# Research Papers — 2026-09 · bundle page 243 of 266 > Full-text markdown of papers #12101–12150 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/243 [← Previous bundle](https://aipod.dev/papers/archive/2026-09/page/242.md) [Next bundle →](https://aipod.dev/papers/archive/2026-09/page/244.md) ### IndicSafeEval: Safety Robustness of Large Language Models under Multilingual Persuasive Jailbreak Attacks Authors: Saikat Mondal, Mamta, Deeksha Varshney, Oana Cocarascu, Asif Ekbal Published: 2026-09-03 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.03781v2 Large language models (LLMs) are increasingly used in multilingual settings, yet their safety is still evaluated primarily in English. This limits our understanding of how alignment failures manifest in low-resource and culturally diverse languages. We introduce IndicSafeEval, a persuasion-based jailbreak evaluation framework for Indian languages. Our benchmark combines ten safety critical content categories with six human-like persuasive strategies across four different Indian languages, such as Hindi, Bengali, Marathi and Punjabi, resulting in 7,200 adversarial prompts. We conduct a systematic black-box evaluation of several open-source LLMs to examine how their safety behaviour varies across languages, persuasion strategies, and risk categories. Our analysis shows that the model does not behave equally safely across all languages and prompt styles. Instead, safety performance depends strongly on both the languages used and the way a request is phrased using persuasive cues. We further observe that different risk categories exhibit different levels of vulnerability, with some types of harmful content being significantly more susceptible to persuasion-based jailbreaks than others. These findings reveal important limitations of current safety evaluations, which are largely English-centric, and underscore the need for multilingual and persuasion-aware benchmarking frameworks to more accurately assess real-world LLM safety. Our implementation is available at https://github.com/MonSaikat/IndicSafeEval. Warning: this paper contains example data that may be offensive or harmful. --- ### Typological Feature Prediction with Large Language Models: An In-Context Learning Approach Authors: Qianwen Wang, York Hay Ng, Aditya Khan, En-Shiun Annie Lee Published: 2026-09-03 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.03775v1 Typological features are widely used in multilingual NLP, and the prediction of such features holds downstream utility. However, existing methods to predict missing values lack interpretable justifications for predictions, while their performance across resource levels and feature types remains underexplored. Given LLMs' abilities in meta-linguistic reasoning and in providing rationales, we investigate LLMs' performance in typological feature prediction via an in-context learning approach with linguistic data from URIEL+ and Glottolog. We find that zero-shot prompting is insufficient, but when given phylogenetic and geographic neighbour evidence, LLMs substantially outperform all baselines without disadvantaging low-resource languages. We further find that most LLM rationales are consistent with the provided evidence, offering a step toward explainable typological feature prediction. --- ### Rethinking World Models for Safety-Critical Embodied Systems Authors: Kailang Ma, Heye Huang, Inhi Kim, Kitae Jang Published: 2026-09-03 Categories: cs.AI, cs.RO arXiv: https://arxiv.org/abs/2609.03774v1 World models have progressed from compact latent dynamics to generative, controllable, and interactive simulators of embodied environments. However, high predictive likelihood and visual fidelity do not necessarily ensure that a model preserves the evidence required for safe decision-making. This perspective identifies three structural mismatches in current world modeling: likelihood versus risk, prediction versus intervention, and finite-horizon prediction versus accumulated consequences. We propose the Risk-Informed World Model (RIWM) as a decision-centric research direction for safety-critical embodied systems. RIWM organizes world modeling around consequences, intervention, epistemic uncertainty, and recoverability, and integrates four interdependent capabilities: decision-relevant representation, counterfactual reasoning, safety-critical episodic memory, and runtime safety assurance. It distinguishes physical, social, and operational consequences while using epistemic uncertainty to qualify the evidence supporting action. We further discuss open challenges in identifying consequential futures, validating counterfactual reasoning, maintaining revisable safety memories, translating learned consequences into executable constraints, and determining when evidence is sufficient to act. This perspective argues that future world models should move beyond predicting likely futures toward identifying which futures matter, revising judgments through experience, and recognizing when to act, revise, sense, defer, or abstain. --- ### RealCADBench: Benchmarking Parametric CAD Modeling from Industrial Design Intents Authors: Guanlin Li, Zhichao Huang, Huimu Yu, Yichen Long, Hongsen Liu, Linxin Cai, Qiuhe Hong, Zongzhen Li, Wei Wang, Xianwen Zhong, Yuchen Wang Published: 2026-09-03 Categories: cs.CV, cs.CL arXiv: https://arxiv.org/abs/2609.03773v2 Parametric computer-aided design (CAD) modeling is difficult to evaluate with a single metric. Existing CAD benchmarks often emphasize synthetic or CAD-native settings, limited input modalities, or executability and IoUs alone. We introduce RealCADBench, a benchmark for intent-to-program CAD modeling from real industrial design intents. It contains 12,632 tasks from 19 factory-automation categories and spans text descriptions, 2D engineering drawings, real product pictures, and rendered images for both Part and Assembly modeling. We report results on a 1,770-task evaluation slice: 1,745 Part tasks across four input regimes and RCB-Assm25, a 25-task assembly study used in every reported assembly comparison. Each method generates FreeCAD API Python, which a shared runtime executes to export the 3D model. We evaluate the exported model using executability, Solid IoU, Surface IoU, and a rubric-based visual-semantic identity Judge. Among the nine standalone frontier large models evaluated, no model leads all four metrics. Across six frontier-scale large models, executability ranges from 0.565 to 0.812, Solid IoU from 0.2841 to 0.5379, and Surface IoU from 0.112 to 0.217 across the four Part regimes. The highest regime-balanced composite comes from a different model than the leaders on the four component metrics. On RCB-Assm25, Codex with GPT-5.5 improves executability and both IoU metrics over standalone GPT-5.5, but lowers the Judge score by 6.98 percentage points, leaving GPT-5.5 as the Judge leader. We also observe recurring failure modes, most notably missing fine structures, loss of part identity, and incorrect assembly placement. These results show that execution alone is insufficient to characterize realistic CAD modeling and that frontier models and agents differ substantially across executability, IoUs, and visual-semantic identity. --- ### OBER+: Continuity-Aware Reporting and Traceable Continuous Improvement in Outcome-Based Education Authors: Elakkiya Rajasekar Published: 2026-09-03 Categories: cs.LG, cs.CL, cs.CY arXiv: https://arxiv.org/abs/2609.03770v1 Institutions practising outcome-based education compute learning outcome attainment routinely, while reviews of curriculum analytics report an absence of evidence on how that computation informs decisions. This paper presents OBER+, an extension of a deployed institutional attainment platform that computes the step from a measured shortfall to an evaluated corrective action. Five connected stages accumulate attainment across deliveries of a course, signal a shortfall and a persistent shortfall, grade it on cutoffs the regulator already uses, record the decision against a catalogue of practices annotated with their evidence, log the change, and quantify the subsequent movement in the shortfall. A further rule compares successive statements of an outcome, so attainment is never read as a series across a point at which the outcome changed. Applying the rules to the live record of two real courses produced three results. Every outcome of a core course was substantively redefined between consecutive deliveries, with subject matter moving between outcome numbers, so a naive reading would have reported a twenty-five point collapse between quantities that do not refer to the same learning. Recomputing the platform's figures from its documented rule showed six of ten differing by more than rounding explains, in a pattern that identified a defect since reported to the institution. Across fifteen statement pairs from three transitions, five were identical character for character, and among the ten that were not, the outcome carrying a given number was nearest to a differently numbered earlier outcome in six, a result resting on an ordering of similarities and requiring no threshold and no labelling. The contribution is a computational design for outcome-based reporting, stated as rules any attainment platform can implement, with evidence of what they make visible in a live institutional record. --- ### From Nowcasting to Forecasting: Adapting a Reanalysis-Trained Authors: Mikko Partio, Leila Hieta, Ossi Laine Published: 2026-09-03 Categories: cs.LG, physics.ao-ph arXiv: https://arxiv.org/abs/2609.03763v1 Accurate cloud-cover forecasts are important for temperature prediction, radiation forecasting, and solar-power operations. Short-range forecasting methods can preserve observed cloud placement during the first forecast hours, but their skill decreases when cloud fields evolve through formation, dissipation and deformation. Longer lead times require accounting for atmospheric evolution, but operational numerical weather prediction (NWP) forecasts may not accurately represent the satellite-observed cloud state at initialization. We develop CloudCast v2, a machine-learning model for 12-hour cloud-cover forecasting from observation-based initial conditions. The model is first trained on the Copernicus European Regional Reanalysis (Ridal2024) to learn cloud-evolution dynamics, and is then adapted to satellite-derived cloud fields using conditional flow matching (Lipman2023), a generative method that transforms noise into cloud-cover forecasts conditioned on the observed initial cloud fields and NWP inputs. CloudCast v2 reduces mean absolute error by 10% relative to its predecessor, CloudCast v1 (Partio2025), over the 1-12 h range. It also overtakes CloudCast v1 in fractions skill score, a neighborhood-based measure of spatial agreement, after approximately 3-6 h, depending on the cloudiness category. These results show that observation-initialized machine-learning forecasts can extend beyond the usual 1-3-hour nowcasting range while retaining spatial detail from satellite cloud fields. --- ### Projected Riemannian Gradient Descent for the Bures-Wasserstein Barycenter: Dimension-Independent Linear Convergence at Unit Step Size Authors: A. Afham Published: 2026-09-03 Categories: cs.LG, math.OC, quant-ph arXiv: https://arxiv.org/abs/2609.03762v1 The computation of the Bures-Wasserstein (BW) barycenter of an ensemble of positive definite matrices arises throughout machine learning, optimal transport, and quantum information. Riemannian gradient descent (RGD) at unit step size -- the fixed-point iteration used in practice -- converges rapidly, yet existing analyses present a dichotomy: unit-step guarantees carry worst-case exponential dependence on the dimension, while dimension-independent guarantees require small step sizes that forfeit the empirical speed. We resolve this dichotomy, not by improving the guarantees for unit-step RGD, but by proposing a Projected RGD algorithm that achieves dimension-independent linear convergence at unit step size. The achieved rate, $(1 - κ^{-3/2})$, where $κ$ is the condition number of the ensemble, also polynomially improves on the best small-step guarantee ($κ^{3/2}$ versus $κ^{5/2}$ iteration complexity). The crux is a novel Projection Lemma: clipping the eigenvalues of a positive matrix to an interval $(α, β)$ is the closed-form, non-expansive (1-Lipschitz) BW-metric projection onto the set $\{S : αI \leq S \leq βI\}$ -- a statement which, unlike its known one-sided counterpart, does not follow from convexity. The projection is moreover free: it reuses an eigendecomposition the next iteration must perform in any case, so the projected and unprojected iterations cost the same per step. The same analysis covers the invariant matrix projection problem of Brahmachari et al. (2025), whose fixed-point algorithm we identify as unit-step RGD on a totally geodesic submanifold, thereby extending the dimension-independent guarantee to that setting verbatim. --- ### ENEAS: Embedding-guided Neural Ensemble for Adaptive Segmentation Authors: Javier del Pino, Salvador Rodríguez, Alejandro Garabito, Javier Álvarez, Chema Garabito Published: 2026-09-03 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.03756v1 We present ENEAS, a unified, text-promptable method for instance tracking and semantic discovery. Text-promptable segmentation models, including the latest foundation models such as SAM 3, still suffer from temporal hallucinations, spatial fragmentation, and semantic misclassification: they fail to report target absence when an object leaves the field of view, segment local textures instead of the complete object during extreme close-ups, and prioritize visual features over ontological reality, so that visually similar artifacts such as statues, paintings, or reflections are segmented as target entities. ENEAS works two ways from a single method: precise tracking and high-quality segmentation of a unique instance, and open-concept discovery of every instance a text query names, resolved by a semantic verification layer. For tracking, we extend the geometrically robust SeC architecture, previously limited to point interactions, with a text-prompting adapter and leverage its temporal memory, so that the target is held through disappearance without drifting to distractors and kept whole even when it fills the entire view. For discovery, the verification layer combines high-speed visual embedding matching with conditional VLM refinement, invoking semantic reasoning only for ambiguous candidates, which filters out the ontological errors that visual-only models cannot distinguish while keeping latency low. Designed with 3D reconstruction in mind, where a single misclassified distractor corrupts the asset, ENEAS unlocks high-quality semantic tracking and segmentation of video, of broad libraries, and of collections of temporally or spatially unordered data, together with the discrimination to tell true instances from their doppelgangers: things that look alike but are not the same. The code and models are available at https://github.com/speridlabs/eneas --- ### SimSkill: A Self-Evolving LLM Agent for Skill and Knowledge Accumulation in Traffic Simulation Authors: Qi Liu, Qinzheng Wang, Can Li, Yiming Bie, Wanjing Ma Published: 2026-09-03 Categories: cs.AI, cs.MA arXiv: https://arxiv.org/abs/2609.03753v3 Cumulative culture enables humans to preserve, reuse, and extend knowledge and skills across experiences and generations. Inspired by this principle, we introduce \textit{SimSkill}, a self-evolving agent built around the Simulation of Urban MObility (SUMO) traffic simulator. SimSkill continually identifies capability gaps, generates and solves environment-grounded tasks, verifies solutions through an action--critic loop, and consolidates experience into episodic, procedural, and semantic memory. Through autonomous exploration, it builds a library of reusable skills and knowledge spanning major stages of the traffic-simulation workflow. We evaluate SimSkill on two held-out benchmarks across three backbone LLMs, with each result independently verified. It improves verified success by up to 25 percentage points, and ablations show complementary contributions from procedural and semantic memory. Its benefits remain backbone- and budget-dependent, as memory does not improve every model or uniformly reduce inference cost. More broadly, SimSkill illustrates a natural-language-centered design paradigm for LLM-based agent systems. Its high-level control logic, operating principles, and accumulated knowledge are expressed in natural language, while an LLM integrates them with executable tools and code to realize precise and reproducible execution. All code and experimental data are publicly available at https://github.com/qiliuchn/SimSkill-V1. --- ### Rent-a-RAG: Embedding-Space Watermarks for Auditing Third-Party RAG Authors: Alexandr Goultiaev Tolstokorov, Kyriakos Mouratidis, Javad Dogani, Nikolaos Laoutaris Published: 2026-09-03 Categories: cs.CR, cs.CL arXiv: https://arxiv.org/abs/2609.03749v1 Third-party retrieval-augmented generation (RAG) marketplaces create a new auditing problem: data providers may license corpora to a RAG operator, yet later have no visibility into whether their documents are being reused without compensation. Auditing this misuse is difficult because the operator is non-cooperative, answers are paraphrased by the generator, and one response may combine evidence from many providers. We propose DirBucket, a provider-side semantic watermarking and black-box auditing framework for document-level reuse in multi-provider RAG. DirBucket watermarks documents by meaning-preserving paraphrases whose embeddings are biased toward provider-bucket secret directions, enabling detection from black-box answers while preserving retrieval utility. On a challenging benchmark that reflects mixed-provider reuse under black-box access, DirBucket is the only method that consistently achieves strong target detection with no non-target activation, detecting non-compliance in every audit within 23 audited answers on our primary benchmark. The watermark survives adversarial post-answer laundering, and none of the evaluated evasion strategies simultaneously defeats detection while preserving user-perceived answer quality. Detection transfers unchanged to a second benchmark built from real clinical, cyber-threat-intelligence, and legal provider corpora. These results suggest that embedding-space watermarking can make document reuse in third-party RAG statistically auditable. --- ### KnowVis: Knowledge-Centric Visual Summarization for Video Lectures Authors: Yi Xu, Yifan Hou, Xiaoyu Zhang Published: 2026-09-03 Categories: cs.CV, cs.CL arXiv: https://arxiv.org/abs/2609.03742v2 Video lectures are valuable educational resources, but their dense and lengthy formats often overwhelm novice learners. This difficulty stems from a fundamental pedagogical mismatch: while videos deliver transient information linearly, human learning requires constructing interconnected cognitive networks, a task that induces severe cognitive overload for novice learners lacking prior domain knowledge. Existing video summarization methods fail to resolve this mismatch, as they primarily produce text-heavy, linear condensations that still demand high cognitive effort. To bridge this gap, we propose KnowVis, a framework that transforms linear video lectures into pedagogically grounded visual narratives. KnowVis first extracts a detailed concept map from multimodal video content to identify important and challenging threshold concepts, then constructs structured knowledge units, and finally synthesizes engaging visual summaries. Alongside the framework, we introduce a curated dataset of 125 educational videos across 10 academic disciplines, paired with 1,079 generated visual summaries. Extensive automated evaluations and a human study demonstrate that, compared to state-of-the-art baselines, KnowVis generates more accurate and clear visuals that successfully reduce cognitive load and significantly improve student learning effectiveness and knowledge retention. --- ### Fill My Mirror: Geometry-Constrained Mirror Inpainting Authors: Ofek Basson, Shimon Vainer, Yacov Hel-Or, Ohad Fried Published: 2026-09-03 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.03740v1 Mirrors are common in real-world images, yet producing geometrically consistent reflections with generative models remains challenging. Unlike most objects, mirror appearance depends on scene geometry and viewpoint, making it hard to synthesize using learned appearance priors alone. We address this in the mirror inpainting setting, where the scene is fixed and only the mirror region is generated. Our key insight is that much mirror content is geometrically constrained by the visible scene and need not be hallucinated. We estimate scene geometry and project visible content into the mirror to recover reflection regions determined by geometry. A generative model then completes the mirror region via a two-mask diffusion strategy balancing geometric constraints with the model's learned priors, reducing projection artifacts and improving reflection consistency. The method is training-free and applicable to complex real-world scenes. We evaluate on MirrorBench-V2 (synthetic) and real images. Using standard and geometry-aware metrics, we show that explicitly using scene geometry improves consistency. --- ### Beyond BLEU: A Case for Redefining Sign Language Translation Benchmarks Authors: Oline Ranum, Edward Fish, Simon Hadfield, Richard Bowden Published: 2026-09-03 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.03734v1 BLEU-4 is the standard metric for evaluating sign language translation (SLT), but spoken-language metrics may not adequately reflect sign language proficiency. The multimodal, low-resource context of SLT allows models to exploit spurious correlations and spoken-language priors, rather than learning stronger sign representations. In this paper, we evaluate the relationship between spatio-temporal understanding and BLEU-4 across six SLT models on Phoenix-2014T and CSL-Daily, showing that gains in BLEU-4 are not on their own evidence of better sign language understanding. This work introduces an alternative inspired by language-learning assessment, using an open-weight-LLM QA protocol that measures salient content preservation. It aligns more closely with human rankings and is six to seven times more paraphrase-invariant than BLEU-4. Applied to SLT, this protocol targets content transfer, is more robust to train-test overlap, and gives a different picture of the field: the five gloss-free systems are largely within noise of one another on Phoenix-2014T, while the gloss-supervised system stands 9.3 points higher, a gap invisible to BLEU-4. --- ### Unfold The World: Factorize 4D Properties in Reinforcing Spatial Reasoning Authors: Yijun Yang, Shenghe Zheng, Wenbo Li, Jianhui Liu, Haoze Sun, Yanbing Zhang, Jiaxiu Jiang, Lin Song, Haoyang Huang, Nan Duan, Lei Zhu Published: 2026-09-03 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.03729v1 Despite the remarkable prowess of Vision-Language Models (VLMs) in general multimodal tasks, they remain fundamentally ``flat'' when reasoning about the physical world. We argue that this spatial bottleneck stems from a profound dimensional mismatch: while VLMs are trained to interpret 2D projections, true spatial reasoning demands the recovery of latent 3D geometry and temporal continuity. To conquer this high-dimensional complexity, we advocate a shift from monolithic learning to a ``divide and conquer'' paradigm. We present FactoSR, a factorized reinforcement learning framework that explicitly interpret the dimensions collapsed by visual projection. At its core, FactoSR decomposes the monolithic problem of world-consistent reasoning into three orthogonal, geometric sub-objectives: planar correspondence ($XY$), depth consistency ($Z$), and temporal reversibility ($T$). By optimizing these verifiable constraints within a unified policy learning mechanism, we effectively transform an ill-posed projection recovery problem into a series of tangible reasoning steps. Extensive evaluations on multi-view and video benchmarks demonstrate that this elegant decomposition yields substantial gains in 3D and 4D reasoning, achieving a 5.9% boost on VSI-Bench and 4.5% on All-Angles-Bench. Our findings suggest that reinforcing explicit, factorized 4D consistency is a critical step toward evolving VLMs into robust, world-aware reasoners. --- ### Proactive Service Agents: A Unified Decision Framework, Methods, and Evaluation Authors: Yan Tang, Tingyu Cao, Yuanbo Tang, Huaze Tang, Keer Hu Published: 2026-09-03 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.03727v1 Large language model agents can plan, invoke tools, and modify external states, yet most systems still take an explicit user instruction as a fixed starting point. Proactive service moves the decision upstream: an agent must infer service opportunities from incomplete environmental and user signals, choose among remaining silent, asking, assisting, and acting, and account for interruption, misunderstanding, overreach, and privacy costs. This survey gives an operational definition centered on initiative and formulates the problem as a partially observable sequential decision process constrained by authorization and risk. The formulation represents timing, content, and delivery within one structured action, while making explicit the option value of waiting, the decision value of questions, and feedback-induced state changes. On this basis, we organize existing methods along one decision pipeline (state and need estimation, intervention gating, action construction, and feedback adaptation) and describe prescribed, predictive, model based, and return optimizing mechanisms as nonexclusive policy-construction components. We further normalize decision units and three-axis evidence descriptors across streaming dialogue, screen, video, software-engineering, and human-agent collaboration resources, and formalize metrics for triggering, timing, calibration, user burden, safety, and policy value. The synthesis shows why offline classification performance alone does not predict deployment benefit and why long-term memory is not a defining condition of proactivity. Reliable proactive service instead requires calibrated incremental intervention value, verifiable authorization, recoverable execution, and counterfactual evidence. --- ### Genetic Algorithms for Tractable Bayesian Network Fusion via Pre-Fusion Edge Pruning Authors: Pablo Torrijos, José A. Gámez, José M. Puerta, Juan A. Aledo Published: 2026-09-03 Categories: cs.NE, cs.LG arXiv: https://arxiv.org/abs/2609.03724v1 Bayesian Network (BN) fusion combines multiple input networks into a single structure, balancing dependency preservation with computational tractability. While unrestricted fusion retains all dependencies, it often results in overly complex networks with high treewidth, which affects inference scalability. Limited fusion mitigates this by pruning edges to control treewidth but risks overfitting to input-specific noise and omitting dependencies from the original BNs. This paper introduces a consensus framework that prioritizes shared structures among input networks while enforcing treewidth constraints, ensuring a good consensus. We propose genetic algorithms with advanced initialization, specialized operators, and a tailored fitness function. Additionally, we adapt existing methods to this problem and implement greedy baselines for benchmarking and further optimization. Experiments on synthetic and real-world BNs show the superiority of the proposed genetic algorithms over the adapted methods and greedy baselines. --- ### Can LLMs Extract Architectural Design Decisions from Source Code Commits? - A Preliminary Exploratory Study Authors: Amey Karan, Rudra Dhar, Mohamed Soliman, Karthik Vaidhyanathan Published: 2026-09-03 Categories: cs.SE, cs.AI arXiv: https://arxiv.org/abs/2609.03721v1 Context: Architectural Design Decisions (ADDs) capture the rationale behind the structure and evolution of software systems but are rarely documented explicitly, and are often hidden inside source code commits. Recovering them is important for Architectural Knowledge Management (AKM). Problem: Extracting ADDs from commits is challenging due to their implicit and unstructured nature. Large Language Models (LLMs) have shown strong capabilities in understanding code and text, yet their effectiveness for this task remains underexplored. Study: We present a preliminary study using four LLMs (Gemini 3 Pro, DeepSeek R1, Kimi K2, Qwen3) with zeroshot and fewshot prompting on 30 developer-written ADDs from open-source projects. We score outputs with ROUGE-L, BLEU, METEOR, and BERTScore, and one author manually reviews the Gemini outputs. Results: All models reach a BERT-F1 above 0.81, and fewshot prompting improves alignment (Gemini BERT-F1: 0.828 to 0.847). However, the generated ADDs are often too long, implementation-focused, and miss the rationale behind the decision. This highlights opportunities for architecture-aware LLM systems and automated AKM. --- ### Opening mind by opening architecture: analysis strategies Authors: Francesco Vitucci, Giuseppe Silvi, Daniele Giuseppe Annese, Francesco Scagliola, Anthony Di Furia Published: 2026-09-03 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.03719v1 In numerical signal processing for electroacoustic composition, the progressive loss of specific development and research environments caused by the increasing use of digital market tools has favoured the dominance of the closed-architecture audio processor model. This model, while powerful, envisions the possibility of describing output data about its perceived characteristics, but at the cost of ignoring its internal process and interacting systems, which become complex, powerful environments but closed in an inscrutable black box, a loss we must consider. Any digital signal processing technique tells a story. Just as the words of a language incorporate social, historical and technical polysemic layers, a signal processor has its own story of implementation, a gradual technological achievement with its inevitable aesthetic consequences. Through the looking-glass of literature, one can access those environments with renewed awareness by reestablishing a scientific method and an attitude to research. In this specific case, starting from the case study of Manfred Schroeder's historical reverbs, we illustrate the process of building analytical evaluation tools, as well as practical implementation, at the basis of a conscious study path. --- ### What Do CAE Simulation Agents Really Need Beyond a Generic Harness? Authors: Jiasheng Shi, Tianhan Zhang Published: 2026-09-03 Categories: cs.CE, cs.CL, physics.comp-ph arXiv: https://arxiv.org/abs/2609.03718v1 Computer-aided engineering (CAE) simulation is among the largest and most demanding areas of engineering, where setting up a solver such as OpenFOAM, FEniCS, or COMSOL takes real expertise. Large language model (LLM) agents promise to turn a natural-language request into a working simulation, and recent CAE agents add simulation-specific machinery: multi-agent decomposition, domain retrieval, and scripted reflection. That machinery suited weak base models; modern harnesses already supply multi-turn reasoning, tool use, and execution feedback. We ask what a CAE simulation agent still needs beyond a generic harness. With information access and repair budget held fixed, a single-agent harness matches or beats multi-agent specialized systems (FoamBench 96.4\% vs.\ 88.2\%). Ablations trace this to capabilities the harness already provides: execution-feedback repair lifts FoamBench from 71.8\% with no repair round to 96.4\%, while scripted reflection adds nothing. The one input that still helps is domain knowledge supplied as solver tutorials, our largest measured gain (80.9\% to 96.4\%). --- ### Artificial Intelligence for Energy Optimization in Data Centers Authors: Mohammed Basharath Ullah, Summaiya Unnisa Begum, Mohammed Nadeem Ullah Published: 2026-09-03 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.03716v1 Data centers are increasingly optimized by artificial intelligence and, at the same time, increasingly loaded by it. The literature treats these as two unrelated problems: control studies model workload as an exogenous arrival process, while sustainability studies model infrastructure as a fixed multiplier. We screen roughly 194 papers retrieved through a documented protocol, code 63 of them, and report what the coding shows. Of 28 primary control-oriented studies, 18 are validated in simulation alone and 5 reach physical hardware or a production facility; none account for water withdrawal, and none account for embodied carbon. Reported savings intervals across four technique families overlap almost completely, which means the field cannot presently rank its own methods. Ten recurring gaps are scored for consequence and tractability, and we set out CLEAR-DC, a framework coupling a control-policy branch to a workload-demand branch through an explicit elasticity term, reads out net rather than direct benefit, and emits a schema-conformant record covering energy, carbon, water, embodied share and validation venue. The framework is an architectural and methodological proposal, not a trained system; the contribution we defend empirically is the corpus analysis and the reporting schema derived from it. Coding sheet, derived statistics and all result artifacts: https://github.com/Kimalice/AI-for-Energy-Optimization-in-Data-Centers-Closing-the-Optimizer-Load-Loop --- ### Counterfactual Routing Using Integer Programming with Constraint Generation Authors: Daniël Vos, Sterre Lutz Published: 2026-09-03 Categories: cs.AI, cs.DS arXiv: https://arxiv.org/abs/2609.03707v1 We present our submission to the IJCAI 2025 'Counterfactual Routing Competition' (CRC 25). The goal of the competition is to find counterfactual explanations for the shortest path problem. This requires deciding what the minimal changes to a road network would make a route chosen by the user the optimal route. This enables explanations such as "Your suggested route would indeed have been optimal, if road X were not a bicycle path." Our solution models the problem as an integer program, iteratively incorporating constraints until an exact solution is found. In the final evaluation on held-out test instances, our method ranked fourth in solution quality and obtained its solution fastest on every instance, with an average runtime of 9.0 seconds compared to 118.8 seconds for the next-fastest submission. --- ### Q-Edge: Symmetry-Reduced Quantum Simulation of Structured Extreme Dependence Authors: Hongrui Zhang, Paolo Recchia, Ying Chen Published: 2026-09-03 Categories: quant-ph, stat.ML arXiv: https://arxiv.org/abs/2609.03706v1 High-dimensional simulation of multivariate extremes is fundamentally limited by the combinatorial complexity of dependence, often more than by the scarcity of extreme observations. We show that symmetry admits a lossless orbit-space representation that preserves structured extreme dependence while replacing an exponentially large dependence space with a compact set of symmetry classes. Based on this principle, we develop Q-Edge (Quantum Extreme Dependence Engine), a symmetry-reduced quantum framework that operates directly in orbit space, enabling scalable simulation and digital twins of structured extreme systems. By transferring symmetry into the data representation rather than the quantum circuit, Q-Edge allows unconstrained quantum generative models to exploit dramatically reduced state spaces. For a 30-dimensional problem, approximately 1.6 million angular states collapse to 256 orbit states, reducing the required quantum representation from about 21 qubits to 8. Our results establish a general computational principle for scalable quantum simulation of structured extreme dependence. --- ### Federated Causal Discovery via Regression-Directed Cumulants Authors: Pablo Torrijos, Fabio Stella, José A. Gámez, José M. Puerta Published: 2026-09-03 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.03705v1 In this paper we study linear non-Gaussian acyclic models (LiNGAM) when used in federated environments. These causal models allow one to go beyond Markov equivalence. However, in many domains data are scarce, and increasing the sample size by centralising data from different clients is not advisable due to regulations such as the GDPR. The federated environment offers an attractive option to balance privacy and causal discovery accuracy. Unfortunately, the standard centralised estimator in the LiNGAM setting, i.e., DirectLiNGAM, cannot be straightforwardly federated. Higher-order cumulant tensors offer a way around this obstacle: they depend only on the joint distribution of the variables involved and add exactly across independent sample groups, so a single communication round suffices in horizontal, vertical, and hybrid partitions. However, FedISHC, i.e., the current federated method along these lines, breaks down under near-symmetric noise. To overcome the above limitation, we introduce the FedRCD family of causal discovery algorithms, and investigate three variants that trade off communication rounds against algebraic noise; two of them are exact federated counterparts of the centralised high-order cumulant (HC) and HC-LiNGAM algorithms, and the single-round variants further effectively support exact unlearning at any granularity, from a single observation to a whole client. Numerical experiments show that at sample sizes typical of real deployments, the entire cumulant-based federated family does not actually rank variables by the population asymmetry that the scores encode at zero. It ranks them by a variance ladder induced by the DAG along its directed paths, the cumulant counterpart of varsortability. Marginal standardisation collapses every cumulant method to near-random ordering, while scale-invariant DirectLiNGAM, not federable under this protocol, is unaffected. --- ### Synthetic Semantic Supervision for Contrastive Code Representation Learning in Small Transformers: An Empirical Study Authors: Kenneth Paulsen, Florian Tambon, Mike Papadakis, Shin Yoo Published: 2026-09-03 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.03702v1 General-purpose code embeddings power tools for code search, classification, and retrieval. Compact transformer encoders for code typically rely on either human-written docstrings (labor-intensive and inconsistent) or mined structural signals such as execution traces (setting-specific and costly to collect). We empirically study an alternative: contrastive pretraining of small encoders with synthetically generated natural-language descriptions emphasizing code functionality and intent, paired with code in a dual-encoder framework at training and discarded at inference. We benchmark this approach against pretraining-based baselines, generalist LLMs, and embedding-specific models on eight retrieval, classification, and generation tasks across C, C++, and Java. Synthetic semantic supervision yields statistically significant gains over pretraining baselines of the same inference-time size on five of eight tasks, with parity on two more; once fine-tuned, it matches or exceeds zero-shot models two orders of magnitude larger on classification, and it stays on par with execution-aware supervision at matched pretraining data, suggesting a scalable, effective alternative to existing code-representation paradigms. --- ### Symmetries and Causality: Causal Effect Identification Beyond IID Data Authors: Martin Rabel, Jakob Runge Published: 2026-09-03 Categories: math.ST, cs.AI arXiv: https://arxiv.org/abs/2609.03697v1 In the natural sciences, symmetries and cause-effect relationships are ubiquitous. Yet for complex machine-learning tasks, like world-modeling in reinforcement learning, they appear difficult to harness. We propose a formal description of statistical systems based on symmetries in data leaving causal mechanisms invariant. The result is an abstract, simple and general mathematical language for causal reasoning. This paper provides formal descriptions of models and queries, setting up this language, and the formal infrastructure and strategies for their mathematically rigorous identification from data within this formalism. This approach reproduces and matches standard theoretical results on IID data and transport of experimental and non-experimental data. But its main purpose is to unify and substantially extend the scope of causal reasoning, in going beyond IID data and in approaching complex causal queries not captured by do- or soft-interventions. This new perspective on causally relevant aspects of data-modeling additionally sheds new light on well-known structures like c-components or hedges but also includes aspects of missing data and is inherently well-suited for the description of transfer and robustness properties. --- ### SignSeek: Learning Transferable Representations for Sign Dictionary Retrieval Authors: Sobhan Asasi, Ozge Mercanoglu Sincan, Richard Bowden Published: 2026-09-03 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.03695v1 Sign language dictionaries are essential resources for sign language learners, yet automatically retrieving a sign from a dictionary, given only a query video, remains a challenging problem due to the natural variability between signers. Existing sign representation learning methods are built for closed-set recognition, producing embeddings that do not generalise to the open-set, signer-independent setting that retrieval demands. \textbf{SignSeek} closes this gap by contrastively learning sign representations with saliency-guided articulator masking. A contrastive objective aligns same-gloss signs across signers, while our Articulator Saliency-Guided Masking (ASGM) pinpoints the single most critical articulator per sign. This drives two complementary objectives, a masked contrastive alignment (MAC) loss that sees the sign through a single articulator and a masked prediction (MAP) loss that reconstructs it in latent space from the surrounding spatio-temporal context. Pretrained on 266K samples ($\sim$5,700 glosses) across multiple sign languages, \textbf{SignSeek} sets a new state-of-the-art performance in cross-corpus retrieval on ASL-Citizen, WLASL, and NMFs-CSL without any downstream fine-tuning. Strikingly, it achieves zero-shot generalisation to an entirely unseen British Sign Language (BSL), surpassing methods explicitly trained on BSL, and transfers seamlessly to isolated sign recognition and subtitle alignment, outperforming prior skeleton-based methods. --- ### Observation-Conditioned Latent Energy Priors for Sparse Implicit Neural Shape Completion Authors: Paul Büschl, Ezequiel de la Rosa, Julia Wolleb, Julian McGinnis, César Nombela-Arrieta, Bjoern Menze Published: 2026-09-03 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.03694v1 Implicit neural representations (INRs) can model continuous 3D shapes with a shared coordinate decoder and per-instance latent codes. At test time, autodecoder-style models commonly freeze the decoder and optimize a new latent code from sparse off-grid SDF samples. When these samples underconstrain inference, the latent can drift toward regions that fit the observations but decode implausible unobserved geometry. We propose a post-hoc observation-conditioned latent energy prior for frozen INR decoders. The energy scores standardized latents conditioned on a permutation-invariant encoding of the sparse observation set and is used as a residual expert alongside an L2 latent prior selected on validation data. We evaluate on a controlled cell-nucleus SDF dataset and a public MedShapeNet-derived SDF completion dataset. The proposed L2 objective augmented with conditional energy improves consistently over a validation-selected L2 baseline in the sparsest cell-nucleus regimes and, on MedShapeNet, outperforms both L2 and a six-component GMM latent-density prior across all reported readouts. A shuffled-context ablation is consistently weaker than matched context, supporting an observation-specific contribution. These results suggest that lightweight conditional energies can make pretrained INR decoders more observation-aware without retraining. --- ### AlcaTRAz - Anchored Tree-Rule Defense Against Jailbreaks Authors: Jakub Reš, Petr Kaška, Martin Perešíni, Martin Ukrop, Kamil Malinka Published: 2026-09-03 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2609.03693v1 Large language models (LLMs) are vulnerable to jailbreak attacks that bypass safety alignment through carefully crafted prompts. Many existing defenses require access to model weights or internals, making them difficult to apply to black-box deployments. We propose AlcaTRAz (Anchored Tree-Rule defense Against jailbreaks), a prompt-level defense based on rule trees that operates exclusively on the input text and requires no modification or retraining of the target model. The method automatically learns a transferable transformation rule that inserts controlled character-level perturbations at selected positions, thereby disrupting structural regularities exploited by jailbreak attacks while largely preserving the model's utility on benign queries. We evaluate the proposed method across 33 open-weight models, 22 jailbreak attack types, and a benchmark of short, single-turn benign questions, comparing against three representative prompt-level baselines (Llama Guard, RA-LLM, Goal Prioritization). Among the compared defenses, AlcaTRAz achieves the best composite security and functionality score in 73.4 % of model-attack combinations and shifts the aggregate score from a modal value of 10 (maximal-severity response to the malicious request) in the undefended setting to a modal value of 2 (near-refusal) after defense, while keeping the mean benign score within 0.27 points of the undefended baseline (8.35 vs. 8.62 on a 0-10 scale). AlcaTRAz substantially reduces but does not eliminate jailbreak success: a high-severity tail remains, and we do not consider adaptive attackers, so we position it as one layer within a defense-in-depth strategy rather than a standalone guarantee. --- ### MetaStructAtlas: A Grounded 3D Vision-Language Dataset and Benchmark for Functional and Structural Reasoning in Whole-Body PET/CT Authors: Chenguang Zheng, Le Xue, Yichi Zhang, Wenbo Zhang, Zehui Ling, Gang Feng, Xin Gao, Yuan Qi, Yuan Cheng, Zixin Hu, Mei Tian Published: 2026-09-03 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.03690v1 The joint interpretation of metabolic function and anatomical structure is essential for clinical diagnosis in whole-body PET/CT. Although recent advances in 3D medical vision-language models have demonstrated remarkable progress, current efforts are limited to regional CT imaging, leaving a critical void in comprehensive whole-body PET/CT analysis. In this work, we introduce MetaStructAtlas, a large-scale dataset for grounded whole-body PET/CT interpretation that synthesizes multimodal imaging with integrated anatomical, metabolic, and semantic annotations. MetaStructAtlas provides 490 co-registered 3D PET and CT volumes with 50,470 organ-level segmentation masks and grounded radiology reports. To facilitate interactive reasoning, we further developed MetaStructVQA, a standardized 3D grounded visual question-answering benchmark containing 100,565 QA pairs. This framework explicitly links diagnostic queries to visual evidence across modalities, encompassing anatomical, morphological, and metabolic characteristics. Finally, we evaluate state-of-the-art 3D medical VLMs on MetaStructVQA, establishing a robust foundation for multimodal representation learning and integrated whole-body reasoning in nuclear medicine. --- ### Semantic-Aware Subgraph State Space Model for WSI Classification in Histopathology Authors: Feixing Chen, Hao Lu, Lin Luo, Yan Xu Published: 2026-09-03 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.03689v1 Histopathological subtyping relies on the recognition of characteristic histological patterns. These patterns may be expressed by individual tissue structures or by the spatial distribution and co-occurrence of multiple structures, and they often span irregularly shaped tissue regions, termed semantic units in this work. However, conventional patch-based representations may fragment such units and fail to explicitly preserve their internal spatial organization, while efficiently modeling relationships among numerous spatially separated units remains challenging. To address these limitations, we propose the Semantic-Aware Subgraph State Space Model (SASG-SSM), a flexible and efficient framework for whole slide image (WSI) classification. Semantic-Aware Subgraphs (SASGs) first approximate irregularly shaped semantic units by adaptively grouping spatially connected patches guided by class-agnostic visual-semantic priors. By representing patches as graph nodes with adjacency edges, SASGs preserve their internal spatial organization rather than treating them as an unordered set. A Subgraph State Space Module (SG-SSM) subsequently combines a graph neural network encoder for intra-subgraph topology encoding with a Mamba-based state space encoder for efficient contextualization across large numbers of subgraphs. This module integrates local structural information within semantic units with global contextual information arising from their distribution and co-occurrence across the WSI, while efficiently modeling a large number of spatially distributed regions. Extensive experiments across four WSI subtyping datasets demonstrate consistent advantages over representative state-of-the-art methods. Further evaluations under small-cohort and few-shot settings demonstrate robustness and data efficiency under limited training data. Code will be released at https://github.com/HLSvois/SASG-SSM. --- ### ToPO: Token-Conditioned Preference Routing for Attention-Based Latent Diffusion Models Authors: Juntao Xu, Shihong Li, Hoi Fan Au, Ning Zhu Published: 2026-09-03 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.03688v1 Pairwise preference labels rank complete images, yet Diffusion-DPO applies their effect over many spatial and denoising-time coordinates. For attention-based, noise-prediction latent diffusion, ToPO (Token-Oriented Preference Optimization) constructs a per-minibatch, detached, separable spatial-temporal route from branchwise squared-residual contrast in a frozen reference denoiser. Preferred-branch cross-attention uses content tokens to modulate the spatial factor, and an auxiliary pixel-midpoint ordering term is added without local labels or a learned reward model. In matched three-seed retrainings with a shared update schedule, ToPO has higher endpoint estimates than Diffusion-DPO on all five reported SD-1.5 metrics and on HPSv2, ImageReward, and CLIP for SDXL. It also receives larger raw win shares in an aggregate blind SDXL A/B study. These findings are scoped to the reported equal-update U-Net protocols rather than an equal-compute comparison. --- ### A Circuit for Plural Reference: How LLMs Represent and Retrieve Singular and Plural Entities Authors: Anh Danh, Rick Nouwen, Massimo Poesio Published: 2026-09-03 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.03687v1 Coreference resolution is an important task in contextual reasoning. In this paper, we investigate the mechanism for representing and retrieving singular and plural entities for plural reference. We use a combination of mechanistic interpretability and attention pattern analysis to study the process in which LLMs predict a pronoun to refer back to previously mentioned entities. Using a range of causal intervention techniques, we find a set of attention heads that are responsible for (1) representing coreference information in the input, (2) identifying entities that form a plural reference, (3) transferring the information to the component that is responsible for selecting the antecedents and predicting the pronoun. We also find that LLMs align with humans in preference for plural pronoun. Specifically, entities in a plural construction are more likely to be referred to as a plural entity if they are ontologically similar and are linked by the conjunction "and". --- ### Resolution-Aware Experimental Design under Partial Identifiability Authors: Sofianos Panagiotis Fotias Published: 2026-09-03 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.03686v1 Experimental design is commonly framed as choosing the experiment expected to provide the most information. Under partial identifiability however, persistent nuisance uncertainty can make the same observation carry different structural meanings. We introduce Resolution-Aware Experimental Design (RAED), which selects an experiment by the smallest expected nonempty structural candidate set achievable subject to false-exclusion control. We prove an exact cross-nuisance aliasing separation: an experiment can be preferred by structural and full-latent information gain, average classification, and nuisance-marginalized informativeness while having arbitrarily poorer valid structural resolution. RAED nevertheless preserves the expected ordering under a genuine composite Blackwell comparison. To make this criterion operational, we develop a learned score-based implementation with finite-sample nuisance-average and positive-tail calibration, and characterize a rare-tail sample-complexity obstruction. Under constrained sensing, two subsurface-flow benchmarks exhibit genuine RAED--expected-information-gain (EIG) experiment-selection disagreements, with the clearest and largest held-out resolution differences in WCA. In a fluvial benchmark, tail protection changes the selected physical experiment and replaces hard-region false exclusions primarily with explicit ambiguity. In a mechanistic methane-oxidation benchmark, a prospectively specified 5\% false-exclusion tolerance also yields a nontrivial finite-sample population guarantee for tail-sensitive nuisance risk, with 95\% joint confidence across all three structural families. --- ### DropClick: Semi-Automated One-Click Segmentation for Agricultural Robotic Data Authors: Patrick Zimmer, Michael Halstead, Chris McCool Published: 2026-09-03 Categories: cs.CV, cs.RO arXiv: https://arxiv.org/abs/2609.03680v1 Labelling vision datasets, especially for segmentation tasks, is a laborious and costly process that stymies novel developments in agricultural robotics. In this paper, we present DropClick, a click-guided segmentation tool that simplifies the annotation process. Our system utilises single-click inputs on objects to generate pseudo-labels, which can replace manual annotations. DropClick stands out as it is a semi-automated approach and does not require a click for every object in the scene. It can therefore further reduce the required amount of user input drastically. We evaluate our method on two challenging agricultural robotic datasets, SB20 and BUP20 for plant and fruit segmentation, respectively. DropClick is first trained on a small subset of just 5 images from the original training data. This DropClick model can then be deployed as a one-click segmentation system and achieves comparable or higher performance than other one-click methods achieving an mIoU of 70.0 and 72.6 points, for SB20 and BUP20 respectively. DropClick then excels at maintaining high performance when clicks are not given (e.g. dropped); when 50% of the clicks are missing it still maintains an mIoU of 68.9 and 71.3 points, for SB20 and BUP20 respectively. We validate DropClick as a pseudo-labelling approach by taking its outputs to train a Mask2Former instance-based segmentation model in a semi-supervised manner. In this process, partially removing user input from DropClick yields similar high performance when compared to providing all clicks, at 70.1 vs 70.7 points AP50 for SB20 and no difference for BUP20 at 77.0 for both models; at the same time saving 46.3% of total input for SB20 and 31.9% for BUP20. --- ### Understanding Autonomous Driving Datasets by Describing Differences between Image Subsets in Natural Language Authors: Julian Truetsch, Felix Hauser, Christoph Stiller, Frank Bieder Published: 2026-09-03 Categories: cs.CV, cs.CL, cs.LG, cs.NE, cs.RO arXiv: https://arxiv.org/abs/2609.03677v1 Understanding the composition of large-scale autonomous driving datasets is essential for safety, robustness, and reliable operation across domains. For example, domain shift between locations could lead to the operating environment being misaligned with the training data, resulting in potentially dangerous performance degradation. Yet, existing data analysis pipelines largely rely on metadata, predefined labels, or manual inspection, which provide limited semantic insight or do not scale. This paper studies set difference captioning: given two subsets of images, the goal is to produce a natural-language hypothesis describing differences between the target and reference set. Building on a two-stage formulation, we adapt the method to autonomous driving by focusing on object-centric patches derived from object detection, which simplifies aggregation and enables attribution of differences to specific object instances or categories. To evaluate this setting in-domain, we introduce a new benchmark, AD-Diff Bench. Low-concentration experiments assess the suitability of set-difference-captioning approaches to sparse, real-world differences. We restrict our experiments to open-weight models to support reproducibility and ease of deployment. The proposed benchmark and analysis provide a step towards practical, human-interpretable dataset introspection for autonomous driving datasets. Our implementation and benchmark dataset are available at https://github.com/KIT-MRT/AD-Diff --- ### CoFiE: Coarse-to-Fine Evidence Selection for Efficient Streaming Video Understanding Authors: Jing Jiang, Yiran Ling, Ruonan Li, Dimitrios Stamoulis, Jie Liu Published: 2026-09-03 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.03675v1 Streaming video understanding requires Vision Language Models (VLLMs) to process growing video streams and answer user questions under tight latency constraints. Existing methods improve efficiency through token pruning and memory-bank schemes, but mainly reduce visual tokens after visual encoding. Consequently, downstream token pruning alone cannot substantially reduce end-to-end latency because the expensive frame encoding cost has already been incurred. We propose CoFiE, a Coarse-to-Fine Evidence Selection framework that decouples evidence selection into a coarse, query-agnostic filtering stage before the vision encoder and a fine, query-specific refinement stage during LLM prefill. CoFiE introduces Novelty-Guided Frame Filtering to retain visually distinctive candidate frames and Query-Specific Evidence Refinement to select the frames most relevant to the user query. This design removes substantial redundancy before frame encoding while preserving query-specific refinement once semantic information becomes available. Experiments show that CoFiE establishes a new state-of-the-art accuracy-efficiency trade-off across multiple video understanding benchmarks, reaching 78.86% accuracy on StreamingBench and 68.72% on OvO-Bench, with improvements of up to 3.15% over prior methods. Even with up to 80% evidence-frame filtering, CoFiE outperforms strong open-source multimodal models while improving end-to-end inference latency by up to 2.54 times. --- ### Do Video Generators Track the World Across Segments? A Benchmark and Method for World-State Reasoning in Video Continuation Authors: Yingmao Miao, Pengfei Zhang, Chaoran Xu, Meng Yu, Jing Tang, Xiangxiang Chu, Chao Shen, Chenhao Lin Published: 2026-09-03 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.03673v1 Video generators build long videos by composing shorter parts, either by generating segments one after another or by autoregressively extending chunks. Each new part usually depends on memories of historical observations, such as recent frames, selected key frames, memory banks, or cached features. These memories preserve visible evidence from the past, but current generators do not reliably turn such evidence into a world-state interface: what holds in the video world after previous actions and how it should change under the next prompt. A past frame remains valid history, but it may not describe the state needed by the next segment; some states must instead be inferred from occluded or implicit changes rather than copied from a directly observed frame. This creates a simple but overlooked question for video continuation: given a previous video, its prompt, and a new prompt, can a model generate a continuation that reflects the state determined by both the historical video and the new prompt? To answer this question, we introduce Statebench, a benchmark that targets this gap by testing continuations over three state categories: past-visible states, occluded-process states, and complex-transition states. We further propose Stateagent, which explicitly maintains an entity-state representation, updates it under the new prompt, grounds the predicted post-action state as a future end frame, and renders the next video. Experiments show that our method improves controlled video continuation by raising the all-case state score (SCS-All) from 45.2 to 69.3, and also benefits story generation at the one-minute scale. Code is avaliable at https://github.com/AMAP-ML/StateAgent. --- ### ARCOS: Zero-shot Boundary Localization for Corneal Layer Segmentation Across Optical Coherence Tomography Devices Authors: Nuno Vivas Brás, Benjamin Memmi, Maëlle Bouhassane, Cristina Georgeon, Vincent Borderie, Karsten Plamann, Anatole Chessel Published: 2026-09-03 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.03668v1 Accurate segmentation of corneal layers in optical coherence tomography (OCT) is essential for quantitative assessment of corneal morphology, including layer thickness and structural changes associated with disease or surgery. However, automatic segmentation remains challenging because corneal interfaces are thin, affected by speckle noise, and variable across acquisition devices. In this work, we propose ARCOS, a patch-based zero-shot boundary localization framework for corneal layer segmentation in clinical anterior-segment OCT images. Rather than performing conventional region classification, the method predicts boundary heatmaps for the main corneal interfaces from overlapping native-resolution patches. Patch-level predictions are stitched across the full B-scan and converted into boundary locations to obtain continuous, anatomically ordered layer segmentations. The network combines multi-scale feature fusion with a self-conditioned refinement module that uses intermediate boundary information to improve local heatmap predictions while preserving spatial detail. The method was evaluated on clinical OCT images acquired from multiple devices and compared with representative segmentation baselines using boundary localization and derived thickness metrics. The proposed method achieved an off-by-one boundary localization accuracy of 95.1% and a mean absolute boundary error of 0.514 pixels on the matched-device test set. In zero-shot cross-device evaluation, it maintained an average off-by-one accuracy of 84.3% and a mean absolute boundary error of 0.855 pixels across unseen acquisition devices, outperforming the baseline models. Thickness estimates derived from the predicted boundaries showed low error across corneal regions, supporting the method's use for quantitative corneal OCT analysis. --- ### Out-of-Distribution Generalisation with Sequence Models in Offline Multi-Agent Reinforcement Learning Authors: Oussama Hidaoui, Omer Ebead, Ulrich Armel Mbou Sob, Siddarth Singh, Juan Claude Formanek, Felix Chalumeau, Omayma Mahjoub, Sasha Abramowitz, Ruan John de Kock, Wiem Khlifi, Louay Ben Nessir, Simon Verster Du Toit et al. Published: 2026-09-03 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.03667v1 Generalising to unseen tasks remains a fundamental challenge in offline multi-agent reinforcement learning (MARL). In this work, we present a principled analysis of zero-shot task generalisation in the offline setting and conduct an extensive empirical investigation into the scaling behaviour governing task diversity, dataset size, and network capacity. To facilitate this study, we extend offline sequence modelling architectures to handle multi-task observation and action spaces alongside variable agent counts across tasks. Our primary finding is that scaling task diversity---rather than sheer dataset size is the dominant factor in achieving robust zero-shot transfer. Through large-scale experiments across four challenging environments (Connector, RWARE, SMAX, and LBF), we demonstrate that our multi-task approach achieves a mean improvement of 3.2x on held-out test tasks compared to single-task models and consistently outperforms strong behaviour cloning baselines. These results suggest that the development of generalisable MARL agents should prioritise the diversity of the training distribution with varying numbers of agents, providing a roadmap for scaling offline MARL effectively. --- ### Cross-Dataset Transfer and Reliability of Explainable Artificial Intelligence for RhythmFormer Remote Photoplethysmography Authors: Louis Chen, Torbjörn E. M. Nordling Published: 2026-09-03 Categories: cs.CV, cs.AI, eess.IV arXiv: https://arxiv.org/abs/2609.03663v1 Background. Remote photoplethysmography estimates the cardiovascular pulse from facial video, and its explanations have rested on inspecting heatmaps rather than on quantitative evidence about where a model reads it. We quantified the explanations and asked whether such explanations transfer between datasets and track model performance. Method. We trained eight condition-specific RhythmFormer models on NCKU-rPPG, recorded under three illumination levels, speaking, rotation, and cycling, estimated one heart rate per 5.12-second clip, and set them beside a UBFC-rPPG reproduction. Raw attention, rollout, attention flow, and Beyond Intuition were assessed by skin coverage and the Salience-guided Faithfulness Coefficient (SaCo). Results. Beyond Intuition ranked highest on both datasets, at median coverage 0.789 and SaCo 0.837 on Static level 3 against 0.826 and 0.917 on UBFC-rPPG; lower ranks differed. Within one participant of one condition, neither measure was related to a clip's heart-rate error, waveform correlation, or signal-to-noise ratio on either dataset: 186 of the 252 coefficients fell below $|ρ|=0.10$ and 28 reached $p<0.05$ against the 13 expected by chance. Across the eight scenarios only Beyond Intuition's coverage followed the three performance measures, at $ρ=-0.43$, $+0.57$, and $+0.43$, while the attention-only methods' SaCo ran opposite to each. It failed at 40 lux alone, its median coverage falling to 0.180 and its median SaCo to $-0.178$, whereas motion degraded the estimates far more without such a drop. Conclusions. Skin coverage and SaCo carry information complementary to the performance measures rather than a proxy for them: attributing to the skin does not guarantee an accurate estimate. What an attribution reveals about a condition is where the model looks rather than how faithfully its map is ordered. --- ### Extracting Forgotten Prompts from Targeted Unlearned Models Authors: Au Ashley Hoi-Ting, Meghdad Kurmanji, William F. Shen, Nicholas D. Lane, Ligang He Published: 2026-09-03 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.03662v1 Recent unlearning methods (e.g. NPO, DPO, LUNAR) make use of refusal alignment to suppress forgotten data. However, it has been shown that refusal responses might leave traces of unlearning, and recent attacks have been able to successfully recover some of the unlearned knowledge. In this paper, we uncover a new vulnerability. Existing attacks typically assume that the forgotten prompts are already known to the adversary and focus on recovering their answers. However, we show that the forgotten prompts themselves can be extracted by using the retained data and black-box access to the model. Our attack, Targeted Active Search (TAS), first identifies the forgotten entities by constructing canonical templates and entity pool, and selectively querying the model using the most informative template-entity pair under a limited query budget. Once the entities are identified, TAS instantiates prompt templates with those entities to probe the unlearned model and reconstruct the forgotten prompts. Experiments across three unlearning methods with three datasets and three LLMs shows that TAS recovers the forgotten entity with $100\%$ accuracy and reconstructs up to $95\%$ of forgotten prompts, all while using up to $99.7\%$ fewer queries than naive probing. --- ### Local Updates, Global Learning (LUGL): Playing Games with non-incremental Learners Authors: David Milec, Spyridon Samothrakis, Michael Fairbank, Dennis J. N. J. Soemers Published: 2026-09-03 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.03660v1 The dominance of Neural Networks (NNs) in RL is partially due to their incremental learning capability, which naturally suits the online, non-stationary nature of self-play training. However, gradient-boosted trees like LightGBM are widely recognised as the state of the art for tabular data in supervised learning, often outperforming NNs in accuracy and efficiency. Game states are inherently tabular---discrete actions, categorical card identities, structured board positions---which makes them an ideal candidate for tree-based methods. We introduce LUGL (Local Updates, Global Learning), a framework that decouples data collection from model fitting, enabling non-incremental learners such as GBTs to operate in RL settings where they would otherwise fail due to distributional shift. LUGL alternates between a local updates phase, where the agent plays self-play games and accumulates tabular updates (Q-values, V-values, policies, or regret values) in a finite table, and a global learning phase, where the table is used to train a function approximator that generalises to unseen states before the table is reset. We test our approach in four standard perfect-information games (Tic-tac-toe, Connect-4, Othello, and Hex) and five imperfect-information games (Kuhn's poker, Leduc Hold'em, Liar's Dice, Goofspiel, and Flop5 Hold'em), and show that our results are competitive with or superior to DQN and DeepCFR. Our experiments demonstrate that the community's strong bias towards NNs in game-playing may be unwarranted, since LightGBM-based agents achieve competitive or superior performance across all tested benchmarks. --- ### Rethinking 3D Noise: Learning 3D-Aware Video Priors via Optimization-Free Morphological Perturbations Authors: Onat Şahin, Mohammad Altillawi, George Eskandar, Carlos Carbone, Ziyuan Liu Published: 2026-09-03 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.03657v1 3D scene representations like NeRF and 3D Gaussian Splatting (3DGS) suffer severe artifacts in sparse-view settings. Recent generative 3D artifact fixers attempt to address this, but rely on paired corrupted and clean renders requiring costly, per-scene reconstructions across varying view configurations. While 2D image augmentations act as instant regularizers, no explicit equivalents exist for 3D representations to preserve spatial consistency across views, an essential property for 3D-aware training. We propose 3D Morphological Perturbations as an optimization-free regularizer that preserves spatial consistency. Leveraging explicit 3DGS, we treat each Gaussian as a fundamental building block - analogous to a 2D pixel - and apply perturbations across its morphological parameter space via scale, rotation, and pruning. Our method eliminates per-scene 3DGS optimization loops from dataset curation while enabling models to learn stronger geometric priors than sparse-view baselines in diagnostic ablations conducted on a lightweight video diffusion sandbox. Scaled to a 14B-parameter video model via ControlNet, our approach maintains visual fidelity while reducing mean depth error by 12.5% over state-of-the-art image-to-image 3D artifact refiners, ultimately boosting downstream robotics policy success rates by up to 8.0% across 3 of 4 manipulation tasks. --- ### PL-SCEA: Reconfiguring Pretrained Attention for Few-Shot Industrial Anomaly Detection Authors: Xiaoyu Yang, Qixing Wu, Huixian Zhao, Changlong Jin Published: 2026-09-03 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.03655v1 Vision Foundation Models (VFMs) provide transferable patch representations for few-shot industrial anomaly detection, but their attention computation is typically inherited from pretraining objectives centered on semantic aggregation. This creates a potential mismatch: token relations that support semantic recognition may not adequately expose the localized texture and structural deviations required for anomaly localization. We therefore investigate the hypothesis that the attention computation of a frozen VFM can be reconfigured as a task-relevant component of anomaly detection. We instantiate this idea with Power-Law Self-Correlation Enhanced Attention (PL-SCEA), which retains the semantic context of pretrained query-key attention while constructing token-adaptive self-correlations over contextualized value features. Positive-correlation filtering and power-law reweighting then emphasize relations that are salient relative to each token's relational background, without introducing additional trainable attention projections. The resulting features are modeled by a lightweight variational autoencoder that provides a fixed-size reconstruction-based representation of category-specific normality. The two stages serve complementary roles: attention reconfiguration shapes how local relational deviations are represented, while reconstruction-based modeling converts deviations from learned normality into anomaly scores. Across MVTec AD and VisA, the complete framework achieves competitive image-level detection and consistently strong pixel-level localization across the evaluated few-shot settings. Ablations further show that PL-SCEA improves localization with either the VAE or a memory bank under the tested setting. These results support the view that task-aligned attention reconfiguration can improve the anomaly-localization capability of frozen pretrained representations. --- ### Enhancing Financial Question Answering: A Novel Benchmark Dataset of Banks' financial statements Authors: Arianna Miola, Bruno Spaccavento, Lorenzo Silotto, Marco Bianchetti, Luca Cagliero Published: 2026-09-03 Categories: cs.CL, cs.AI, cs.CE, cs.IR, q-fin.GN arXiv: https://arxiv.org/abs/2609.03654v1 The comparative analysis of banks' financial statements poses significant challenges for automated question answering systems due to their complexity, substantial length, technical language, and inhomogeneity of both textual and numerical content across different jurisdictions and institutions. We introduce FinRAG-QA, a novel benchmark dataset for financial question answering, which comprises 999 practitioner-curated questions on 10 standardised indicators, grounded in 209 annual and Pillar 3 reports from 24 major European and U.S. banks spanning 2019-2023. Unlike prior financial QA benchmarks, which centre on U.S. filings and single-institution analysis, FinRAG-QA targets cross-institutional retrieval over documents averaging 198k words, longer than any existing financial QA resource. On this benchmark we evaluate a multi-stage RAG pipeline and isolate the contribution of each component. Contextual chunk enrichment combined with a retrieval-optimised embedding model raises NDCG@10 from 0.322 to 0.710; conditional on the ground truth being retrieved, a reasoning-optimised generator raises answer accuracy from 44.6% to 79.0% (+34.4 percentage points), at roughly 20x the generation latency. We further show that cross-encoder reranking degrades retrieval when the first-stage ranking is already strong, and that a single top-ranked chunk outperforms larger contexts at generation time. Experiments were run in late 2024-early 2025 with the models available at that time. --- ### The Impact of Synthetic Data Augmentation on Discourse-Pragmatic Function Classification Authors: Sara Sorahi, Kevin Tang, Reza Kazemian Published: 2026-09-03 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.03652v1 Synthetic data augmentation has become a common strategy for addressing class imbalance in NLP, but most approaches focus on the quantity and diversity of generated examples rather than their geometric relationship to real training data. We investigate this question in the context of discourse pragmatic function classification, a task where data sparsity is a structural feature rather than a collection artefact. Using 410 manually annotated instances of the English word look drawn from the British National Corpus, spanning four functions: Attention Signal, Directive, Discourse Marker, and Interjection. We generate synthetic training examples with Llama 3.1 and partition them by their cosine distance from real training data in RoBERTa embedding space. We compare six training conditions that differ in the placement of synthetic examples relative to the empirical decision boundary, while holding augmentation quantity constant across conditions. All augmented conditions improve macro F and accuracy over the real only baseline, but core proximal examples (NEAR) yield the largest gains in macro F (0.113), while a distance balanced mix achieves the highest accuracy (0.748). No condition improves AUC, indicating that augmentation shifts the decision boundary rather than improving the model's underlying probability estimates. These findings suggest that where synthetic examples land in representation space matters as much as how many are generated, with implications for low resource pragmatic classification more broadly. --- ### Relative Prime Factorization and Finite-State Presentations under Fixed Finite-Monoid Observation Authors: Takayuki Kuriyama Published: 2026-09-03 Categories: cs.FL, cs.LG arXiv: https://arxiv.org/abs/2609.03643v1 Let $L\subseteqΣ^*$ and fix a morphism $h:Σ^*\to M$ into a finite monoid. We study exact factorization and canonical presentation in the relative syntactic congruence $θ_{L,h}:=\equiv_L\cap\ker h$. We separate unique factorization from finite direct presentation. An exhaustively computer-checked $36$-element quotient has a unique exact prime factorization for every live non-unit class, yet its valid prime-return rules contain an infinite family, so unique factorization does not imply the finite relative presentation property (FRP), even for a finite quotient. We lift the same defect to a nonregular context-free language with an infinite relative quotient and finite prime spectrum. To isolate the obstruction, we introduce the finite-state relative presentation property (FSRP), in which canonical valid right-hand-side languages are represented by finite residual controllers, and prove $\mathrm{FRP}\subsetneq\mathrm{FSRP}$. We then introduce prime-target left-division determinism (PTLD), which implies unique exact factorization, tail exactness, tail determinism, and a quadratic bound on valid rules. A nonregular deterministic context-free example with a finite group observer satisfies PTLD while lying outside every fixed $(k,\ell)$-substitutable class. Finally, for fixed $h$ we give a strong positive-data learner for the canonical PTLD presentation with polynomial-time hypothesis updates and a finite characteristic sample, together with a limit reconstruction of the canonical FSRP controller from weakly behaviorally correct CFG-valued learners. --- ### Tree-Structured Vector Quantization For Efficient And Progressive Image Compression Authors: Xinkun Wang, Tianyi Xu, Qingyu Luo, Mingming Ma, Changzhe Jiao, Fu Li, Yi Niu Published: 2026-09-03 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.03641v1 Vector-quantization based image compression has achieved strong rate--distortion performance, yet most of them still produce a separate compressed representation for each target bitrate. Such variable-rate behavior allows one model to operate at multiple rates, but it does not necessarily provide a progressive bitstream whose prefixes are themselves decodable and can be refined by appending additional bits. We propose \textbf{Tree-VQ}, a progressive tree-structured vector quantization framework for learned image compression. Tree-VQ organizes discrete codewords as a hierarchical binary tree and represents each latent token by a routed root-to-leaf path. Crucially, every prefix of this path corresponds to a valid quantized representation, so shallow internal nodes serve as coarse reconstruction codes and deeper nodes provide successive refinements. This allows a compressed image to be decoded from an early prefix and progressively improved as more branch symbols are received, rather than being re-encoded for different target rates. To make this structure practical for compression, we introduce a prefix-compatible tree entropy model that codes progressive continuation decisions and routed branch refinements using only causally available decoded contexts. We further use rate-aware refinement scheduling to decide which spatial blocks should receive additional tree bits under a given prefix budget, and hierarchical prefix supervision to ensure that internal nodes are directly decodable at low rates. Experiments show that Tree-VQ achieves a superior performance--efficiency trade-off, delivering the best perceptual compression results with much fewer parameters and lower latency than competing methods. --- ### Stabilizing Camera-Controlled Novel View Synthesis at Inference Time Authors: Prajwal Singh, Arjun Badola, Seema Kumari, Hajime Nagahara, Shanmuganathan Raman Published: 2026-09-03 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.03639v1 Training-free, camera-controlled novel view synthesis from a single image using pre-trained video diffusion models often becomes unstable under large camera motion and long generation horizons. Existing approaches commonly combine several inference-time components, making it unclear which design choices are most important for stability. We show that the main source of stability is simple. Decomposing camera motion into small autoregressive steps limits per-step geometric distortion and reduces error accumulation. A controlled camera-step study shows that performance remains stable for small motions and degrades more strongly as the per-step motion approaches $18$-$20^\circ$. We further evaluate geometry-constrained spatial attention and low-frequency appearance anchoring as supporting refinements, together with an efficient registration-free warping pipeline. Across RealEstate10K and MegaScene, CamTrol++ improves temporal and geometric consistency, downstream 3D reconstruction quality, and generation efficiency over training-free baselines. The method remains effective for 56-frame generation and under substantial controlled depth corruption. These results show that careful control of camera motion at inference time can substantially improve the stability of camera-controlled novel view synthesis without retraining or modifying the diffusion backbone. --- ### Analysis of Prompt Engineering for Drug Toxicity Prediction Authors: Mia MacGregor, Aakash Welgamage Don, Mark Bartlett Published: 2026-09-03 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.03635v1 Clinical trials in the UK can cost up to £1.3 million, with approximately 90% drug failure rate. Toxicity is a major contributing factor in drug failure. Testing is time and cost intensive. In recent years, the use of artificial intelligence has been increasingly explored to aid in the prediction of drug toxicity, with extensive use of large language models (LLMs). However, LLMs can show considerable variation when minor changes are made to prompts, which raises concerns about their sensitivity to prompt engineering. Prompt engineering is used to optimise a prompt given to an LLM to generate the desired output. This paper proposes a method to analyse prompt engineering for drug toxicity prediction. The aim of the paper is to investigate the importance of prompt phrasing for drug toxicity prediction. LLMs were prompted to identify chemical properties of significance when predicting drug toxicity. Prompts were constructed to investigate; job role, prompt structuring, and rule interpretation. LLMs were then used to generate datasets, using the identified features from initial prompting, which were then passed to machine learning algorithms. The experiments show that the natural variance which occurs in LLMs outweighs any fine-tuning of prompts. There were, however, substantial improvements in model performance when using chemoinformatic code to extract features instead of using LLM-generated values. The proposed analysis methodology is applicable to a wide range of prompt types across different areas of bioinformatics. ---