# Research Papers — 2026-09 · bundle page 228 of 266 > Full-text markdown of papers #11351–11400 of 13298 published in 2026-09, newest first. AI-readable mirrors of the full 13298-paper 2026-09 archive (50 papers per file). Canonical page: https://aipod.dev/papers/archive/2026-09/page/228 [← Previous bundle](https://aipod.dev/papers/archive/2026-09/page/227.md) [Next bundle →](https://aipod.dev/papers/archive/2026-09/page/229.md) ### Beyond Cross-Lingual Transfer: Benchmarking Propagation Boundaries in Multilingual LLM Unlearning Authors: Pengyang Shao, Chuanpeng Lu, Wei Qin, Yanzheng Jin, Xiaohao Liu, Xi Ai, Kenji Kawaguchi, Richang Hong Published: 2026-09-05 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.05976v1 Large Language Model (LLM) unlearning aims to suppress target knowledge while preserving general capabilities. In multilingual settings, unlearning must additionally propagate within its intended linguistic scope. However, existing evaluations mainly measure cross-lingual transfer and cannot distinguish insufficient from excessive propagation. We introduce CLLPU (Cross-Lingual and Language-Bound Protocol for LLM Unlearning), a multilingual benchmark that formulates this problem through two settings: common-goal forgetting, where target knowledge should be suppressed across all languages, and language-conditioned forgetting, where suppression should remain confined to a designated language. CLLPU combines goal-guided topic pairing, schema-aware relation matching, and dual-anchor multilingual translation to construct 800 matched knowledge-unit pairs and 72,000 QA instances across ten languages. Experiments with six representative methods on Llama-3.1-8B-Instruct reveal opposite failure modes: forgetting remains incomplete when universal suppression is required, yet spreads beyond the intended boundary when language-conditioned confinement is required. We further find that general multilingual utility can conceal damage to neighbor knowledge. These findings establish propagation control as a central challenge for multilingual LLM unlearning. We publicly release CLLPU together with its construction pipeline. --- ### Intent Drift at SME Scale: Deployment Practice, Not Model Capability, Determines Agentic Compliance Authors: Ilia Voroshilov Published: 2026-09-05 Categories: cs.CY, cs.AI arXiv: https://arxiv.org/abs/2609.05975v1 We introduce Chain of Intent, a governance framework for agentic AI at small regulated firms, and validate it against a failure it was built to address. Existing agentic governance research assumes enterprise infrastructure that small firms do not have. In a simulated Hong Kong asset manager with 415 synthetic contact records, an agent performing a routine client-communications task was subjected to ordinary managerial pressure to increase its reach. With its authorised constraints written into its configuration, the agent held: it identified every ambiguity in the firm's records, cited privacy legislation it had never been shown, and refused six successive requests, breaching in two of fifteen runs. With the same task, data, pressure and model, but its purpose left unstated as resource-constrained firms routinely leave it, it breached in thirteen of fifteen runs, contacting up to 220 individuals of whom 94 per cent had no demonstrable marketing consent - conduct carrying a maximum of three years' imprisonment under Hong Kong law. Chain of Intent applies four controls requiring no security engineering: a machine-readable purpose, constrained tool access, a scope ledger, and a pre-action check. It eliminated unlawful contact in every run while preserving task completion, and ablation shows each control independently sufficient by a different mechanism. We further show that drift must be measured at two stages - agents widened their candidate sets in every pressured run while acting on them in roughly one in seven - and that governance applied at the point of intent costs roughly half as much as governance applied at the point of action. --- ### Efficient and Robust Camera-independent Multiview 3D Geometric Reconstruction from Noisy Monocular Depth Estimation and Multiple Point Matching Authors: Marius Leordeanu Published: 2026-09-05 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.05972v1 We present an efficient and robust method for 3D geometric reconstruction that is based solely on the camera-independent linear relationships among a given set of points, which are stable over time and robustly estimated using multiple point matches. We essentially learn, from correspondences between points across several frames, a linear geometric auto-regression matrix $\mathbf{W}$, which establishes how a point in 3D can be expressed as a linear combination of all the others. This matrix is constant and does not depend on the world coordinate system or the camera pose---it is an intrinsic property of the point set. We also show that the principal eigenvectors of $\mathbf{W}$, which all have eigenvalue $1$, provide a homogeneous representation of the 3D point configuration. The first version of our method takes advantage of noisy monocular depth maps in order to obtain, from multiple frames, a robust geometric auto-regression matrix $\mathbf{W}$ of linear relationships between the 3D points. Thus, we build on recent advances in deep learning, which now provide monocular depth estimation models that are fast but very often noisy. Our approach handles noise through robust linear estimation over several frames. The second version of our method does not need monocular depth estimation maps. It applies in cases of weak-perspective projection, when the linear combinations between the 3D points can be robustly estimated from their 2D projections in the image. Note that the camera projection matrix is never used in our derivations. Consequently, our method does not recover camera pose, but only 3D structure. This is a key difference between our method and the related literature on 3D geometric reconstruction. --- ### Machine Learning for Pre-Culture ESBL Risk Stratification to Guide Empiric Antibiotic Selection: A 12-Hospital Study of Enterobacteriaceae Cultures Authors: Aravind V. Kuruvikkattil, Lalitha Pranathi Pulavarthy, Rashmita Kudamala, Saptarshi Purkayastha Published: 2026-09-05 Categories: cs.LG, stat.AP arXiv: https://arxiv.org/abs/2609.05970v1 Empiric antibiotic therapy for suspected ESBL-producing Enterobacteriaceae must be selected 48-72 hours before culture results, forcing clinicians to choose between undertreating resistant infections and overusing carbapenems that drive further resistance. We developed a cost-sensitive XGBoost model predicting an ESBL phenotype (resistance to ceftriaxone, ceftazidime, cefepime or piperacillin-tazobactam) at culture ordering using 45 pre-culture EHR features across 132,955 cultures from 72,217 patients at 12 hospitals (14.41% with the ESBL phenotype). Cultures were partitioned at the patient level. At 90% sensitivity, the model achieved 95.8% NPV, reducing post-test ESBL probability to 4.2%, a threshold that may support safe carbapenem-sparing in non-ICU settings, while sparing 307 of every 1,000 cultures an unnecessary broad-spectrum course at the cost of 14 missed ESBL cases per 1,000. SHAP analysis identified prior ESBL colonization as the dominant predictor, ahead of prior organism burden and neighborhood deprivation; removing deprivation features caused minimal performance loss ($Δ\text{AUROC} = -0.020$), enabling equitable bedside deployment. Discrimination was unchanged under a strict IDSA ESBL-E definition (AUROC 0.766), with specimen type added as a predictor (0.764) and without any class-imbalance correction (0.762), and ranged from 0.71 to 0.78 across organism strata. --- ### Test-Time Weak-to-Strong Alignment: Transferring Implicit Rewards from Weak to Strong Flow Models Authors: Xin Xie, Fan Zhang, Dong Gong Published: 2026-09-05 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.05968v1 Aligning a text-to-image generation flow model with a reward makes it follow objectives that the training data alone does not provide. Alignment fine-tuning delivers this by reinforcement learning (RL) or preference optimization, but it must be repeated for every checkpoint and returns a model fixed at the reward and strength it was trained with. Test-time alignment instead steers a frozen model during sampling, allowing task-specific and sample-specific guidance. Existing methods obtain this only by drawing the per-step signal from the reward function itself, through its gradient, or through a separately trained value function. We propose changing the supervision source: let a pair of weak models, not a reward function, supply the supervision. A source aligned model, kept together with its base as a source alignment pair, stores its training reward as an implicit, step-wise, KL-anchored signal expressed in the sampler's own coordinates. We explore whether this model-form supervision can cross scale, and show that it does: our method, AlignGraft, aligns a larger, frozen, never-tuned model by adding the pair's velocity difference during sampling. The transport is exact under a shared noising kernel and needs neither the reward nor its gradient at test time. The method has no schedules, only a single scalar that controls the alignment strength and can extrapolate it beyond that of the source alignment pair. Across image and video flow models (Stable Diffusion 3.5, FLUX, and Wan), the transfer lifts the frozen large model on preference, compositional, and text-rendering rewards, can exceed the source aligned model itself, and preserves the large model's fidelity at a small constant sampling overhead. Extensive experiments show that one alignment run on a weak model produces supervision that the whole model family can reuse at test time. --- ### On-the-go Forgetting without Explicit Unlearning via ERASE Authors: Kushal Chakrabarti, Mayank Baranwal Published: 2026-09-05 Categories: cs.LG, stat.ML arXiv: https://arxiv.org/abs/2609.05966v1 Existing unlearning approaches typically rely on post hoc weight adaptation or distillation, leading to duplicated memory costs, degraded generalization, and limited scalability. In this work, we introduce ERASE, Erasure via Reconstructive Adversarial Signal Editing, a framework for on-the-go forgetting that suppresses the observable influence of private data without modifying model weights. ERASE leverages structured, class-conditioned input perturbations to induce selective forgetting during inference, eliminating the need for retraining, fine-tuning, or model copies. We rigorously characterize sufficient conditions when ERASE provably achieves functional forgetting of designated subclasses while preserving predictions across other subclasses within the same superclass. This analysis offers a principled foundation for inference-time forgetting under mild regularity assumptions. Across diverse architectures and benchmark datasets, ERASE maintains the best observed balance between forgetting efficacy, computational efficiency, and retention fidelity over recent unlearning-based methods. By reimagining data removal as forgetting without unlearning, our work establishes a scalable, regulation-aligned pathway for continual, privacy-conscious learning. --- ### FineHOI: Part-Aware Dense Representations for Zero-Shot Human-Object Interaction Detection Authors: Francesco Tonini, Lorenzo Vaquero, Mohammad Mahdi Derakhshani, Cees Snoek, Elisa Ricci, Cigdem Beyan Published: 2026-09-05 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.05959v1 Human-Object Interaction (HOI) detection aims to localize humans and objects in images and classify their interactions. Zero-shot HOI focuses on recognizing interactions that are not observed during training, requiring models to generalize beyond seen verb-object compositions. Recent approaches leverage Vision-Language Models (VLMs), benefiting from rich semantic representations. However, they often rely on global or detector-centric features that compress interaction cues and hinder fine-grained spatial reasoning. To overcome this limitation, we propose FineHOI, a zero-shot HOI framework that explicitly models interactions from dense patch-level features. Our approach is motivated by the observation that human-object interactions are defined by localized spatial relationships, which are not preserved by global and detector-centric representations. To this end, we introduce an Adaptive Part-Level Attention module that decomposes humans and objects into semantically coherent parts via unsupervised clustering, and re-weights them based on their interaction relevance. These representations are then integrated through a Region-Aware Interaction Transformer that integrates part-aware and global features and produces the final HOI embedding. Extensive experiments demonstrate that FineHOI consistently outperforms existing zero-shot HOI methods, achieving particularly strong gains on unseen interactions. Code is available at https://github.com/francescotonini/fine-hoi. --- ### STP-BENCH: A Unified Systematic Benchmark for Virtual Spatial Transcriptomics from Histopathology Images Authors: Youngmin Chung, Ji Hun Ha, Andrew H. Song, Cristina Almagro-Pérez, Chaeyoung Seo, Won Jun Suh, Jeong Won Beom, Kyoung Bin Oh, Eytan Ruppin, Faisal Mahmood, Joo Sang Lee Published: 2026-09-05 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.05956v1 Spatial transcriptomics (ST) provides unprecedented insights into tumor heterogeneity by capturing spatially resolved gene expression, yet its high experimental cost hinders large-scale adoption. Consequently, computational approaches that predict spatial gene expression directly from hematoxylin and eosin slides, termed virtual ST, have rapidly emerged. Despite this progress, assessing advances in the field remains difficult due to insufficient benchmarking: prior studies rely on small, heterogeneous datasets, inconsistent training and inference pipelines, and limited evaluation of biological interpretability and model robustness. To address these gaps, we present STP-BENCH, a standardized benchmark for virtual ST models. STP-BENCH comprises six cancer types spanning two ST platforms (Visium and Xenium), with each training dataset containing more than 30,000 spots and at least 15 slides to ensure statistical reliability. We evaluate 21 predictive approaches, re-implemented with a unified pathology foundation model as the morphological encoder when architecturally applicable. Beyond conventional benchmarks that report average predictive accuracy on highly variable genes, we systematically examine which genes and gene sets are recoverable from histomorphology. We further evaluate the downstream biological utility of predicted profiles through cell-type deconvolution and spatial domain identification, and assess model reliability under domain shifts and data scaling. Notably, unified morphological encoding substantially re-orders model rankings established in prior studies, indicating that architectural innovations and image encoding have been conflated in previous evaluations. We publicly release STP-BENCH to support reproducibility and serve as a community benchmark at https://github.com/NEXGEM/STP-Bench. --- ### LoGIC: Budgeted Context Construction for Node-Level Graph In-Context Learning with Tabular Foundation Models Authors: Mingqi Yang, Zidong Guo, Jihui Yang, Wenming Zuo Published: 2026-09-05 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.05955v1 Tabular foundation models have become powerful graph learners. Systems such as G2T-FM and GraphPFN encode each node as a feature row and make predictions through in-context learning (ICL), with labeled rows serving as the prompt. Current protocols employ the complete training table as context, causing attention to scale quadratically with the labeled pool and introducing preprocessing and memory bottlenecks. We investigate context construction for node-level graph ICL: which labeled nodes and auxiliary unlabeled nodes should constitute the prompt for specified queries. We formulate this allocation in terms of two resources: a labeled-context budget for predictive evidence and an unlabeled-halo budget for adapter message passing without using label capacity. We present LoGIC, which retrieves labeled nodes via structural, feature-based, and coverage channels, shares each context across the queries in a graph-local cluster, incorporates an unlabeled halo for adapter backbones, and chooses the channel and context budget without test labels. Across three backbone configurations drawn from two model families on GraphLand, budgeted contexts maintain locally runnable full-context performance, stay competitive with published large-dataset results, and markedly lower peak memory requirements compared with full-context and whole-graph inference. They further permit frozen graph ICL on million-node graphs without retraining. Our analysis identifies when retrieval channels work best and connects their behavior with graph properties. --- ### ProtoRAG: Prototype-Based Retrieval Augmentation for Few-Shot Fine-Grained Remote Sensing Object Detection Authors: Jian Wang, Yuxiang Hong, Chufeng Zhou, Chao Pang, Xiaokang Zhang Published: 2026-09-05 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.05953v1 Few-shot fine-grained object detection (FGOD) in remote sensing imagery is challenging because limited annotations must support both object localization and discrimination among visually similar subcategories. Although multimodal large language models (MLLMs) provide strong coarse object localization, they lack explicit visual evidence for reliable fine-grained recognition. To address this limitation, we propose ProtoRAG, a prototype-based retrieval-augmented framework that decouples coarse localization from fine-grained recognition by equipping MLLMs with an external object-level visual memory. To construct a reliable visual memory from limited support samples, we introduce Discriminative Prototype Space Learning (DPSL), which encourages discriminative and prototype-stable representations through supervised contrastive learning and prototype-consistency regularization. We further develop an uncertainty-guided candidate-constrained reasoning strategy that augments MLLMs with retrieved candidate-specific visual references and invokes multimodal reasoning only for ambiguous instances. Extensive experiments show that ProtoRAG consistently surpasses representative baselines in nine few-shot settings, outperforming the strongest baselines by 14.80, 2.27, and 4.04 mAP$_{50}$ on MAR20, HRSC2016, and FAIR1M-2.0, respectively. --- ### QGB-W$k$NN: Quantum Granular-Ball Learning for Robust Classification Authors: Suzhen Yuan, Dehang Chen, Lifeng Shen, Shuyin Xia, Jeremiah D. Deng Published: 2026-09-05 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.05952v1 Nearest-neighbor classification is widely used in machine learning, yet existing methods often suffer from low computational efficiency and limited robustness in noisy environments. To jointly address these challenges, this paper proposes an efficient and reliable weighted $K$-nearest neighbor classification framework based on quantum granular balls, termed QGB-W$k$NN. The proposed framework improves computational efficiency by integrating quantum-enhanced granular-ball representation with hierarchical nearest-neighbor search, while enhancing classification reliability through a purity-aware weighted decision mechanism. Specifically, quantum-kernel granular balls are constructed to reduce retrieval redundancy and strengthen nonlinear feature representation under limited quantum resources. A granular-ball purity-guided HNSW optimization strategy is developed to exploit structural reliability for hierarchical graph construction during neighbor retrieval, alleviating the local optimality issue caused by conventional random layering. Finally, a weighted voting mechanism jointly incorporating granular-ball similarity and purity is introduced to produce more reliable classification decisions in noisy environments. Extensive experiments on benchmark datasets demonstrate that QGB-W$k$NN achieves competitive classification accuracy while exhibiting favorable Pareto trade-offs between classification performance and computational cost. Moreover, the proposed framework consistently improves robustness under various noisy conditions, suggesting that reliability-aware quantum granular-ball learning provides a promising paradigm for efficient and robust nearest-neighbor classification. --- ### The Blindness of Document-Level Translation Evaluation Authors: Ahrii Kim, Vilém Zouhar, Chanjun Park, Seong-heum Kim Published: 2026-09-05 Categories: cs.CL, cs.HC arXiv: https://arxiv.org/abs/2609.05949v1 Document-level machine translation (MT) evaluation extends segment-level protocols by presenting full documents to annotators, on the assumption that such presentation elicits document-level judgments. We test this assumption with a counterfactual condition (MIX) in which each document combines segments drawn from different systems, preserving document-level presentation while breaking cross-segment consistency. Across 18,420 expert Englis-to-Korean annotations and 14 automatic metrics, scores, system rankings, and error annotations are statistically equivalent between coherent and incoherent documents. Perception does not explain this: shown matched passages, raters identify the coherent one as the work of a single translator in 87.3% of trials. Document presentation does change how annotators work, but that change does not reach the recorded output. What is blind is the protocol, not the annotator. The concern is not that scores fall short, but that the resources invested in document-level systems, metrics, and annotation may not be measuring what they are intended to measure. --- ### Beyond Final Decisions: A Process-Centric Benchmark for Transparent AI-Assisted Peer Review Authors: Siming Yuan, Xueyi Zhang, Wangze Ni, Tianfang Xiao, Shimin Di, Jia Zhu, Zhuoren Jiang, Rong Tan, Lei Chen, Kui Ren Published: 2026-09-05 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.05947v1 Peer review is central to quality control in science. However, existing evaluations of AI-assisted peer review mainly focus on the overall quality of generated reviews or the accuracy of final decisions. They therefore provide limited evidence about whether model decisions are supported by sufficient and reliable review evidence. We introduce a process-centric diagnostic benchmark for AI-assisted peer review. It uses (x,$z_s$,$z_c$,$z_r$,y) to represent the paper content, summary, critique, suggestion, and decision. We convert heterogeneous review records from PeerRead, NLPeer ARR-22, and OpenReview-ICLR into process-aligned data. Our benchmark uses direct decision prediction from the paper content (Direct) as its baseline. It compares the decision value of Gold-process variables and Predicted-process variables, and conducts stage-level evaluation, chain-consistency evaluation, and interventional sensitivity analysis. Experiments across three datasets and six models show that Gold-process variables generally have higher decision value. For the main analysis model, the Gold--Predicted gap remains stable across datasets and random seeds. This gap is also reproduced in most model--dataset combinations. Although model-generated intermediate review texts show relatively high local consistency across adjacent stages, the final decisions are not consistently supported by the preceding review evidence. Our benchmark targets AI systems designed to assist rather than replace human reviewers. It provides a transparent and auditable diagnostic tool for evaluating the reliability of their review processes. --- ### Interpretable and Fair Generalized Additive Neural Networks via Multi-objective Learning Authors: Ziming Wang, Changwu Huang, Ke Tang, Yew-Soon Ong, Xin Yao Published: 2026-09-05 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.05946v1 Interpretability and fairness are two of the most emphasized dimensions in trustworthy artificial intelligence (AI). Various explainable AI methods have been introduced to improve interpretability. This paper focuses on neural network (NN)-based generalized additive models (GAMs), a class of self-interpretable models. While most existing research has prioritized improving the accuracy of NN-based GAMs, their interpretability remains largely underexplored. To address this gap, this paper introduces explicit quantitative metrics for evaluating the interpretability of NN-based GAMs, empirically examines their effectiveness, and explores strategies for improving interpretability within these models. In addition, the simultaneous and explicit optimization of both interpretability and fairness, along with their trade-offs and the underlying reasons, remains underexplored. To address this, we propose a multi-objective neural basis model (MONBM) framework based on multi-objective evolutionary learning to consider accuracy, interpretability, and fairness simultaneously. A partial retraining strategy is further developed to facilitate the practical application of evolutionary multi-objective optimization to deep model architectures. Based on MONBM, this paper reveals the complex relationships between these dimensions and the reasons behind these intricate relationships. This analysis demonstrates how multi-objective optimization can be combined with self-interpretable models to reveal relationships among trustworthiness objectives. In addition, MONBM obtains a set of models with different trade-offs between dimensions, and the competitiveness of the approach is validated by comparing it with state-of-the-art methods. --- ### SurveyAgent-HKA: A multi-agent framework for scientific survey generation with LLMs and human knowledge augmentation Authors: Tong Bao, Mir Tafseer Nayeem, Yi Zhao, Davood Rafiei, Chengzhi Zhang Published: 2026-09-05 Categories: cs.CL, cs.DL, cs.IR arXiv: https://arxiv.org/abs/2609.05938v1 Automatic scientific survey generation has become an important task in scientific document processing. The common approach of retrieving literature from a single source (e.g., arXiv) and generating surveys through a one-pass large language model (LLM) call often leads to limited reference coverage and, more importantly, fails to replicate the expert-driven revision process that is crucial for writing high-quality surveys. In this paper, we introduce SurveyAgent-HKA, a multi-agent framework that improves end-to-end scientific survey generation by incorporating knowledge derived from published surveys and peer-review comments. The framework decomposes survey generation into well-defined sub-tasks handled by LLM-powered agent. It first retrieves relevant papers from multiple sources and identifies key topics through clustering to construct an initial outline, which is then refined using outlines from related human-written surveys. Based on the refined outline, topic-focused papers are retrieved and re-ranked to select for drafting a well-grounded survey. Then, we identify common issues raised by experts in peer-review comments from published surveys to guide the revisions and finalize the survey. Experiments on two domains show that our approach outperforms mainstream baselines in citation quality, structural consistency, and content quality. Furthermore, our framework is efficient in both time and cost, making it a practical solution for broader AI-assisted scientific writing applications. --- ### Beyond Classification: Structured Supervision Aligns Visual Evidence with Medical Semantics Authors: Hexiang Bai, Hanyang Xu, Xiaoxue Li, Xiaoliang Wu, Shangde Gao, Hongxia Xu, Ke Liu Published: 2026-09-05 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.05937v1 Vision Transformers (ViTs) have shown immense potential in medical image analysis. However, standard pre-training via global image classification suffers from spatial collapse, where models rely heavily on background shortcuts rather than localising critical foreground lesions. To overcome this limitation and align visual evidence with precise medical semantics, we systematically investigate alternative pre-training paradigms.Specifically, we evaluate three independent forms of structured supervision: topological priors via graph self-supervision, dense pixel-level constraints via segmentation, and cross-modal semantic grounding via image-text pairs. Notably, our empirical analysis reveals that while all three forms of structured supervision successfully alleviate the global pooling bottleneck and steer visual attention towards foreground regions, image-text alignment achieves the most superior performance. By embedding high-dimensional diagnostic logic, the cross-modal approach not only anchors attention on precise visual evidence but also enables profound abstract reasoning. Extensive experiments demonstrate that this semantically enriched pre-training fundamentally enhances the model's feature representation. Consequently, when fine-tuned for downstream clinical classification tasks, our models achieve superior accuracy and yield highly interpretable attention maps focused on true pathological features, vastly outperforming vanilla classification baselines. --- ### Rethinking the Evaluation of Efficiency Methods for Multi-Agent Systems Authors: Jiamu Zhang, Lingxi Zhang, Pengjun Lu, Qiyue Zhang, Yu-Neng Chuang, Zhengchen Li, Shuai Xu, Vipin Chaudhary, Hanjie Chen Published: 2026-09-05 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.05933v1 Efficiency is increasingly important for Large Language Model (LLM)-based multi-agent systems (MAS), as larger models and more agents introduce substantial execution costs. Recent methods aim to make MAS cheaper by pruning agents, removing communication edges, or searching for compact structures. However, we argue that existing evaluations may overestimate their true ability to improve MAS efficiency. Reported gains are often measured under method-specific prompts and starting topologies, making them difficult to attribute to the proposed structural changes. Moreover, many reported successes appear in non-MAS-demanding settings, where a single agent or a randomly pruned system can already preserve strong performance. To study these issues, we introduce a controlled and MAS-demanding diagnostic benchmark for representative MAS efficiency methods. We evaluate methods under a shared backbone model, agent registry, and runtime, across controlled variations in topology, scale, depth, and tool use. Our analysis shows that many reported gains are setup-dependent and may arise from structural collapse, disabled tool pathways, or starting systems where random pruning already preserves accuracy, rather than robust improvements in MAS efficiency. --- ### Solving versus Verifying: Catching Contradictions in Tax Reasoning Systems Authors: Albert Sadowski, Jarosław A. Chudziak Published: 2026-09-05 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.05928v1 Large language models now compute correct tax liabilities on over 90% of well-formed cases in statutory benchmarks, which makes them candidates for the tax-advisory and compliance systems that consume such an answer directly. Real legal inputs, however, are frequently defective: required facts are missing, or stated facts contradict one another. Accuracy on clean benchmarks says nothing about how a model behaves then, and a system that computes straight through a defective input returns a confident number with no sign that anything is wrong. This raises two questions: does a model asked to solve a case abstain when the input is defective, and when it does not, can the same model catch the defect when asked instead to verify the input? We study six recent models on SARA-derived tax cases under missing-fact and contradictory-fact perturbations. The strongest models abstain when a fact is missing but compute through injected contradictions, returning the clean-input answer 63-76% of the time with no signal of the conflict; asked instead to verify the same input, they flag most of those contradictions. We wire that verification call into a simple contradiction gate: one extra call that abstains when the model reports a conflict. Across all six models it recovers most of the missed contradiction abstention at a clean-accuracy cost of at most about 5 percentage points, with no training and no external tooling. High accuracy on well-formed inputs is therefore an incomplete measure of reliability, and the detection the solver misses is cheaply recoverable with a single self-check. --- ### AVSplat: Dense-View Feed-Forward 3D Gaussian Splatting with Assist-View Preconditioning Authors: Muyu Xu, Fangneng Zhan, Yu Wei, Hanspeter Pfister, Shijian Lu Published: 2026-09-05 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.05925v1 Pose-free feed-forward 3D Gaussian Splatting enables novel view synthesis from uncalibrated multi-view images. Although more views should improve performance, existing methods often degrade with dense-view inputs because global aggregation spreads attention over many tokens, and naive voxel fusion averages many Gaussians into overly smooth representations. We present AVSplat, a framework that turns additional views into reliable signals for both aggregation and representation. Before global attention, each view performs a single lightweight interaction with a small set of Assist Views chosen for relevance and diversity, and the cached features provide a focused scene context that stabilizes correspondence. For representation, we use adaptive temperature-aware voxel fusion that sharpens attribution under high occupancy, guided by occupancy and point confidence. Crucially, AVSplat restores positive view scaling where performance remains stable or improves as more input views are added, instead of degrading in the dense-view regime. Ablations show that Assist View Preconditioning is primarily responsible for preventing dense-view degradation, while Occupancy-guided Voxel Fusion contributes most of the single-point image-quality gains. --- ### Versioned Transitive Dependency-Closure Binding and Operation-Time Effect Governance for Agent Skills: ClosureBound Authors: Genliang Zhu, Chu Wang Published: 2026-09-05 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2609.05920v1 Agent Skills combine instructions with files, packages, tools, models, and services, so operational identity can exceed a signed directory. Recursive or lazy dependencies may change while root-level evidence remains valid, and different surfaces may reach the same durable effect. We present ClosureBound, a reference monitor that prevents authorization transfer across material changes to this heterogeneous closure. Its resolver commits typed graph nodes and topology. Each grant binds an exact closure root, effect ceiling, purpose/provenance, validity, and epochs. At durability, it re-resolves closure and state, normalizes the operation into an external-effect IR, and admits it only if a joint witness satisfies every bound. Supported equivalent paths share one ceiling. Assuming complete mediation and discovery, authenticated freshness, sound normalization, cryptographic binding, and authoritative linearization, we establish metadata non-authority, closure determinism, version non-inheritance, effect non-amplification, bound-value freshness, and path invariance. We do not establish program equivalence or remote-service honesty. A provider-free implementation matches 40 frozen lifecycle fixtures; 18 kernel contracts and six mutants cover binding and downgrade cases. Full-profile exploration reaches 84,608 states and 530,752 transitions without a declared invariant violation; six weakened profiles yield witnesses. A lexical audit of 549 public Skills (4,872 unique files) finds that 21 of 526 roots with bundled files name every non-manifest path verbatim, 67 contain links resolving outside their roots, and no root declares a frontmatter dependencies field. These observations motivate conservative closure discovery and define concrete targets for broader runtime, interoperability, efficacy, and production validation. --- ### A dictionary learning framework for graphs via filters and optimal transport Authors: Jinchuan Liao, Dai Hai Nguyen Published: 2026-09-05 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.05919v1 We propose a graph dictionary learning (GDL) framework where each graph is represented as a zero-mean Gaussian distribution derived from its filtered Laplacian. Each observed graph is approximated by a barycenter over learned atom graphs, computed under the filter graph distance (fGOT), a graph comparison metric sensitive to global structural properties. The reconstruction error between the observed graph and its barycenter is measured by the surrogate fGOT (sfGOT) distance, a tractable approximation of fGOT that handles graphs without known node correspondence, and is minimized end-to-end via backpropagation. We further provide a novel interpretation of sfGOT through the lens of the Hilbert-Schmidt Independence Criterion, showing that minimizing the sfGOT distance between two graphs is equivalent to maximizing statistical dependence between the spectral embedding of their nodes. Experiments on benchmark datasets demonstrate competitive performance over existing GDL methods on graph clustering and classification tasks. --- ### PAI-Actor: Cinematic Multi-Character Replacement in Dynamic Scenes Authors: Bangxun Tang, Heyuan Gao, Yiren Song, Guian Fang, Zijian He, Jie Yang, Mike Zheng Shou Published: 2026-09-05 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.05918v1 We present PAI-Actor, a cinematic multi-character animation framework for character replacement in dynamic movie scenes. Unlike conventional animation systems that mainly drive a single static image or a single subject, our goal is to replace and animate multiple characters within real video clips while preserving the original scene dynamics, camera motion, and background content. This setting is particularly challenging because the generated characters must remain consistent with the source performance in motion and interaction, while also matching the surrounding background in lighting, shadow, composition, and overall cinematic appearance. To address this, we formulate multi-character animation as a structure-guided human recovery problem and build a movie-driven training pipeline from high-quality film data. Furthermore, to support practical cinematic production, we introduce a bidirectional-to-autoregressive distillation framework: we first train a bidirectional diffusion transformer for high-quality short-clip generation at 1080P resolution, and then distill it into an autoregressive video-to-video model for efficient inference and longer video generation. Experiments show that PAI-Actor enables high-fidelity multi-character animation with strong scene consistency, cinematic visual quality, and efficient long-form generation. --- ### STAR-Pro: Stage-Wise Token Adaptive Reduction with Progressive Refinement for Efficient Large Vision-Language Models Authors: Yichen Guo, Tinghao Wang, Qizhe Zhang, Lingbei Meng, Yuan Zhang, Jiajun Cao, Hao Jiang, Chenwei Wu, Jixian Wu, Sixiang Chen, Tao Luo, Hongyang Cheng et al. Published: 2026-09-05 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.05916v1 Large vision-language models (LVLMs) achieve strong multimodal understanding, but the hundreds to thousands of visual tokens they process impose substantial computational overhead, motivating training-free visual token pruning. In this work, we conduct two complementary analyses of visual token pruning. First, we measure the feature-space coverage of tokens retained before cross-modal fusion and find that aggressive pruning discards substantial visual information. Second, we track text-to-visual attention across decoder layers and find that the visual tokens considered important change substantially with depth, making one-shot pruning decisions unreliable. Together, these findings show that effective pruning should preserve broad visual coverage before fusion and progressively refine the retained tokens as cross-modal evidence evolves during fusion. We therefore propose STAR-Pro (STage-Wise Adaptive Token Reduction with Progressive Refinement), a training-free two-stage framework. Its Adaptive Stage applies pivoted QR to construct an over-budget feature-coverage candidate pool, while its Progressive Stage uses evolving text-to-visual attention at selected decoder layers to prune a nested survivor set under a target layer-average token budget. Extensive experiments across seven LVLMs spanning multiple architectures and 18 image and video benchmarks demonstrate the effectiveness of STAR-Pro under aggressive pruning. On LLaVA-Video-7B, STAR-Pro reduces visual tokens by 90.5%, retains 92.7% of baseline performance, and achieves a $2.24\times$ measured inference speedup. Code is available at https://github.com/EasonAI-5589/starpro. --- ### Neuron-Guided Fine-Tuning: Unlocking Efficient Alignment Mechanisms for Large Language Models Authors: Zeyu Wu, Junchao Wu, Shudong Liu, Runzhe Zhan, Xin Chen, Shu Yang, Yichao Du, Longyue Wang, Weihua Luo, Jinsong Su, Derek F. Wong Published: 2026-09-05 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.05913v1 Existing Supervised Fine-Tuning paradigms, particularly Full Parameter Fine-Tuning are often plagued by parameter redundancy, inconsistent data quality, and catastrophic forgetting, which current methods typically address in isolation and lack a unified optimization signal to bridge data selection, parameter updates, and knowledge preservation. To address this, we propose Neuron-Guided Fine-Tuning (NGFT), a holistic framework that leverages neuron activation patterns as a universal proxy to unify the fine-tuning lifecycle. NGFT operates via three synergistic mechanisms: (1) Adaptive Task-Specific Neuron Selection, which identifies essential neurons in a single forward pass to concentrate updates and reduce redundancy; (2) Activation-Based Data Selection, which prioritizes information-dense samples that maximize contribution to key neurons; and (3) Neuron Activation Alignment, a novel loss function that anchors activations to pre-trained states, deepening representation learning and preserving general knowledge. Experimental results across three models across both domain-specific and general benchmarks demonstrate that NGFT significantly outperforms existing mainstream fine-tuning methods in both efficiency and performance, while effectively mitigating catastrophic forgetting. --- ### FreeTransformSR: Efficient Lightweight Image Super-Resolution via Free Low-Rank Learnable Transform Authors: Hongji Li, Yunhui Li Published: 2026-09-05 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.05912v3 Single image super-resolution aims to reconstruct high-resolution images from low-resolution inputs. This paper proposes FreeTransformSR, a novel lightweight super-resolution network based on a channel-wise free low-rank learnable transform. The transform learns task-adaptive basis functions in a data-driven manner, enabling adaptive feature modulation with minimal parameter overhead. To further enhance high-frequency detail recovery, we introduce a local feature modulation branch that complements transform-domain processing with depthwise convolution. In addition, a soft complexity adaptive module dynamically fuses the outputs of local convolution and window self-attention branches through a lightweight gating network, adaptively adjusting the fusion ratio based on regional texture characteristics. An adaptive intensity modulation strategy is also incorporated to adjust transform-domain response strength at the sample level, enabling the network to dynamically adjust processing intensity according to input features. Extensive experiments on five benchmark datasets demonstrate that FreeTransformSR achieves competitive PSNR/SSIM performance with significantly fewer parameters and FLOPs. Specifically, FreeTransformSR achieves 32.41 dB on BSD100 x2 and 27.00 dB on Urban100 x4 with only 595K parameters, while delivering faster inference speed than competing methods, making it well-suited for deployment in resource-constrained scenarios. Source code is available at: https://github.com/HJiLi/FreeTransformSR. --- ### Structurally Close, Temporally Distant: Measuring Security Exposure in Long-Horizon LLM Agents Authors: Md Jafrin Hossain, Nur Al Hasan Haldar Published: 2026-09-05 Categories: cs.CR, cs.AI, cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.05911v1 Long-horizon LLM agents interact with untrusted content, persistent memory, external state, and sensitive tools. Existing analyses often characterize attacks by the number of execution steps between malicious input and a downstream action. We show that temporal remoteness can overstate security separation in stateful agents. We introduce a provenance-aware execution graph linking agent events through deterministic state, identifier, and tool provenance, and define \emph{influence distance} $\DI$ as the shortest structural path from an untrusted source to a sensitive action. We compare it with \emph{sequence distance} $\DT$, the shortest injection--sink path in the ordered trajectory. Since the influence graph contains every sequence edge, $\DI \leq \DT$; $\Gap=\DT-\DI$ measures the separation hidden by step count. Across 454 injection--sink pairs from 360 long-horizon AgentDojo trajectories over OpenAI's \texttt{gpt-4o-mini} and \texttt{gpt-4o} and Claude's Haiku 4.5 and Sonnet 4.6, $\Gap>0$ for 96.9% of pairs, with a median gap of 9 hops; 91.0% remain decoupled after removing the largest provenance-only edge class. On AgentDojo's banking suite, 33.8% of 231 pairs from 377 trajectories decouple through different provenance mechanisms. Among 274 OpenAI pairs, $\Gap$ does not independently predict attack success after controlling for $\DT$, attack family, and backend ($β_{\Gap}=0.066$, $p=.088$). At matched thresholds $k=2,3$, a deterministic $\DI$-based pre-execution gate blocks five attack sinks missed by a sequence-only gate with no additional benign blocking, although the paired gain is not significant ($p=.0625$). Execution structure therefore reveals proximity hidden by step count and can support targeted runtime intervention. We measure candidate influence pathways rather than causal attribution. --- ### UniRRM: Unified Reasoning Reward Models Across Languages and Evaluation Paradigms Authors: Peng Lai, Yichao Du, Junchao Wu, Weibo Gao, Linan Yue, Longyue Wang, Weihua Luo, Derek F. Wong, Guanhua Chen Published: 2026-09-05 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.05910v1 Reinforcement learning (RL) excels on tasks with verifiable rewards, but in open-ended tasks, the reliability of reward models remains a key challenge. Existing solutions either depend on costly proprietary LLM-as-a-Judge systems or opaque scalar reward models that lack interpretability. Recent works on generative reward models offer a promising alternative, but they remain constrained by static evaluation criteria, fragmented evaluation paradigms, and limited multilingual support. To address these challenges, we introduce \textbf{MixReward}, a large-scale multilingual dataset spanning six domains and 103 languages, containing both pairwise and listwise data, and propose \textbf{UniRRM}, a unified reasoning reward model supporting multiple languages and evaluation paradigms. UniRRM uses a staged reasoning chain to dynamically generate task-generic and instruction-specific criteria, enabling fine-grained, input-adaptive judgments while maintaining consistency across languages. Experiments demonstrate that UniRRM-8B and UniRRM-14B achieve performance close to the state-of-the-art for models of comparable size across multiple benchmarks, and are effective for unseen evaluation paradigms. In addition, ablation studies validate the reliability and effectiveness of UniRRM. --- ### From Narrative to Auditable Forecasts: A Structured Scaffold for Agentic Forecasting Authors: Yuanpu Cao, Yongkang Du, Yurui Chang, Lu Lin, Jinghui Chen Published: 2026-09-05 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.05905v1 LLM agents are increasingly used for live forecasting, where they retrieve up-to-date information and produce estimates for unresolved future events. However, current agentic forecasting often relies on implicit narrative aggregation: agents collect evidence, discuss it in prose, and often assign a probability without an explicit update path from evidence to forecast. This limits both forecasting accuracy and auditability. We propose AuditForecast, an agentic scaffold for structured probabilistic forecasting. AuditForecast first anchors the forecast with a suitable quantitative baseline model, uses model-guided data retrieval to derive a base probability, and then applies situational factor updates outside the model's scope through mechanical aggregation in odds space. This turns forecasting from a prose-based judgment into a structured process with explicit intermediate objects. Across multiple live forecasting benchmarks, AuditForecast improves forecasting accuracy and calibration relative to strong agentic baselines, surpasses market-implied references in several settings, and outperforms substantially more expensive deep-research agents while remaining Pareto-dominant in the cost--accuracy tradeoff. Beyond performance gains, AuditForecast produces an auditable forecasting report that makes forecast construction explicit and supports systematic post hoc analysis. --- ### GenPuzzle: Benchmarking Visual Reasoning in Image Generation Models Authors: Changpeng Zhao, Yiren Song, Jinpeng Wang Published: 2026-09-05 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.05902v1 Recent image generation systems increasingly combine multimodal understanding, reasoning, and synthesis, suggesting that they may do more than render plausible scenes. Yet existing evaluations emphasize aesthetics, prompt alignment, compositionality, or text-based answers, leaving unclear whether these systems can solve visual problems and faithfully express solutions in pixels. We introduce GenPuzzle, a benchmark for reasoning-centric image generation. GenPuzzle contains 2,005 problems across 12 tracks, spanning pattern completion, spatial construction, mazes, Sudoku, nonograms, tangrams, board games, matchstick puzzles, orthographic projection, and mathematical visual proof. Each task provides a visual puzzle and requires an image output that preserves the input state while executing a logically valid solution. GenPuzzle uses task-specific evaluation protocols: discrete grid outputs are transcribed and verified programmatically, while visually complex outputs are assessed with tiered, multidimensional, or binary multimodal large language model (MLLM) rubrics. We further select the automatic judge by measuring agreement with human reference scores. Across three frontier generators, the strongest model reaches only 40.57 Macro Overall, revealing frequent failures in logic, geometry, state preservation, and instruction execution. GenPuzzle provides a testbed for measuring progress from image rendering toward visual problem solving. --- ### From Review to Authorization: Key-Isolated Threshold Signing for LLM Agents Authors: Yu Zheng, Qizhi Zhang Published: 2026-09-05 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2609.05901v1 Autonomous LLM agents can turn untrusted content into effectful actions such as payments and permission changes. If the same process interprets this content and controls a reusable signing credential, prompt injection can cross the judgment boundary and reach execution authority. We present KITA, a review-to-authorization architecture that keeps the user's personal secret signing key and every threshold signing-key share outside all LLM processes. Under threshold signature unforgeability and our system assumptions, compromising the proposer and fewer than t reviewer-signer domains cannot produce a valid authorization for a new action without signing contributions from t distinct domains. Thus, any such authorization includes a share from an uncompromised domain, bound to the canonical action and released only after authenticated reviewer approval. This establishes execution-bound authorization integrity. We implement the complete reviewer-to-executor path with a structured-output LLM adapter and threshold BLS. Six system tests validate quorum gating and message binding at this interface, while cryptographic microbenchmarks measure the online signing path and its scaling behavior. --- ### AlignDiff: Exploiting Model-Intrinsic Information for Better Preference Data Selection Authors: Peng Lai, He Zhu, Zhiwen Ruan, Dongdong Zhang, Yun Chen, Peng Li, Furu Wei, Yang Liu, Guanhua Chen Published: 2026-09-05 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.05899v1 Aligning large language models with human preferences remains a challenge, primarily due to the critical role of preference data quality in effective alignment. Existing datasets are frequently plagued by inherent noise and distribution shifts, which inherently limit model performance. To bridge this gap, we propose AlignDiff, a preference data filtering framework driven by intrinsic model signals. AlignDiff first identifies samples with clear preferences using both positive and inverse signals, then prioritizes the more challenging samples based on the average negative log-likelihood gap, encouraging the model to learn richer information from them. AlignDiff is evaluated on two widely used model families (LLaMA and Qwen) and three benchmarks widely adopted in the alignment community (AlpacaEval 2.0, Arena-Hard, and MT-Bench). Across all settings, it consistently outperforms seven strong baselines. We conduct comprehensive ablation studies to validate the effectiveness of AlignDiff, and further show that difficulty-based curriculum learning improves model performance. --- ### Polarity-Asymmetric Structural Calibration for Link Sign Prediction Authors: Qiqi Gao, Wenzhuo Song, Xueyan Liu Published: 2026-09-05 Categories: cs.SI, cs.LG arXiv: https://arxiv.org/abs/2609.05896v1 Link sign prediction (LSP) aims to infer the positive or negative polarity of unobserved links in signed networks. Signed Graph Neural Networks (SGNNs) usually rely on signed-graph structural priors, including structural balance and homophily-like similarity, to guide message passing and prediction. These priors describe population-level tendencies, not guarantees for individual target edges. Their failures are especially costly under severe sign imbalance, where errors on minority and locally conflicting relations are harder to detect and correct. We propose Polarity-Asymmetric Structural Calibration (PASC), a target-edge structural-prior calibration framework for signed link prediction. PASC constructs a structure-only prior representation, estimates a target-edge structural prior score, and compares this score with a local signed-context cue to derive a conflict residual. The residual calibrates signed attention aggregation, target-edge gated fusion, and regime-adaptive optimization. Experiments on five real-world signed network datasets show that PASC consistently achieves the best Macro-F1 among representative baselines, with competitive AUC, Binary-F1, and Micro-F1. Structural-shift experiments further suggest reduced dependence on dense-neighborhood and local-closure shortcuts. Source code is available at https://github.com/iqqGGGGGGG/PASC-for-LSP. --- ### A First-Order Learning Algorithm for Online Resource Allocation with Constant Regret Authors: Menglong Li, Jiawei Zhang Published: 2026-09-05 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.05895v1 We study a finite-horizon online resource allocation problem with initial resource capacities proportional to the horizon. In each period, a request type is observed and one action is chosen from a finite menu. Each action earns a reward and consumes a vector of resources. The arrival types are independent and identically distributed, but their probabilities are unknown. We present a primal first-order learning policy that, in each period, performs one gradient ascent update of the action coordinates associated with the current request type. The policy achieves $O(1)$ expected additive regret relative to the hindsight optimum, with a bound independent of the horizon $T$. It does not solve any linear program, and the regret bound does not require a nondegeneracy assumption on the fluid linear program. --- ### The End of AI Exponentiation: Fluttering Inside and Outside AI Bubble Authors: Victor Kebande Published: 2026-09-05 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.05894v1 The exponentiation of Artificial intelligence (AI) in the recent past has entered a transformative era that has been driven by the growth in large language models (LLMs), large-scale compute infrastructures, and autonomous reasoning systems. However, the rapid acceleration of AI has increasingly shown technological, societal, economic, ethical and infrastructural challenges associated with peak data limitations, rising computational demands, synthetic data recursion, valuation inflation, and societal instability. The traditional scaling paradigms that have powered the modern AI systems are gradually encountering friction in sustaining continuous exponential growth. This paper views ``the end of AI exponentiation,'' thus exploring how it flutters inside and outside the bubble, where instability emerges within the AI ecosystem through compute and data-center races, speculative investments, and the rat-race toward superintelligence, and outside the ecosystem through labor disruption, governance concerns, public uncertainty, and geopolitical acceleration surrounding future intelligent systems and infrastructures globally. --- ### Multimodal Resource-Exhaustion Attacks on Vision-Language Models via Joint Pixel-Prompt Optimization Authors: Zhaoxiong Ni, Yatie Xiao, Chi-Man Pun, Fei Peng, Qingxiao Guan, Keke Tang Published: 2026-09-05 Categories: cs.AI, cs.CR, cs.CV arXiv: https://arxiv.org/abs/2609.05889v1 Resource-exhaustion attacks against autoregressive vision-language models (VLMs) typically assume unimodal threat models, treating the image branch as the primary optimization surface while holding user-visible prompts fixed. Even recent loop-centric variants remain confined to this single-channel paradigm, leaving the exploitation of availability unexplored as a cross-modal optimization problem over jointly controllable input surfaces. We introduce Joint Pixel-Prompt Optimization (JPPO), the first compound adversarial framework elevating the visible prompt to a first-class adversarial variable alongside image perturbations. Under a restricted joint-input threat model, JPPO performs coupled, stagewise optimization over both the pixel and prompt surfaces. This produces synergistic cost amplification, mechanistically distinct from loop-dependent failures, exhibiting negligible loop incidence in our experiments. Evaluating five open-source VLM families on MS COCO and ImageNet under an 8/255 infinity-norm budget, JPPO achieves over 4.6x latency and 5.3x energy amplification on Qwen2.5-VL-7B, and over 36.6x latency with 32.7x energy amplification on BLIP-2. This represents the strongest cost amplification among directly compared baselines while requiring substantially fewer optimization iterations. Ablations confirm this amplification arises from multimodal coordination rather than prompt length or isolated modalities. These findings reveal structural blind spots in current VLM serving defenses, motivating cost-aware robustness evaluation as a first-class security requirement for multimodal deployments. --- ### UniFusion: Sparse-View 4D Reconstruction via Unified Spatio-temporal Depth Alignment Authors: Yongzhe Lyu, Shaofei Wang, Yixin Chen, Siyuan Huang Published: 2026-09-05 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.05888v1 In this paper, we address the challenging problem of 4D reconstruction from sparse-view videos. This setup usually relies on monocular depth estimation to provide priors for the reconstruction model. A key challenge arises from limited cross-view overlap and temporal variation, making monocular depth predictions inconsistent across views and time. Existing methods align spatial and temporal dimensions in separate stages, requiring foreground segmentation masks while failing to leverage temporal cues for cross-view alignment. Contrary to these methods, we propose a unified spatial-temporal depth alignment framework that jointly resolves cross-view and cross-time inconsistencies without distinguishing foreground/background. Our method represents depth maps across views and time as a set of spatio-temporal neural fields. This representation not only yields fast convergence, but also captures spatio-temporal correlation among depth maps implicitly, without dependence on external segmentation/tracking models. We also propose a multi-view depth-order loss while leveraging the classic scale-and-shift-invariant loss to further improve the final depth quality. The aligned depths initialize and supervise Gaussian splatting models for 4D reconstruction. Experiments on Ego-Exo4D and EgoHuman demonstrate that our improved depth alignment substantially benefits dynamic Gaussian-splatting-based reconstruction methods for novel-time/view synthesis and geometry accuracy/consistency. --- ### One Rate Is Not Enough: Adaptive Anisotropic Learning Rates for LoRA Fine-Tuning Authors: Huiyi Wang, Daijiao Liu, Lina Yao, Dong Gong Published: 2026-09-05 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.05885v1 Low-rank adaptation (LoRA) has become the standard for parameter-efficient fine-tuning of large language models. Most LoRA variants follow a uniform-LR convention, applying a single global learning rate across every rank-one component of every adapter. We show that this convention overlooks substantial within-module heterogeneity, where the rank-one components of a LoRA adapter update at highly uneven rates and low-velocity modules converge to concentrated singular spectra that underutilize the nominal rank budget. To address this, we propose an adaptive anisotropic learning-rate model that assigns each rank-one component its own effective learning rate, computed online from training-time signals and mean-normalized per module to preserve the global LR budget. AnLR-LoRA instantiates this model with two signals available during AdamW optimization, namely function-space velocity and Adam SNR, as a lightweight scheme with no extra trainable parameters. Across commonsense reasoning, natural language generation and visual instruction-tuning benchmarks, AnLR-LoRA consistently improves over LoRA while encouraging broader use of rank capacity, with gains that remain robust across a wide range of global learning rates and transfer cleanly to other LoRA variants. --- ### CALM: Class-wise Agreement and Label-gated Disagreement Modulation for Decentralized Federated Learning Authors: Yifan Ying, Qing Tian Published: 2026-09-05 Categories: cs.LG, cs.DC arXiv: https://arxiv.org/abs/2609.05884v1 Conventional federated learning relies on parameter averaging, which forces clients to be doubly homogeneous: all must run an identical architecture, and accuracy degrades when local data are non-IID. Decentralized federated distillation sidesteps both: each client runs its peers' model snapshots as teachers on its own local data and distills from their soft predictions, with no server, no public data, and no shared architecture. Under severe non-IID skew, however, the trustworthiness of the aggregated teacher target is a matter of degree, yet existing pipelines make hard, all-or-nothing decisions: outlier teachers are discarded by threshold, and whatever target survives is trusted in full. We propose CALM, which replaces every hard decision with a smooth trust gate at three levels: per class, teachers are weighted by agreement with the peer consensus; per sample, distillation is scaled by the teachers' divergence from that target; and a label gate scales it by how strongly the target supports the sample's true label. None of this adds communication or auxiliary data. On CIFAR-10, SVHN, OrganAMNIST, and Google Speech Commands with heterogeneous client architectures under Dirichlet label skew, CALM consistently outperforms uniform and hard-filtered distillation and matches or exceeds competing heterogeneous-FL methods. --- ### What if LLMs Ate Their Words: Causal History Effects in Multi-Turn Interaction Authors: Jinnan Li, Zheren Fu, Yue Wang, Jinzhe Li, Yuan Wu, Yi Chang Published: 2026-09-05 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.05882v1 Multi-turn interaction creates a feedback process in which an LLM's previous responses become context for later behavior. Prior work shows substantial multi-turn degradation and that assistant-generated history can affect later behavior. However, it remains unclear how these effects manifest across models, tasks, turns, and inside a model. We study these gaps across six task families and five models. Degradation from fully specified single-turn input (FULL) to progressively revealed multi-turn interaction (SHARDED) is clearly task- and model-dependent, and stronger one-shot performance does not imply greater interaction robustness. We then retrospectively analyze completed SHARDED conversations by replaying the user messages already observed in each trajectory while editing only assistant-generated history. Replacing prior assistant responses with neutral content (termed neutralization) changes downstream min-max normalized performance by +.027 across 2,973 trajectories. On a prespecified length-controlled subset, short and length-matched neutralization yield nearly identical effects (+.069 versus +.068), showing that simple context shortening is insufficient to explain the effect of history editing. Turn Surgery further intervenes on one assistant turn at a time. Among 237 selected degraded trajectories, 63.7% contain at least one beneficial intervention, while most tested positions remain unchanged; for binary tasks, 48.4% admit a fail-to-success reversal. An open-weight case study links behaviorally consequential history changes to measurable downstream state differences, but finds task-dependent rather than universal internal signatures. Overall, assistant-generated history has active but selective effects on multi-turn performance, motivating selective rather than uniform history management. --- ### Broken on Arrival: Silently Defective LLM Artifacts in Public Model Registries and How to Catch Them Authors: Aditi Patodiya Published: 2026-09-05 Categories: cs.SE, cs.LG arXiv: https://arxiv.org/abs/2609.05881v1 Developers increasingly run large language models locally by pulling quantized GGUF artifacts from public registries, yet nothing in the distribution pipeline functionally tests these conversions before they reach users. We executed 327 quantized code-capable model artifacts: 305 from the official Ollama library, spanning 15 model lines at every eligible quantization level at or under 8 GB, and 22 from the most-downloaded community repositories on HuggingFace. Each ran a 15-task smoke suite calibrated so that healthy artifacts pass while a known-broken one fails; suspects then faced full 164-task evaluation, a second inference backend, an independent distributor's conversion of the same model and quantization as referee, and, for community files, re-testing under the artifact's own template. The official library carries five silently defective artifacts, a batch of four Qwen2.5-Coder-3B conversions and one phi3.5-mini conversion, that solve zero of 164 tasks and zero of the smoke suite on both backends while independent conversions of the same models work: 1.6% of official artifacts, 2 of 29 model-and-size conversion groups. The adjudication chain cleared small-model artifacts that a naive threshold would condemn as broken when they are merely collapsed by extreme quantization, and it exposed two older community conversions that degrade badly on CUDA yet pass on Metal: not defective files but backend-dependent failures, a third phenomenon no registry currently tests for. Two confirmed defects produce output whose surface statistics sit inside the healthy range, invisible to any low-noise heuristic short of execution. We release the audit dataset, the quantcheck acceptance-testing tool, and disclosure reports for every confirmed defect (https://github.com/aditi-p31/quantcheck), and argue that model registries need the acceptance gate that package registries already run. --- ### Grounded and Faithful P&ID Reasoning: Constraining Vision-Language Models with Recovered Evidence Graphs Authors: Prathamesh Gadekar, Sagar Srinivas Sakhinana, Venkataramana Runkana Published: 2026-09-05 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.05880v1 Piping and Instrumentation Diagrams (P&IDs) are the authoritative maps of process plants: isolation, maintenance, and HAZOP decisions depend on what connects to what. Vision-language models describe these sheets fluently, yet they often invent or miss process connections---and an invented or missed link can reverse an isolation or reachability call, so a plant decision cannot trust a fluent answer that was never checked against the linework. We instead recover an explicit graph of the drawing---its symbols, the process connections between them, and the tags that name them---and then require the model to answer only by querying that graph through seven read-only operators, so a topology claim is returned only when it cites the query results that support it. On TopoPID-VQA, a new suite of 3000 topology questions over these sheets, Graph-Grounded Harness (Ours) raises exact match accuracy from 36.7--41.3% under image-only prompting to 74.3--76.0% for Qwen3-VL-4B, Qwen3-VL-8B, and Gemma-4-E4B. It does so on an imperfect substrate: on Digitize-PID dataset the recovered graph scores F1 0.742 on exact process connections, and 0.801 once symbols and tags are pooled in. The residual errors track that gap---grounding pays off where the recovered graph is right, and perception error still breaks topology questions where it is not. --- ### A budget-dependent crossover between coverage- and response-based training-set selection for machine-learned interatomic potentials Authors: Jia Bi, Alin-Marin Elena Published: 2026-09-05 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.05877v1 Selecting compact training sets for machine-learned interatomic potentials requires deciding whether to preserve structural diversity or target configurations on which models disagree. The better choice can depend on how much data is retained, making a comparison at one training-set size insufficient. Here we link selection criteria to prediction accuracy through a budget-resolved comparison of retrained MACE models on GAP-20 Carbon and pooled revised MD17. Structural coverage is compared with a response-guided selector that targets disagreement between a coverage-trained model and a full-data reference. This retrospective response witness tests the value of model disagreement for compressing an already labelled pool. At 5\%, coverage gives smaller absolute deviations from the full-data error than random sampling across four force endpoints in both datasets. The witness has larger deviations than coverage at 1\% and 5\%, but the ordering reverses at 20\%. At 20\%, witness-selected models also lower direct held-out force errors by 0.46--5.89\% relative to coverage, with all eight paired training-seed intervals favouring the witness. Six errors fall below the full-data reference. Mean force-error reductions are 0.164--0.167~meV~$\textÅ^{-1}$, with larger gains for tail and masked endpoints. Complementary analyses show that learned similarity preserves the coverage ranking, while selecting by frozen-model error gives higher error than embedding coverage. These findings establish retained-data budget as a deciding variable in atomistic training-set selection and provide a direct test of when response-guided compression improves on structural coverage. --- ### FACT: A Forensic Agent with Compiled Tool-Use Trajectories for AI-Generated Image Detection Authors: Jiaoyang Chen, Bin Hu, Jingyu Hu, Kun Zhou, Qin Zhang, Zhengzhe Liu Published: 2026-09-05 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.05876v1 AI-generated image detection is increasingly open-world: new image generators produce highly realistic images that make visual artifacts harder to identify. Existing detectors usually rely on a fixed set of forensic cues, so a detector that works well for one generator family may fail on another. We introduce FACT (Forensic Agent with Compiled Tool-use Trajectories), which learns an image-conditioned tool-use policy for forensic analysis. Instead of applying a fixed detector, FACT decides which forensic tools to call, interprets the returned evidence, and stops when sufficient evidence has been collected. FACT follows an Evolve--Distill--Refine pipeline: it evolves an execution-verified forensic skill, compiles the skill into action--observation tool-use trajectories, distills them into a compact agent, and refines the policy with cost-aware GRPO. Across two internal and four public benchmarks, FACT achieves the best performance among all compared methods, including on recent unseen generators, deepfakes, and manipulated images. --- ### Where Does the Sound Go? Tracing Acoustic Information Loss in Audio-Conditioned LLMs Authors: Song-ha Jo, Sehyun Lee, Soyoon Kim, Jaesik Choi, Sanghyuk Choi Published: 2026-09-05 Categories: cs.SD, cs.AI arXiv: https://arxiv.org/abs/2609.05871v1 Audio-conditioned language models often underuse acoustic cues such as prosody, emotion, and non-speech sounds, raising the question of whether ASR-supervised frontends discard this information before it reaches the LM. We test whether the frontend is responsible by comparing Whisper-Tiny and Whisper-Small with EnCodec, DAC-VAE, and WavTokenizer in a shared Qwen3.5-4B audio-LM pipeline on ASR, emotion recognition, and sound captioning. Encoder replacement alone does not resolve this underuse: Whisper variants remain strongest overall, including on emotion and environmental sound captioning. To localize the failure, we trace task-relevant information through the encoder, projector, LM layers, and LM head. Linear probes and geometric analyses show that discriminative acoustic structure remains recoverable at the final LM layer, even when MCQA accuracy trails probe accuracy by up to 83 points. Because the answer format and decoding procedure are controlled, this task-dependent gap points to content-specific readout failure rather than generic format bias. LogitLens analyses and a targeted LM head intervention support the conclusion that acoustic underuse is not explained solely by encoder-side information loss and that readout alignment can be a dominant bottleneck. --- ### Hierarchical Prompt Injector for Domain Generalization Segmentation Authors: Xin Kun Lin, Ruoyu Guo, Jiaqi Guo, Maurice Pagnucco, Yang Song Published: 2026-09-05 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.05864v1 Domain Generalized Semantic Segmentation (DGSS) is a challenging task, as vision models often rely on low-level appearance cues that change across domains. In contrast, structural attributes exhibit cross-domain stability, motivating the use of structural priors for DGSS. Existing methods use prompt learning to transfer such priors into DGSS models, but typically encode each class as a single holistic prompt. Moreover, these methods apply prompts uniformly to all pixels, offering no mechanism to adapt when only a subset of object regions is visible due to viewpoint changes, occlusion, and environmental variation. We address this with \textbf{Spatial Hierarchical Prompts (SHP)} that enrich each class with region-level geometric anchors capturing structural appearance from distinct viewing angles, ensuring complementary coverage under arbitrary viewpoints. Additionally, we propose the \textbf{Hierarchical Prompt Injector (HPI)}, which enables spatially adaptive prompt injection in foundation models. HPI spatially grounds prompts by modeling their semantic relevance and spatial influence with visual features. Considering the difficulty of learning spatially and semantically aware prompt injection, we further introduce auxiliary supervision to align hierarchical prompts with their corresponding object regions. We achieve 70.62\% and 72.74\% mIoU on synthetic-to-real and real-to-real benchmarks, respectively. Code and checkpoints are released at https://github.com/MosukFate/HPI --- ### Budgeted Task-Aware Acquisition of Dynamic Networks Authors: Zihe Zhou Published: 2026-09-05 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.05862v1 Learning on dynamic graphs is difficult when changes in the underlying network are only partially observed. Acquiring current graph information incurs observation and computational costs, making complete updates impractical under limited resources. This paper focuses on budgeted task-aware acquisition on dynamic networks, where a model needs to decide which stale graph information to refresh for a downstream task. We propose Scout, a lightweight framework that learns the task value of querying each node from the maintained graph and observation history. Our evaluation covers one synthetic and four real-world dynamic networks, two downstream tasks, nine acquisition baselines, and several query budgets. Scout achieves the highest mean downstream performance in 19 of the 21 benchmark-budget settings. Task-utility supervision also outperforms structural-change supervision in 13 of the 16 real-world settings. On the same dynamic network, task-matched acquisition improves link-prediction AUC by 0.012-0.016 and node-classification accuracy by 0.064-0.09 over task-mismatched acquisition. These results show that useful graph observations depend on the downstream task and that limited observation budgets can be allocated more effectively by learning directly from downstream utility. --- ### SLA-Safe Energy Control for AI-Native NG-RAN Using Stability-Aware Constrained PPO Authors: Dharmendra Kumar Published: 2026-09-05 Categories: cs.NI, cs.LG, eess.SP arXiv: https://arxiv.org/abs/2609.05861v1 One important AI-for-RAN use case is energy saving, in which radio resources and cell energy modes must be dynamically controlled without violating user quality-of-service (QoS) or service-level agreement (SLA) requirements. However, aggressive sleep-state or deactivation decisions may reduce energy consumption at the cost of throughput degradation, delay increase, SLA violations, and unstable mode switching, especially under time-varying and bursty traffic conditions. This paper proposes a stability-aware constrained reinforcement learning framework for SLA-safe energy control in 5G NG-RAN. The problem is formulated as a constrained Markov decision process in which an AI-native controller selects closed-loop energy-saving actions based on cell load, queue status, active-user information, current energy mode, and SLA-related indicators. The proposed framework uses constrained proximal policy optimization with adaptive Lagrangian penalties to account for throughput, delay, and SLA constraints. To improve operation under traffic distribution shift, the controller is trained using mixed nominal and stress traffic regimes, while a switching-stability penalty is introduced to reduce oscillatory transitions between active and low-power modes. Simulation results in a seven-cell NG-RAN environment show that the proposed controller reduces energy consumption by approximately 41.4% under nominal traffic, 10.5% under stress traffic, and 22.9% under unseen-stress traffic relative to the Always-On baseline. Under stress and unseen-stress traffic, the controller preserves zero SLA violation and zero throughput loss, indicating service-preserving operation under challenging conditions. The proposed method also reduces switching activity compared with basic threshold-based energy saving. --- ### Beyond Arbitrary Geometry: Topology Generalization In neural PDE Operators Authors: Peiyao Chen, Zhouyuan Xu, Ran Ding, Jiansheng Fan, Chen Wang Published: 2026-09-05 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.05860v2 Neural operators that accept arbitrary meshes are often treated as geometry-general, but unseen domain topology changes both the invariant and decaying subspaces of a PDE operator. We use Hodge heat flow as a controlled lens on this distinction and introduce TopoBox-3D, where tunnels and cavities vary Betti support while the exact Hodge decomposition separates the harmonic kernel from the positive spectrum. Across six architectures, models that infer topology implicitly suffer excess matched degradation in 37 of 45 model--task topology-OOD cells, yet cases that change harmonic dimension are not more strongly penalized on average. The dominant difficulty is instead spectral: the initial Rayleigh quotient is the most stable predictor of error, and spectral broadening adds information for edge and face cochains. Most strikingly, controlled probes show that explicit incidence and harmonic coordinates do not yield the best kernel-identity accuracy; nevertheless, TNO ranks first in mixed-input nonharmonic accuracy on all six tasks with nontrivial harmonic support. Together, these results establish topology as a distinct generalization axis beyond arbitrary-geometry compatibility and show that its influence extends across the Hodge spectrum rather than remaining confined to the harmonic kernel. More broadly, they suggest that global, low-frequency structural priors may help organize predictions in the faster-decaying complementary component, offering a new perspective on how neural operators may generalize across topology as well as geometry. --- ### Selective Posterior Margin Regularization for Forward-Corrected Classification Authors: Zexing Zhang, Jichao Li, Tianyang Lei, XiongYi Lu, Yang Kewei Published: 2026-09-05 Categories: cs.LG, cs.CV, stat.ML arXiv: https://arxiv.org/abs/2609.05859v1 Learning with class-conditional label noise often relies on a transition model from latent clean classes to observed annotations. Forward correction embeds this transition in the likelihood, yet finite-sample networks may still memorize corrupted labels. The corrected likelihood also induces a reverse posterior over the clean classes that could explain each annotation. When its leading class differs from the annotation, the model and transition matrix provide evidence against that annotation, but the leading alternatives can remain nearly tied. We introduce Selective Posterior Margin Regularization (SPMR), which preserves the Forward objective and converts this disagreement into a graded update on the clean classifier. SPMR selects the leading reverse-posterior class, scales a detached pairwise margin by the separation between the two leading posterior classes, and assigns correspondingly little influence to diffuse conflicts. The gap factorizes into transition- adjusted pairwise separation and the posterior mass carried by the leading pair. The active margin follows the locally minimum-norm logit direction that enlarges the selected pairwise margin. Across five known-transition benchmarks, SPMR improves full-length Forward by 2.5-7.0 percentage points and remains 0.7-2.5 percentage points above Forward with Mixup and early stopping. Matched interventions support distinct gains from the posterior-space coefficient, transition-adjusted target, and pairwise action. The same design transfers to estimated transitions, human annotations, architectural changes, and stronger Forward recipes. The formulation uses latent-class evidence already available inside Forward correction without promoting every posterior conflict to a corrected label. --- ### SAFEGuard: Detect Optimization-Based Jailbreak Attacks Through Harmful Semantic Analysis and Fluency Measurement Authors: Quoc Viet Vo, Trung Le, Damith C. Ranasinghe, Ehsan Abbasnejad Published: 2026-09-05 Categories: cs.LG, cs.AI, cs.CR arXiv: https://arxiv.org/abs/2609.05850v1 Despite the significant efforts devoted to aligning large language models (LLMs) with human values and ensuring safe deployment, recent work has revealed that LLMs remain vulnerable to adversarial jailbreak attacks that can bypass safety guardrails and elicit harmful responses. Many defense methods are proposed to detect jailbreaks but they are limited in their effectiveness to counter wide-range optimization-based jailbreak mechanisms that can yield highly fluency-optimized or harmful semantic obfuscated prompts. To tackle this challenge, we propose a unified detection framework SAFEGuard which incorporates a hybrid fluency measurement based on cross-layer distribution distance and perplexity, and the analysis of harmful semantics through gradient matching. Our method is grounded in a paramount observation: high fluency prompts maintain their malicious intention close to harmful prompts while harmful semantic obfuscated prompts often inject gibberish token sequences. Our evaluation demonstrates that SAFEGuard consistently outperforms state-of-the-art baselines and achieves significant improvement in accuracy across different optimization-based jailbreaks. This underscores the effectiveness of SAFEGuard against evolving jailbreak attacks. ---