight particle characteristics in APS. We introduce three different video-derived feature representations and evaluate them using Tabular Prior-Data Fitted Networks (TabPFN), convolutional neural networks (CNN), and classical regression baselines including Random Forest, Gradient Boosting, Support Vector Regression, and XGBoost. Experiments are conducted using grouped leave-one-out cross-validation on 126 labeled pre- and post-spray video recordings from 63 APS spray runs. Across the engineered feature experiments, TabPFN achieves the most consistent performance for temperature prediction, reaching R2 = 0.86 using the combined feature representation. CNN models particularly perform stronger for velocity prediction, achieving R2 of 0.81. In addition, we evaluate models operating directly on raw video frames using pretrained CNNs and find that the highest performance is achieved by a pretrained CNN with a regression head with R2 of 0.90 and 0.82 for temperature and velocity, respectively. The results demonstrate that video-derived plume information provides a promising and scalable foundation for non-invasive APS diagnostics and real-time process monitoring.
arXiv:2606.04278v2 Announce Type: replace-cross
Abstract: Global infrasound monitoring provides a persistent means of detecting energetic bolide atmospheric entries, complementing optical observations and extending coverage over remote regions. We present a global assessment of the physical factors governing bolide infrasound detectability by correlating 623 bolide events reported by the Center for Near-Earth Object Studies between 2007 and 2025 with waveform data from the International Monitoring System. We identify 311 events with confirmed infrasound detections, corresponding to a detection rate of approximately 50%, substantially higher than inferred from earlier surveys, reflecting both the maturation of the global infrasound network and advances in automated, multi-frequency array processing. Analysis of flight parameters shows that infrasound detectability is selective rather than uniform across the bolide population. Detected events are preferentially associated with steeper entry angles and lower-altitude energy deposition, while shallow, high-altitude trajectories are less consistently observed. Very high-energy events remain detectable regardless of geometry, but for the more common lower-energy regime, observability depends on specific combinations of entry parameters and propagation conditions. This geometric dependence persists across comparable energy ranges and atmospheric conditions, indicating that entry angle exerts a primary control on detectability, with energy and propagation acting as secondary modulating factors. These results provide new physical constraints on bolide-atmosphere interactions and improve interpretation of global infrasound observations for planetary defense and atmospheric-entry studies.
arXiv:2605.21731v2 Announce Type: replace
Abstract: Deep learning models are increasingly used in scientific prediction tasks where strong benchmark performance is often interpreted as evidence of scientifically meaningful behavior. This interpretation is fragile, as models may exploit shortcut features, dataset-specific regularities, or distributional biases that are predictive on held-out data but not aligned with domain-relevant structure. To address this limitation, we introduce the \textsc{I-SAFE} (Interventional Secure, Accurate, Fair and Explainable) framework, a post-hoc distributional auditing framework for scientific AI models centered on the Wasserstein Coherence Metric (WCM). Given a trained black-box predictor and an external structural prior encoding domain knowledge about task-relevant input structure, \textsc{I-SAFE} evaluates raw model outputs under structurally guided perturbations of the input. The proposed audit measures output-distribution coherence through three complementary metrics: a Quantile-Based Metric (QBM) for location-level coherence, the WCM for ordinal coherence, and a translation-invariant WCM variant for shape coherence. We instantiate \textsc{I-SAFE} on drug--target interaction (DTI) prediction using the Davis kinase benchmark, KLIFS (Kinase--Ligand Interaction Fingerprints and Structures) binding-pocket annotations, and three sequence-based DTI models: DeepConvDTI, DeepDTA, and TAPB. Although the models operate in a comparable predictive regime, \textsc{I-SAFE} reveals substantially different distributional response profiles, a distinction invisible to accuracy-based evaluation. The framework is model-agnostic and applicable to any domain where inputs admit a structured decomposition and an external prior is available.
arXiv:2606.07309v1 Announce Type: new
Abstract: Instruction-following audio language models (ALMs) can be augmented with explicit acoustic cues, yet it remains unclear whether such cues are used in a grounded way when the raw audio is already available. We study this question in speech emotion recognition (SER) by deriving six interpretable acoustic concept tokens from the standardised eGeMAPS paralinguistic feature set. These tokens summarise energy, pitch, dynamics, brightness, formants, and voice quality, and are appended to the textual prompt while the audio input is kept unchanged. Across the widely used FAU-Aibo and IEMOCAP benchmarks, aligned tokens improve unweighted average recall (UAR), whereas shuffled, conflicting, or corrupted tokens reduce performance relative to aligned tokens and shift confusions toward neutral. Importantly, predictions do not collapse under strong token perturbations, suggesting that the models are sensitive to the symbolic cue channel but remain partly anchored to the audio signal. We argue that token-only interventions provide a practical way to probe audio-grounded cue use, robustness, and interpretability in ALM-based affective computing.
arXiv:2606.07324v1 Announce Type: new
Abstract: Bright squeezed vacuum (BSV) light has a vanishing mean optical electric field yet can strongly enhance strong-field nonlinear responses beyond the conventional semiclassical paradigm. Here we examine this scenario in the light-matter strong-coupling regime by investigating resonant multiphoton ionization of atoms driven by BSV, using a fully quantum treatment of both the electron and the field. Our results show that the photoelectron energy spectrum exhibits an anomalous Autler-Townes splitting whose magnitude grows with the Above-threshold-ionization (ATI) order, rather than remaining essentially ATI-order independent as in the case of coherent driving. This behavior reflects a general scaling with the number of absorbed photons and originates from the broad photon-number fluctuations of the driving field together with the resulting electron-field entanglement. We further show that the BSV-induced enhancement of ionization yields evolves with intensity, crossing over from the $g^{(p+1)}$ limit to the $g^{(p)}$ limit as Rabi oscillations become established. These results identify a quantum regime of strong-field ionization governed by the interplay of photon statistics, nonlinear transitions, strong coupling, and nonseparable light-matter dynamics.
arXiv:2606.07314v1 Announce Type: new
Abstract: Quantum software bugs often yield silent, incorrect outputs rather than explicit errors, making them particularly difficult to detect and repair with conventional techniques. Although large language models (LLMs) have shown strong performance on classical software engineering tasks, their ability to debug quantum code remains largely unexplored. To bridge this gap, we propose QBugLM, a multi-agent framework that automates the quantum software debugging pipeline, from taxonomy-driven bug injection to LLM-based detection and repair, and finally to simulation-based validation, for framework-agnostic OpenQASM 3.0 programs. We further conduct a comprehensive case study using QBugLM to benchmark two LLMs, Claude 4.6 Sonnet and Qwen3 Coder Next, across different prompting strategies, bug categories, and quantum programs. Our results show that iterative feedback is critical, as a single retry raises Pass@1 from below 25% to above 80%. Moreover, simpler structured prompting can even outperform Chain-of-Thought and ReAct for reasoning-capable models under fixed-resource constraints. Our work takes initial steps toward benchmarking LLM capabilities for debugging quantum programs and offers practical insights to support future efforts in automated quantum software repair.
arXiv:2606.06785v1 Announce Type: cross
Abstract: We study finite-sample change detection for one-dimensional noisy dynamical systems using partition-based empirical approximations of stationary behaviour. Given observations from an interval-valued process, we partition the state space, estimate a finite transition matrix from observed transitions between partition elements, and apply a small Doeblin-type regularisation to ensure a unique stationary distribution.
From an initial reference segment, we compute a baseline empirical stationary distribution \(\widehat{\pi}_{0,\rho}\). For each later sliding window, we compute \(\widehat{\pi}_{t,\rho}\) and define the score \[ S_t=\|\widehat{\pi}_{t,\rho}-\widehat{\pi}_{0,\rho}\|_1. \] Large values of \(S_t\) indicate a change in stationary behaviour relative to the baseline. The statistic detects changes in invariant density or stationary law, but not all possible changes in transition dynamics.
Under explicit assumptions on empirical transition concentration, finite-state stationary distribution stability, partition approximation, regularisation bias, and noise stability, we derive a finite-sample bound for the empirical stationary density. The bound separates sampling error, regularisation bias, partition approximation error, and noise bias. We then obtain a single-window false-alarm guarantee and a sufficient detection condition when the invariant density changes by more than the estimation error. We illustrate the method on synthetic noisy beta-map change-point experiments.
arXiv:2606.06957v1 Announce Type: cross
Abstract: Predicting outputs that are located in non-Euclidean spaces, such as probability distributions, networks, and symmetric positive-definite matrices, is becoming increasingly important in modern data analysis, particularly when inputs are high-dimensional. We propose DeSI (Deep Single-Index Fr\'echet Regression), a semiparametric framework for regression with metric space-valued outputs and multivariate inputs that assumes a single-index structure for the conditional Fr\'echet mean. DeSI estimates an interpretable index direction, which quantifies the relative importance of inputs, using a deep neural network, and performs Fr\'echet regression along the resulting one-dimensional index in the target metric space. This structure mitigates the curse of dimensionality while retaining interpretability, which stands in contrast to standard deep neural networks. We establish theoretical guarantees for DeSI, including uniform approximation and convergence rates, and demonstrate its strong predictive performance through simulations on distributions, networks, and symmetric positive-definite matrices, as well as an application to compositional mood data from New Jersey.
arXiv:2509.01494v2 Announce Type: replace
Abstract: Automated Code Review (ACR) is crucial for software quality, yet existing benchmarks often fail to reflect real-world complexities, hindering the evaluation of modern Large Language Models (LLMs). Current benchmarks frequently focus on fine-grained code units, lack complete project context, and use inadequate evaluation metrics. To address these limitations, we introduce SWRBench , a new benchmark comprising 1000 manually verified Pull Requests (PRs) from GitHub, offering PR-centric review with full project context. SWRBench employs an objective LLM-based evaluation method that aligns strongly with human judgment (~90 agreement) by verifying if issues from a structured ground truth are covered in generated reviews. Our systematic evaluation of mainstream ACR tools and LLMs on SWRBench reveals that current systems underperform, and ACR tools are more adept at detecting functional errors. Subsequently, we propose and validate a simple multi-review aggregation strategy that significantly boosts ACR performance, increasing F1 scores by up to 43.67%. Our contributions include the SWRBench benchmark, its objective evaluation method, a comprehensive study of current ACR capabilities, and an effective enhancement approach, offering valuable insights for advancing ACR research.
arXiv:2511.04567v2 Announce Type: replace
Abstract: Constructing reduced models for turbulent transport is essential for accelerating profile predictions and enabling many-query tasks such as parameter exploration and design optimization. This work investigates machine-learning-driven reduced models for Electron Temperature Gradient (ETG) turbulence in the Wendelstein 7-X (W7-X) stellarator. We develop physics-guided scaling laws to predict the ETG heat flux at seven radial locations as functions of three key plasma parameters: the normalized electron temperature gradient ($\omega_{T_e}$), the ratio of normalized electron temperature and density gradients ($\eta_e$), and the electron-to-ion temperature ratio ($\tau$). The model coefficients are determined through regression combined with an active learning strategy. The procedure initializes the scaling laws using low-cardinality sparse-grid training data and iteratively enriches the training set by selecting maximally informative samples from an existing simulation database. The predictive performance of the models is assessed using out-of-sample datasets comprising more than $393$ points per radial location. Using the coefficients identified at the seven training radial locations, we further derive regression-based parameterizations for the scaling-law coefficients as functions of radial position. The resulting models are then evaluated at three additional radial locations not used during training, including both interpolation and moderate extrapolation cases. Overall, our reduced models demonstrate good predictive performance and achieve accuracy comparable to the original reference simulations, including in interpolation and moderate extrapolation regimes. An important finding is that a single radius-independent model cannot adequately describe ETG transport across the W7-X core, suggesting the presence of geometry-dependent physics not captured by the present formulation.
arXiv:2601.22574v2 Announce Type: replace
Abstract: Although Video Large Multimodal Models have achieved strong performance in video understanding, they still suffer from hallucination. Existing inference-time intervention methods usually modify videos under the contrastive decoding framework, but their heuristic designs bring limited improvements and increase inference latency. To address these issues, we propose ViSSRes, an inference-time intervention method that enhances video representations through a lightweight MLP-style network. Specifically, we use a contrastive random walk approach to characterize the spatiotemporal consistency of video representations, and introduce conditional mutual information to associate video representations with the model's semantic understanding. With the model backbone kept frozen, ViSSRes learns residuals for video representations and optimizes them from both spatiotemporal and semantic consistency perspectives. During inference, ViSSRes requires only a single forward pass and introduces no substantial additional inference cost. Experiments show that ViSSRes reduces the hallucination rate of LLaVA-NeXT-Video on EventHallusion by 40.69% and improves video understanding on MMVU by 18.36% under the CoT setting, demonstrating its effectiveness in mitigating hallucinations.
arXiv:2606.07393v1 Announce Type: new
Abstract: As AI systems transition from experimental prototypes to mission-critical tools, their dependence on dynamic data, evolving models, and governance raises questions about whether existing acquisition pathways can keep pace. The U.S. Department of Defense has modernized its acquisition processes through the Adaptive Acquisition Framework, with the Software Acquisition Pathway (SWP) serving as the primary mechanism for acquiring software-intensive capabilities. This paper evaluates whether SWP is sufficient to address the unique demands of AI acquisition.
In this work, we perform a scenario-based evaluation that traces a notional AI-enabled program through key SWP planning activities to assess how policy translates into program artifacts and decisions. We use Policy Scenario Analysis to examine whether the SWP-centered governance stack provides sufficient actionable support for AI acquisition. The governance stack provides a viable foundation for iterative delivery and AI testing. However, we identify a recurring actionability problem in the core guidance. AI-specific controls for data provenance, lifecycle management, and human oversight remain distributed across supplemental documents rather than embedded in the program-facing mechanisms through which SWP is executed. This disconnect leaves program offices reliant on inconsistent local interpretation. We conclude by recommending an AI-supporting sub-path and targeted artifact refinements to better bridge this policy-to-artifact gap.
arXiv:2606.07433v1 Announce Type: new
Abstract: Video understanding is being rapidly transformed by multimodal large language models (MLLMs), as research moves from short clips to long, multimodal, and knowledge-intensive video scenarios. These scenarios require models to handle sparse evidence, long-range dependencies, multimodal alignment, and reliable inference under limited computational budgets. This work presents a human-view perspective on LLM-based video understanding, organized around three functional abilities: watching, remembering, and reasoning. Rather than treating video tasks as isolated benchmarks, this view provides a unified structure for analyzing how video MLLMs acquire evidence, preserve context, and produce grounded outputs. We introduce a formulation that characterizes video understanding systems by their perceptual representations, memory states, reasoning traces, and final predictions. Based on this formulation, we identify challenges in spatio-temporal perception, efficient long-video processing, memory modeling, streaming understanding, and faithful reasoning. Representative methods are organized by their roles in video MLLM systems. Watching covers fine-grained, comprehensive, audio-visual, and efficient perception. Remembering includes offline and streaming memory, while reasoning covers text-only reasoning and thinking with videos. We further examine application domains such as egocentric, sports, instructional, medical, and narrative videos, and cover training datasets and evaluation benchmarks across task types, supervision formats, modalities, and capability dimensions. Finally, we outline open problems and future directions for scalable, memory-aware, and evidence-grounded video intelligence. Related works will be continuously traced at https://github.com/marinero4972/Awesome-HumanView-VideoUnderstanding.
arXiv:2606.07412v1 Announce Type: new
Abstract: LLM-driven software engineering agents have become a central testbed for real-world language-model capability, yet their training remains limited by the availability of high-quality SWE tasks. Existing synthetic data methods typically create tasks through fixed mutation or bug-injection procedures, making the resulting distributions largely independent of the agent's own weaknesses and training progress. We introduce Socratic-SWE, a closed-loop self-evolution framework that reuses the agent's historical solving traces as a source of training signal. Rather than treating traces only as evidence for reward computation, Socratic-SWE distills them into structured agent skills that summarize recurring failures and effective repair patterns. These skills then guide the generation of targeted repair tasks in real repositories. Candidate tasks are checked through execution-based validation and scored with a solver-gradient alignment reward, so that the retained tasks are both verifiable and useful for improving the Solver. The updated Solver produces new traces, enabling the task curriculum to adapt over successive rounds. Across SWE-bench Verified, SWE-bench Lite, SWE-bench Pro, and Terminal-Bench 2.0, Socratic-SWE consistently improves over self-evolving baselines under the same compute budget, reaching 50.40% on SWE-bench Verified after three iterations. These results suggest that solving traces can serve as a scalable substrate for self-evolving SWE agents.
arXiv:2606.07434v1 Announce Type: new
Abstract: Modern prediction markets face two limitations that restrict their applicability in a range of settings:~(i)~they reveal what the crowd believes but not the evidence or reasoning behind those beliefs, and~(ii)~they require an event with an external ground truth that resolves at a known future date. We address these twin challenges by introducing evidence markets, a generalization of prediction markets that incentivizes the submission of evidence alongside beliefs and can be endogenously resolved using the crowd-sourced evidence if external resolution is not possible. At its core, the market uses a logarithmic market scoring rule whose liquidity parameter changes dynamically with the accumulated evidence quality. We prove that platform loss is bounded, evidence is rewarded proportional to the current market uncertainty, and can be equivalently implemented through an automated market maker. In the case where the marker resolves endogenously based on submitted evidence, we characterize how withholding evidence shifts a trader's belief about resolution and use it to prove truthful belief and evidence reporting is a always an $\varepsilon$-dominant strategy incentive compatible (DSIC) strategy. To address operational considerations, we propose evidence verification via an LLM-as-a-Judge framework with staking and give an asynchronous execution algorithm that is not bottle-necked by verification. Throughout the work, we use LLM evaluations -- determining which model is best for a given task -- as a salient and representative running example for our proposed market.
arXiv:2605.20950v2 Announce Type: replace
Abstract: Vision-Language Models (VLMs) face a bottleneck of prohibitive computational costs arising from massive visual token sequences during inference. Existing vision token reduction methods alleviate this burden, but they unintentionally preserve the isolated visual subject strictly aligned with the user's query, which fails to substantially explore salient subjects and their contextual relationships. In this paper, we propose SPpruner, a subject-centric progressive reduction paradigm that emulates the \textit{Focus-then-Context} mechanism of the human visual perception system. Specifically, we first construct a focus identification module to explicitly model the interplay between visual saliency and semantic relevance. Herein, it can excavate the comprehensive visual subject spectrum to ensure a high-fidelity representation of visual input. Subsequently, a context-aware structural scanning module is developed to aggregate contextual cues from neighboring regions. As such, it can effectively restore global relational dependencies to uphold the structural integrity of the preserved subjects. Extensive experiments demonstrate that our paradigm consistently outperforms SOTA methods, achieving up to 2.53 times speedup with only 22.2% of visual tokens retained in Qwen2.5-VL and a 67% FLOPs reduction on LLaVA with a negligible 0.6% accuracy drop.
arXiv:2606.07464v1 Announce Type: new
Abstract: Monolithic vision-action models represent an emerging paradigm in autonomous driving. However, this architecture produces token sequences that quickly exceed real-time computational budgets when encoding extended temporal context for complex interactions. While approaches like linear transformers and external memory try to make the context lightweight, token compression is most compatible with the architecture as it requires no backbone modifications. Yet existing compression adopts rule-based heuristics like temporal decay, decoupled from planning, risking loss of decision-critical information. We propose COMPACT-VA, a planning-aligned working memory framework built on conditional VQ-VAE, compressing extended context into bounded representations. Compression is conditioned on both historical trajectory and a learned planning intent that the posterior encoder distills from future trajectories during training, while the prior encoder learns to predict it from compressed observations. The compressed memory, concatenated with the predicted latent, feeds the policy for end-to-end optimization, planning with retained decision-critical information. We evaluate on high-signal dynamic scenarios where historical context is most critical for behavior correctness (e.g., stop, yield, or proceed), and accordingly design behavioral metrics. Under comparable token budgets, we achieve $>$6% improvement (68.3%) on success rates with consistent gains across metrics. Ablations validate planning-aligned coupling effectiveness. Closed-loop evaluation confirms that COMPACT-VA maintained general driving performance with 3.3* speedup and 2.7* memory reduction over uncompressed processing.
arXiv:2606.00603v2 Announce Type: replace
Abstract: LLM agents are increasingly used in moderation-relevant public forum workflows, where their choices to answer, acknowledge, repair, or decline are routinely challenged by users, platforms, and regulators. The same agent often returns different responses on identical content, so any defense based on the agent's behavior cannot be reliably reproduced. The variation is structural. Four deployment choices typically invisible to the operator each shift the agent's response rate, and their combinations can produce substantially different interventions on the same forum posts. The four choices are (1) which model version is currently served, which can change between calls without notice; (2) the model's weight-release status (open-weight, with weights publicly downloadable, vs. closed-weight, with weights held by the provider); (3) which provider serves the request; and (4) which system-prompt policy is in force. Across LLMs spanning both open-weight and closed-weight families, we find that the previously reported tendency to decline more on visible than hidden challenges aligns with the open/closed weight boundary in our panel more than with access surface. Every closed-weight cell declines more on visible challenges; every open-weight cell reverses this or shows no gap. Auditable forum-agent governance requires awareness of all four choices, not just the model name, since each independently shifts behavior.
arXiv:2606.01120v2 Announce Type: replace
Abstract: In RAG-based fact-checking, LLMs are increasingly used as verifiers to check given claims against retrieved evidence. Their parametric knowledge can induce pre-evidence tendencies that may conflict with the retrieved context, yet existing evaluation frameworks do not characterize such prior-context discrepancy or measure how verifiers arbitrate between parametric and contextual signals. We introduce \textsc{PAVE} (\emph{Prior-Aware Verifier Evaluation}), a diagnostic testbed that stratifies an LLM verifier into four epistemic states based on the correctness and confidence of its pre-evidence prior and evaluates its arbitration behavior on this new benchmark, i.e., whether it persists in correct prior under misleading evidence, and whether it corrects wrong prior when accurate evidence is provided. Experiments across seven LLMs reveal unreliable and highly model-dependent prior-context arbitration, highlighting the importance of verifier selection for real-world RAG-based fact-checking applications. Based on these findings, we propose a lightweight JSD-based test-time arbitration method that improves factual reliability without modifying the underlying model, achieving competitive performance across diverse LLM families.
arXiv:2606.03163v3 Announce Type: replace
Abstract: This yellow paper describes the technical architecture of OpenAgenet / OAN. OAN is a protocol-neutral trust layer for open Agent interconnection and discoverable AI resource products. It specifies the role architecture, \texttt{did:oan} identity objects, registration workflow, governance-backed Root lifecycle enforcement, Root-verified package model, authorization-aware Discovery, Root-issued infrastructure authorization VCs, signed trusted invocation, verification requirements, state transitions, security properties, implementation boundaries, and deployment considerations. The design is intended to support heterogeneous Agent frameworks and interaction protocols, including MCP, A2A, ANP-like systems, domain-specific Agent protocols, Skills, MCP Servers, and Tool/API resources. OAN does not define the entire business conversation among Agents or the native protocol of every resource; it defines how resource identities become admissible, discoverable, verifiable, and safe to approach before protocol-specific interaction begins.
arXiv:2606.03382v2 Announce Type: replace
Abstract: While Proximal Policy Optimization (PPO) demonstrates strong performance in stationary settings, we show that its standard optimization paradigm struggles in continual and non-stationary environments. The failure does not stem from insufficient model capacity or overly restrictive clipping. Instead, PPO performs persistent, directionally inefficient local updates, which indicates a lack of geometry-aware guidance for accumulating meaningful behavioral change and ultimately hindering transitions toward new behavior patterns. Although divergence-based regularization introduces partial geometric awareness, its monotonically increasing penalties implicitly discourage large policy deviations, even when such shifts are necessary for effective adaptation. To address this limitation, we propose Gaussian Trust Region Policy Optimization (GTR), which reshapes the trust region using a Gaussian kernel. The resulting constraint is bounded and non-monotonic, providing strong local stability while progressively relaxing under sustained high-advantage updates. To further improve robustness, we introduce a Mixture Gaussian Anchor that adapts to recent policy trajectories, reducing variance induced by stale references. GTR is architecture-agnostic and achieves strong performance across games, simulated robotic control, open-world exploration, and language model post-training. These results demonstrate that geometry-aware trust-region design can be a promising direction for robust reinforcement learning in complex non-stationary environments. Our code is available at https://anonymous.4open.science/r/GTR_demo/README.md.