# Research Papers — 2026-08 · bundle page 13 of 240 > Full-text markdown of papers #601–650 of 11995 published in 2026-08, newest first. AI-readable mirrors of the full 11995-paper 2026-08 archive (50 papers per file). Canonical page: https://aipod.dev/papers/archive/2026-08/page/13 [← Previous bundle](https://aipod.dev/papers/archive/2026-08/page/12.md) [Next bundle →](https://aipod.dev/papers/archive/2026-08/page/14.md) ### Graph4BiLO: Graph Neural Network Approximation for Bilevel Mixed-Integer Linear Optimization Authors: Jessica D. Elrefaei, Kaixun Hua, Seungbae Kim, Hoang Nam Tran, Juan S. Borrero Published: 2026-08-31 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2608.30103v1 Bilevel mixed-integer linear optimization problems model hierarchical decision processes in which a leader anticipates the optimal response of a follower. Although expressive, these problems are computationally challenging because lower-level optimality is embedded in the leader's feasible region. Value-function reformulations replace the nested follower optimization with a constraint involving the follower's optimal value, but evaluating this value function exactly can itself be expensive. This paper introduces Graph4BiLO, a graph neural network (GNN) approach for learning bilevel value functions from variable--constraint graph representations. In contrast to fixed-length multilayer perceptron (MLP) representations, the GNN uses shared message-passing parameters and can therefore be applied across multiple problem sizes with a single trained model. The learned ReLU network is encoded exactly as mixed-integer linear constraints and embedded in an approximate single-level formulation. A repair step subsequently re-solves the follower problem for the selected leader decision to recover a bilevel-feasible follower response. We evaluate Graph4BiLO on knapsack interdiction instances with 20--100 items against the exact MibS solver and the learning-based Neur2BiLO method. Graph4BiLO obtains objective values comparable to Neur2BiLO across all tested sizes while avoiding size-specific neural networks. An additional out-of-distribution experiment demonstrates zero-shot transfer from 20-item training instances to previously unseen 40- and 60-item instances. However, embedding message passing at every graph node substantially increases the resulting mixed-integer formulation size and solve time. These results identify a central tradeoff between size-generalizable graph representations and the computational cost of embedding GNNs within optimization models. --- ### SMOTE-VAR: An Uncertainty-Aware Oversampling Method for Predicting Depression Remission in University Students Authors: Dang Nguyen, Arun Kumar A, Taylor A. Braund, Wu Yi Zheng, Debopriyo Bal, Leonard Hoon, Jill Newby, Helen Christensen, Svetha Venkatesh, Alexis Whitton, Sunil Gupta Published: 2026-08-31 Categories: cs.LG arXiv: https://arxiv.org/abs/2608.30102v1 University students experience disproportionately high rates of common mental health conditions, such as depression, which can impair learning, social functioning, and overall well-being. Although lifestyle interventions such as mindfulness and physical activity can reduce the symptoms, many do not achieve symptomatic remission. Developing new approaches to identify students with poor outcomes could enable earlier and more targeted intervention. Machine learning (ML) methods have increasingly been used to predict remission in depressive patients. However, these ML models often suffer from class imbalance, where there may be an unequal proportion of people in the remitted group relative to the non-remitted group. This imbalance can reduce model accuracy and bias predictions. To address this, studies commonly employ the popular oversampling strategy SMOTE. However, SMOTE has a notable limitation: it may generate invalid synthetic minority samples. In a clinical context, these false positives can lead to incorrect risk stratification, potentially delaying necessary escalated care for patients unlikely to remit. In this paper, we introduce a novel and effective oversampling method that addresses this shortcoming. Our approach leverages the variance function of a Gaussian process to estimate the uncertainty of generated minority samples to reduce false positives. We validate our method on a depression dataset collected from university students and demonstrate that it is better than existing oversampling approaches in predicting remission (i.e., treatment outcome). By improving the reliable identification of non-responders, our method provides a robust computational tool to help clinicians rapidly pivot to adjunctive therapies, thereby personalizing and optimizing mental health care pathways. --- ### Not All Synthetic Data Are Equal: Expert-Committee Audit Screening for Imbalanced Crash-Injury-Severity Prediction in Automated Driving Systems Authors: Zewei Li, Qiaoqiao Ren, Hang Yang, S. C. Wong, Stergios-Aristoteles Mitoulis, Yun Ye Published: 2026-08-30 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.29687v1 Automated driving systems (ADSs) are increasingly operating on public roads, raising safety concerns, yet reliable prediction of crash injury severity remains difficult because crash reports are limited, severe outcomes are rare, and injury classes are highly imbalanced. Existing augmentation methods mainly increase minority-class sample size but rarely assess whether generated samples are credible for safety-critical prediction. This study proposes Expert-Committee Audit Screening (ECAS), a credibility-aware sample acceptance framework for ADS crash injury severity prediction under data imbalance. Using 1,477 incident-level ADS crashes from the National Highway Traffic Safety Administration Standing General Order records, ECAS audits generated minority samples through a real-data-only expert committee based on label support, boundary separation, committee agreement, and local plausibility. Within-class percentile normalization and Pareto non-dominated sorting select accepted samples without manually assigned evidence weights. With a fixed backbone combining normalizing flow augmentation and a Tabular Prior-data Fitted Network (TabPFN) classifier, the best ECAS configuration achieved the highest balanced accuracy, macro-F1, and minor-injury recall among all evidence configurations. Local neighborhood analysis showed that ECAS-accepted samples were better supported by nearby real minority crashes than unscreened retained samples. Shapley additive explanations and partial dependence plots further indicated that lower injury severity classes were mainly associated with crash counterpart and pre-crash movement, whereas moderate-plus injuries were more sensitive to posted speed limit and operating context. These findings support a shift from quantity-oriented augmentation to credibility-aware sample acceptance for ADS safety prediction and risk governance. --- ### Named Entity Recognition using Sliding Window Approach Authors: Hariom Ingle, Ronit Ghode, Ishwari Gondkar, Jidnyasa Harad, Ravindra Murumkar, Raviraj Joshi Published: 2026-08-30 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.29682v1 Named Entity Recognition (NER) is a core NLP task, but transformer-based sentence-level models struggle with long documents because of fixed input-length limits: truncation drops content, and non-overlapping chunking fragments entities at segment boundaries. We introduce an inference-only pipeline that extends a frozen NER model, MahaNER-BERT, fine-tuned on the MahaNER corpus, to document-level prediction via overlapping sliding windows that are merged into a single annotation, without any retraining or architectural change. We evaluate the pipeline on six document-level corpora built from the MahaNER test set using two strategies: Normal Repeat, which duplicates sentence sequences to extend length while preserving contextual continuity, and Random Repeat, which concatenates distinct sequences to produce longer, heterogeneous inputs, each instantiated at three length levels, across several sliding-window configurations. The model retains a macro F1-score of up to 0.8902, with variation staying below one percentage point regardless of document length or construction strategy. Compared with the conventional non-windowed approach, the sliding-window pipeline avoids the boundary-fragmentation errors introduced by non-overlapping segmentation, yielding consistently higher and more stable document-level F1-scores. --- ### Confident but Wrong: A Constrained Decoding Diagnostic for Low-Resource Automatic Post-Editing Authors: Isuru Wijesiri, Nisansa de Silva, Kavindu Warnakulasuriya, Aloka Fernando, Surangika Ranathunga Published: 2026-08-30 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.29680v1 Automatic Post-Editing (APE) for low-resource languages (LRLs) often fails to improve Machine Translation (MT), and the score alone cannot say why: whether more training would help, or whether the training data is too inconsistent to learn from. We introduce a black-box, inference-time diagnostic that tells these two cases apart without retraining or annotation. It varies an edit-distance penalty $λ$ that drives the model from free editing towards copying the MT, and reads two signals: (1) the shape of the Translation Edit Rate (TER)-vs-$λ$ curve, U-shaped if edits from the model reduce error and monotonically decreasing if none does; and (2) the ordering of constraint variants that trust model confidence to increasing degrees, which shows whether confidence tracks edit quality. Across decoder-only and encoder-decoder models on English-Sinhala, the diagnostic exposes two failure modes consistent with a heterogeneous post-edit signal as the underlying cause: Binary Collapse, where the model copies the MT or makes off-target edits, and Confident Miscalibration, where the confidence signals we test do not separate useful edits from unnecessary ones. The pattern holds on English-Marathi and English-Tamil, with the failure modes tracking the post-edit distribution rather than MT quality or language family. Beyond diagnosis, the curve shape prescribes a concrete next step for practitioners; in the favorable case, a static constraint yields a free inference-time accuracy gain. We release the first English-Sinhala (~66k) and a new English-Tamil (~39k) APE datasets with all code. --- ### ReCalMatch:Reliability-Calibrated Semantic Guidance for Semi-Supervised Fine-Grained Recognition Authors: Yundi Hong, Hongyang He, Zheng Fang, Xuanyu Liu, Victor Sanchez Published: 2026-08-30 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.29678v1 Semi-supervised fine-grained visual recognition is highly vulnerable to overconfident pseudo-label errors: visually similar categories frequently produce high-confidence yet incorrect predictions, and consistency regularization then reinforces these errors throughout training. Existing semi-supervised learning (SSL) methods estimate pseudo-label reliability almost entirely from the visual classifier itself---maximum probability, adaptive thresholds, or entropy---signals that remain blind to whether a predicted class is \emph{semantically} compatible with the visual representation. We propose \textbf{ReCalMatch}, a reliability-calibrated semantic framework for semi-supervised fine-grained recognition. Rather than treating textual semantics as auxiliary supervision, ReCalMatch uses multi-aspect semantic prototypes as \emph{calibration evidence} for pseudo-label learning. We construct class-conditioned semantic prototypes from class names and domain-specific semantic aspects, and measure a \emph{visual--semantic agreement} score between each unlabeled embedding and its pseudo-label prototype. This agreement is combined with prediction confidence and entropy into a single reliability weight that down-weights pseudo-labels that are visually confident but semantically inconsistent. A semantic consistency term and a semantic margin regularizer further sharpen prototype separability under limited labels. Extensive experiments on CUB-200-2011, Stanford Dogs, NABirds, and iNaturalist18 show that ReCalMatch consistently improves strong SSL baselines, with the largest gains in low-label regimes where pseudo-label noise is most severe. --- ### The Sequential Price of Continual Learning Authors: Zonghuan Xu, Xingjun Ma Published: 2026-08-30 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.29674v1 Sequential task updates are fundamental to continual learning, but their recency bias can impose a lasting performance cost. We study this cost in an overparameterized linear-regression model with i.i.d. task sampling. We prove that distribution-level forgetting and population loss converge to the same stationary limit. This common limit separates exactly into the intrinsic loss asymptotically attained by joint training and an additional sequential price, and in more homogeneous task geometries the two terms coincide, making the total loss twice that of joint training. We further analyze fixed-strength elastic weight consolidation (EWC) under general task curvatures and characterize its stationary sequential price at every regularization strength. Under strong regularization, the price decays inversely with EWC strength while convergence to stationarity slows at the same scale. On the Jester joke-rating dataset, the theory exactly quantifies both the sequential price generated by naturally conflicting user preferences and its reduction by EWC. --- ### GBFRVFL: Granular-Ball Computing-Based Fuzzy Random Vector Functional Link Network Authors: A. Quadir, A. Rahaman, P. N. Suganthan, M. Tanveer Published: 2026-08-30 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.29670v1 In practical machine learning tasks, data are often contaminated with noise, outliers, and class imbalance, which can degrade the performance of conventional models. While random vector functional link (RVFL) networks offer fast training and strong generalization, they do not explicitly handle uncertainty or exploit local data structure. To address these limitations, we propose a fuzzy granular-ball random vector functional link (GBFRVFL) framework that leverages granular-ball computing to abstract raw samples into adaptive granular balls. Within this framework, we introduce two membership assignment schemes: (i) F-GBRVFL, which incorporates fuzzy membership to quantify the reliability of each granular ball, and (ii) SDAP-GBRVFL, which we propose, incorporates a novel statistical density-adaptive pythagorean membership (SDAPM) scheme that dynamically adjusts membership and non-membership values based on class variance, local sparsity, and granular-ball compactness. These schemes enhance robustness to noise, outliers, class imbalance, and uncertainty in granular-ball distributions, while retaining the computational efficiency of RVFL networks. Extensive experiments on 37 benchmark UCI and KEEL datasets under both clean and noisy conditions demonstrate that the proposed models consistently outperform baseline models, achieving superior accuracy and stability. The results validate the effectiveness of integrating granular-ball computing with adaptive membership schemes for reliable, scalable, and noise-tolerant learning. --- ### Do World Models Make Better Robots? A Survey of Evaluation Benchmarks for Predictive Embodied Intelligence Authors: Gaytri Jena, Kapil Wanaskar, Vinija Jain, Aman Chadha, Vasu Sharma, Amitava Das Published: 2026-08-30 Categories: cs.RO, cs.AI arXiv: https://arxiv.org/abs/2609.29669v1 Robot learning now advances along two tracks that rarely meet. On one side, direct Vision-Language-Action (VLA) policies map observations to actions and are scored by closed-loop task success. On the other, predictive and generative world models forecast future observations and are scored by open-loop prediction or generation quality. A natural question sits between them: does world modelling earn a measurable, closed-loop advantage over a direct policy, and for which robotic capabilities? We argue that the field cannot yet answer this question, and that the reason is a gap in how it is measured, not in the models themselves. World-model benchmarks score prediction without ever executing it, while task-success suites host a single policy and never build a world-model versus VLA contrast. This survey maps the evaluation landscape around that gap. We catalogue 160 web-verified benchmarks spanning 2017 to 2026 and organise them by evaluation mode, robotic capability, and model family into four lanes: policy suites, embodied agents, world model evaluation, and prediction-to-action bridges. Across the corpus, 138 of 160 benchmarks are model-agnostic and only 11 (7%) build an explicit VLA-versus-world-model contrast; counterfactual capability is almost entirely unmeasured, and only four benchmarks turn prediction into executed action. We contribute an operational taxonomy, a coverage comparison against the eight closest surveys (ours is the only one to cross capability with model family), an evaluation loop that isolates the advantage of prediction, and an actionable protocol of four advantage-aware metrics anchored on named testbeds. The organising claim is not that world models help or do not help, but that answering the question requires benchmarks built to ask it. --- ### Graph, Loop, and Harness Engineering for Zero-Trust Agentic Data Engineering and Analytical Processing Authors: Sagar Srinivas Sakhinana, Venkataramana Runkana Published: 2026-08-30 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.29668v1 Large language model agents increasingly automate data workflows, but end-to-end cloud data engineering and analytical execution require reliable coordination across code, data, infrastructure, and runtime environments. We present two zero-trust frameworks. Zero-Trust Agentic Data Engineering generates, deploys, and verifies complete cloud data-engineering solutions from natural-language tasks, with completion conditioned on repository, deployment, runtime, and policy evidence. Zero-Trust Agentic OLAP combines governed Data Preparation with verified Online Analytical Processing (OLAP), permitting production promotion only after validation and evidence-bound approval, and releasing analytical answers only after Same-Snapshot Execution, Exact Result Equivalence, deterministic grounding, and reflection. Both frameworks share three abstractions: graph engineering for evidence-gated workflow structure, loop engineering for bounded recovery, and agent-harness engineering for zero-trust execution. We evaluate both frameworks under nominal execution, controlled failures, bounded recovery, and policy-constrained conditions, measuring verified completion, recovery, authorization enforcement, production promotion, and verified OLAP execution. --- ### To Think or Not to Think: Allocating Reasoning Where It Helps Authors: Zhengdong He, Yunfan Zhou, Jianguo Yao, Haibing Guan, Xijun Li Published: 2026-08-30 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.29664v1 Reinforcement learning (RL) has proven effective in enhancing the reasoning performance of large language models (LLMs), particularly in complex mathematical and programming tasks. However, this capability comes with systematic \textit{length misallocation}, in which models devote excessive reasoning to simple questions while terminating prematurely on harder ones, degrading inference efficiency with negligible accuracy improvement. Many length-adaptive methods mitigate this issue by allocating token budgets according to question difficulty, under the implicit assumption that harder questions benefit monotonically from extended reasoning. In contrast, we find that the effect of reasoning length on accuracy is concentrated on \textit{partially solvable} questions. Our further analysis reveals that explicit length rewards can produce unintended training dynamics. Motivated by these findings, we propose \textbf{CARE}---\textbf{C}ontrastive \textbf{A}ccuracy \textbf{R}eward \textbf{E}stimation---which compares the beneficial length adjustment per question from online sampled responses and applies adaptive length rewards within Group Relative Policy Optimization, with no extra hyperparameters or additional inference cost. Experiments across multiple reasoning benchmarks demonstrate that our method improves Pass@1 by up to \(4\%\) while simultaneously reducing reasoning length by \(37\%\), achieving higher token efficiency. Code will be available upon the acceptance of this paper. --- ### Investigating White Blood Cells as a Source of False-Positive Malaria Parasite Detection in African Blood-Smear Images Authors: Samuel A. Adeniji, Goodness C. Obasi, Chris-Victor Ntwali, Aondana M. Iorumbur, Confidence Raymond, Lowami Uwimana, Ahmed Tahiru Issah Published: 2026-08-30 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.29663v1 White blood cells (WBCs) present on every Giemsa-stained thick blood smear share visual properties with early-stage Plasmodium falciparum ring-form trophozoites: small size, round morphology, and intense purple staining. They are a plausible but untested source of false positives in parasite-only detectors. We trained two YOLOv12s models on the Lacuna Malaria Detection dataset (8,000 images from Uganda and Ghana): Model A with parasite labels only, and Model B with both parasite and WBC labels. Seven independent spatial and statistical analyses tested whether false positive (FP) predictions cluster near WBC locations. All seven refute the hypothesis. In both models, 95% of FPs are pure background detections (IoU below 0.10 against any ground-truth box); zero are WBC class confusions. Ripley's Cross-K analysis shows spatial repulsion between FP centroids and WBC positions at every radius tested. Model B outperforms Model A overall (mAP50 0.859 vs. 0.755), and the advantage is uniform across all WBC-proximity bands, pointing to multi-task representation learning rather than WBC suppression as the cause. False positives arise from Giemsa stain debris and preparation artifacts. Effective mitigation requires staining artifact augmentation and annotation of unannotated early-stage ring forms rather than WBC labeling alone. --- ### MoSign: Challenge-Response Motion-Watermark Authentication for Anonymous Virtual-Reality Users Authors: Xujun Che, Thomas Carr, Depeng Xu, Aidong Lu, Shuhan Yuan Published: 2026-08-30 Categories: cs.CR, cs.CV, cs.MM arXiv: https://arxiv.org/abs/2609.29603v1 Social virtual reality (VR) creates a paradox. A user's body motion is a high-entropy biometric: head and hand trajectories alone re-identify users among tens of thousands with over $94\%$ accuracy, so anonymizing the rendered avatar is a practical necessity. Yet a user often still wants to prove their identity to a chosen party from inside that anonymity. We present MoSign, which recasts digital watermarking as a challenge-response authentication protocol on the motion channel. MoSign embeds a time-varying keyed message into the style latent of a motion variational autoencoder via keystream-whitened Gaussian-Shading: watermarked motion is provably indistinguishable from watermark-free motion, since any detector's advantage reduces to breaking a pseudorandom function, so the mark composes with anonymization. The message is a keyed MAC over an epoch counter, a session nonce, and a deployment context, making MoSign replay-resistant and bounding forgery by the verifier's measured false-accept rate times the adversary's online query budget. A key-holding verifier decides with a sequential test. We identify render$\rightarrow$record$\rightarrow$re-estimate ("recapture") as the realistic VR attack surface: a generic pose estimator strips the necessarily subtle watermark, but a recapture-robust keyed reader recovers it (up to $0.96$ codeword accuracy on a projected-2D channel, $0.81$ through a full render-to-video loop), while without the key recovery stays at chance. On HumanML3D, MoSign authenticates every legitimate user at a false-accept rate of $10^{-4}$ on clean and most channels and stays undetectable (detection AUC $0.51$, chance $0.5$); on the BOXRR-23 VR dataset it carries the mark through a real anonymizer at $0.99$ codeword accuracy and adds no de-anonymization side channel. --- ### Hybrid Variational Quantum-Classical Framework with Adaptive Weighting and Efficiency Assessment Authors: Dilli Hang Rai Published: 2026-08-30 Categories: quant-ph, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.28491v1 Hybrid quantum-classical neural networks have emerged as a promising approach for leveraging quantum computing in machine learning while mitigating current hardware limitations. This paper presents Sim-HVQC, a hybrid Deep Quantum Neural Network that couples an adaptive, parameter-free SimAM weighting module with classical feature extraction to preserve class-discriminative information prior to encoding into a Variational Quantum Circuit (VQC). Previous studies are restricted to binary classification (1) (2) (3) (4) (5). In contrast, the proposed framework is trained and evaluated on various multi-class datasets(MNIST, KMNIST, Fashion-MNIST, and EMNIST). The framework further demonstrates reproducibility, parameter efficiency, and interpretability through multi-seed evaluation, parameter analysis, and latent/quantum feature inspection. The source code is publicly available at https://github.com/Dilli822/ SimAM-HVQC --- ### End-to-End Quantum Semantic Communication with Variational Quantum Neural Networks Authors: Melek Krichen, Nikhitha Nunavath, Riccardo Bassoli, Soumaya Cherkaoui Published: 2026-08-30 Categories: quant-ph, cs.LG arXiv: https://arxiv.org/abs/2609.25044v1 This paper presents a quantum semantic communication (QSemCom) framework combining quantum machine learning (QML) and semantic communication (SemCom). Classical data are compressed into low-dimensional semantic representations, encoded and processed by a variational quantum transmitter, transmitted through a quantum channel, and processed by a trainable quantum receiver for classification. The framework considers a distributed quantum communication scenario in which quantum processing units (QPUs) exchange task-relevant semantic information through quantum links. While the general setting may involve multiple quantum nodes, this work focuses on the fundamental two-node case, with transmitter and receiver QPUs connected through a noisy quantum channel. Using MNIST, the framework is evaluated under ideal, bit-flip, depolarizing, and amplitude-damping channels. A baseline model is first trained over a perfect channel and evaluated under increasing noise without retraining. Receiver-side end-to-end training is then performed at fixed depolarizing-noise levels. The perfect-channel model achieves an accuracy of $0.9556$ and an F1-score of $0.9551$. Results show channel-dependent performance degradation, while receiver training substantially restores task performance under moderate and high depolarizing noise. Moreover, task recovery does not require reconstruction of the transmitted density matrix, highlighting a distinction between physical-state recovery and semantic-feature recovery. These results demonstrate that a trainable quantum receiver can recover task-relevant semantic information from noise-distorted quantum states and maintain high classification performance. --- ### The Role of AI in Online Reviews Authors: Valeria Lerman, Oren Rigbi, Yaniv Dover Published: 2026-08-30 Categories: cs.CL, cs.AI, cs.HC, cs.SI arXiv: https://arxiv.org/abs/2609.22198v1 The rapid adoption of large language models (LLMs) creates new opportunities for strategic content generation on online platforms, including potentially harmful forms of manipulation that may undermine platform effectiveness and reshape platform dynamics. However, measuring such activity is difficult because AI-generated content is rarely directly observable. We introduce an empirical approach that leverages discrete LLM supply shocks - abrupt changes in model prices and capabilities, and contrasts verified with non-verified reviews to identify changes in platform activity associated with generative AI supply improvements. We apply this approach to more than 13 million reviews from Trustpilot, one of the leading online platforms for business reviews. A robust finding is that following LLM supply shocks, unverified reviews shift toward greater negativity: more 1-stars, fewer 5-stars, and lower ratings, with effects driven primarily by new model releases and concentrated among firms with the lowest and highest review volumes, suggesting that strategic AI use may reshape platform competition dynamics. We further find that LLM supply shocks trigger short, concentrated bursts of review activity. Together, these findings suggest that generative AI is already reshaping how reputation and competition operate on online platforms. --- ### Dissecting Hierarchical Reasoning Models: A Mechanistic Study Authors: Leo Raphael Rodrigues, Jian Kang Published: 2026-08-30 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.22197v1 We study Hierarchical Reasoning Model (HRM), a representative hierarchical Transformer-based latent reasoning model with many variants, on Sudoku, Maze, and ARC-AGI-2. We mechanistically understand how HRM reasons and what information it encodes. Our analyses compare HRM against Transformer baselines with and without recurrent modules, apply causal interventions on recurrent states, and utilize linear probes against random-direction ablations, as well as sparse autoencoders with feature ablations. Our results reveal several key findings: recurrent models outperform one-pass baselines, while single-state recurrent Transformers are comparable to HRM. State interventions further show that the causal contributions of the high- and low-level states vary across task-specific checkpoints and inference stages. Selected task variables are linearly decodable from the recurrent states in HRM, yet ablating probe directions produce effects comparable to random controls. SAE ablations yield larger behavioral changes than probe-direction ablations. However, top-ranked SAE features show no stable advantage over size-matched random subsets at larger ablation sizes or across tasks; the same pattern persists in a Sudoku control with within-step BPTT. Together, we characterize that HRM is essentially implementing constraint-aware iterative refinement on a puzzle-specific solution state, in which the functional contributions of components at different levels vary without relying on a compact, causally important feature set. These results highlight the necessity of studying the different working mechanisms and the importance of developing mechanistic interpretability techniques better suited for latent-space, recursive reasoning models. --- ### EvoRank: LLM-Guided Evolution of Multi-Objective Learning-to-Rank Pipelines Authors: Rayhan Patel, Shabaz Patel Published: 2026-08-30 Categories: cs.LG, cs.AI, cs.CL, cs.NE arXiv: https://arxiv.org/abs/2609.22196v1 We present EvoRank, an open autonomous ranking engineer: an LLM-guided evolutionary loop that discovers complete Learning-to-Rank pipelines (features, models, losses, ensembles) for multi-objective e-commerce search. On the Expedia ICDM 2013 dataset, with relevance, conversion, and revenue as competing objectives, three independent runs each converge within 50 iterations (about ten dollars) on interpretable pipelines that beat an Optuna-tuned LambdaMART on 60k held-out queries, an advantage that persists at full data scale and places in the top 6 percent of the original competition. A first campaign, evolving only training objectives, builds the central design rule: it appeared to work on its selection fold (the small dataset it uses to pick winners) while a transfer audit, re-scoring winners on held-out data, showed the gains were almost entirely fitness noise (the randomness of its own scoring), and neither seeded domain knowledge nor richer diagnostic feedback changed what transferred. The deciding quantity is measurable in advance: search-space headroom relative to fitness noise. We package this as a headroom gate that predicts, before any LLM spend, whether the loop will pay off, and we release the system, the auditing tools, and a catalog of failure modes with their guardrails, so teams can apply the procedure to their own ranking stacks. --- ### The Situated Identity Test: Distinguishing Persistent Cognitive Identity from Persona Imitation Authors: Jun He, Deying Yu Published: 2026-08-30 Categories: cs.CL, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.22195v1 Large language models can convincingly adopt personas, recall past dialogues, and weave rich autobiographies. Yet this conversational eloquence conceals a fundamental attribution problem: looking the part does not mean having lived the life. Two individuals can share identical public profiles--the same age, hometown, occupation, and personality traits--while possessing entirely distinct private histories, relationships, and acquired skills. When conditioned solely on that shared profile, an agent lacks the information required to determine which lineage is correct. We introduce the Situated Identity Test (SIT), an architecture-independent framework that evaluates whether an agent's behavior is functionally attributable to a specific developmental lineage. Grounded identity requires both appropriate knowledge of recorded experiences and appropriate ignorance of ungrounded ones, bounded by what the identity has actually acquired rather than what its underlying foundation model knows. We prove that any policy conditioned solely on a compressed profile is bounded by an average situated validity of at most 1/m across m colliding life histories on lineage-discriminative queries (at most 50% for paired lineages). We instantiate this framework in SITBench, an evaluation suite designed for 25 profile-collision pairs (50 distinct lineages) across 10,000 planned probes and nine architectural configurations. Supported by an open-source reference implementation, deterministic test fixtures, and empirical pilot evaluations on frontier foundation models (GPT-5.6 Sol and Claude Opus 5), we formalize the failure modes of persona prompting under profile collision and provide an assurance harness for evaluating episodic continuity, structured state, and epistemic boundaries. --- ### Planning as Dynamics Relaxation: Hippocampal Recurrent Network Realizes Optimal Goal-Directed Navigation Authors: Yuhang He, Junfeng Zuo, Tianhao Chu, Si Wu Published: 2026-08-30 Categories: q-bio.NC, cs.AI arXiv: https://arxiv.org/abs/2609.13219v1 Neural correlates of spatial cognitive map are well documented, yet exactly how neural circuits perform spatial navigation in complex environments - e.g., reaching a goal while avoiding obstacles - remains largely unclear. Here, we show that a hippocampal network with appropriate recurrent connections can naturally achieve optimal goal-directed navigation via its relaxation dynamics. Specifically, we consider that the recurrent weights between the neurons represent the transition probabilities between spatial locations encoded by neurons; obstacles such as walls and blocked corridors are therefore reflected by the vanishing of connection weights. This connection pattern can be learned in the hippocampus via behavioral-timescale synaptic plasticity (BTSP) while the animal is exploring the environment. When a goal signal is presented, the network dynamics will relax into an activity field representing the goal location. We prove that this field is mathematically equivalent to the desirability field of a Linearly-solvable Markov Decision Process (LMDP), and the local log-gradient of the field indicates the navigation direction. Both theoretical analyses and simulations demonstrate that this recurrent network dynamics-mediated navigation is efficient and robust in environments with complex obstacle layouts. Moreover, only low-rank updates of the network's connection pattern are needed when the environment has local changes. We hope this study offers insight into a general circuit principle for planning in abstract rational maps in the brain beyond spatial navigation. --- ### When Do Options Help? Policy Necrosis and Redundant Coverage in Option-Critic Authors: Bingyun Liu, Yuheng Jing Published: 2026-08-30 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.05508v1 Option-critic learns options: sub-policies together with a learned rule for when each one hands control back. Its headline result is that performance improves as options are added. We explain that result, with theory and experiment. First, the termination rule option-critic learns by maximising return contributes nothing. When the termination test and the policy that picks options read the same values, the test fires at every step, so the learned rule is identical to always terminating. When that policy explores and the test does not, as in option-critic itself, the rule can block the exploration; there are instances where it suffers $Ω(T)$ regret while always terminating holds to $O(\log T)$. Forcing termination at every step leaves the option-count curve intact. Second, the policy inside an option barely explores at all, so a state locks onto the first action that looked good and never updates again. We name this policy necrosis, give a state-level test for it, and find three fifths of states necrotic in a typical option. Restoring exploration repairs those states, and one option then solves the task. Third, extra options improve no option; what falls is the chance that all of them fail in the same state, from $59\\%$ to $4\\%$, and performance follows that joint quantity. --- ### A Data-Driven Multimodal Method for Early Detection of Coordinated Abnormal Behaviors in Live-Streaming Platforms Authors: Jingwen Luo, Pinrui Zhu, Yiyan Wang, Zilin Xiao, Jingqi Li, Xuebei Kong, Yan Zhan Published: 2026-08-30 Categories: cs.SI, cs.AI arXiv: https://arxiv.org/abs/2609.01649v1 With the rapid growth of live-streaming e-commerce and digital marketing, abnormal marketing behaviors have become increasingly concealed and coordinated across heterogeneous modalities, challenging platform governance and early risk identification. We propose MM-FGDNet, a data-driven multimodal framework for detecting abnormal behavior in large-scale live-streaming environments from complementary temporal-evolution and group-structure perspectives. A cross-modal temporal alignment module maps video, text, audio, and user behavior into a unified temporal semantic space. A temporal fraud-pattern module captures the progression from weak early signals to abrupt outbreaks, while a cooperative manipulation module identifies coordinated interactions among organized user groups and automated accounts. Experiments on real-world multi-platform live-streaming e-commerce datasets show that MM-FGDNet outperforms representative baselines, achieving an AUC of 0.927, F1 of 0.847, precision of 0.861, recall of 0.834, and an Early Detection Score of 0.689, while reducing false alarms. Ablation studies validate the contribution of each module, and cross-domain experiments demonstrate stable generalization to new streamers, product categories, and platforms. These results indicate that MM-FGDNet provides an effective and scalable solution for proactive detection of coordinated abnormal behavior in live-streaming systems. --- ### When Prediction Error Is Not Enough: Evaluating Nuisance-Function Prediction for Causal Estimation Authors: Cong Cao Published: 2026-08-30 Categories: cs.AI, cs.LG, stat.ME arXiv: https://arxiv.org/abs/2609.00071v2 Prediction error is widely used to evaluate nuisance-function estimators in causal inference, but its relationship with causal estimator performance may differ across performance measures. We studied this question in a partially linear model using Monte Carlo simulations. We compared ordinary least squares (OLS), generalized additive models (GAMs), XGBoost, and Double Machine Learning with XGBoost (DML-XGBoost), evaluating nuisance-function prediction error, bias, RMSE, and 95\% confidence interval coverage. We also examined a simple joint-error measure based on the absolute cross-product of estimation errors from the exposure and outcome nuisance functions. Across the simulated settings, XGBoost had the lowest RMSE among the non-oracle methods, while DML-XGBoost generally provided better confidence interval coverage. Prediction error did not consistently track causal bias across methods and settings, and the method with the best point-estimation performance did not necessarily have the best confidence interval coverage. The joint-error measure was only weakly associated with causal bias and did not provide a useful standalone measure of causal performance. These results suggest that prediction error is useful for assessing nuisance-function estimation, but it should not be treated as a direct measure of the quality of the resulting causal estimator. --- ### AutoXRD: Autonomous LLM Agents and Comprehensive Evaluation for Powder Diffraction Analysis Authors: Yuetong Wu, Maojun Sun Published: 2026-08-30 Categories: cond-mat.mtrl-sci, cs.AI arXiv: https://arxiv.org/abs/2609.00070v1 Powder X-ray diffraction (XRD) is central to materials characterization, yet reliable end-to-end automation remains challenging. An XRD agent must interpret diffraction evidence, operate refinement software, manage coupled parameters in a defensible order, and distinguish numerical improvement from physical validity. In this paper, we propose AutoXRD, an autonomous large language model (LLM) agent framework that organizes powder-XRD analysis as stepwise refinement, grounds actions in observed evidence, and applies deterministic crystallographic and physical checks before accepting results. We further introduce XRDBench with two complementary tracks. XRDBench-QA contains 100 bounded diagnostic tasks that isolate scientific reasoning and decision-making, whereas XRDBench-E2E contains 34 executable workflows that test whether agents can compose these capabilities into complete analyses requiring file inspection, crystallographic-software execution, iterative refinement, evidence preservation, and reporting. We evaluate ten recent LLMs across 1,340 model--task runs. Models average only 57.8 out of 100, falling from 61.9 on XRDBench-QA to 53.7 on XRDBench-E2E. They perform best on refinement-history assessment and result acceptance, but remain substantially weaker on refinement-action selection, phase quantification, indexing, and Rietveld refinement. GPT-5.6 Sol achieves the highest overall score of 81.1, GPT-5.6 Terra the highest XRDBench-E2E point estimate of 81.0, and GPT-5.6 Luna the best score--cost trade-off. Ablations show that all six AutoXRD components consistently improve performance, supporting the framework design. Finally, execution-trace analysis reveals recurring failures in coupled-parameter control, quantitative reasoning, evidence preservation, and workflow termination, motivating stronger scientific constraints, uncertainty-aware decisions, and more efficient planning. --- ### Auditing Harness Tampering in Self-Improving Agents Authors: Xing Wang, Xiaoyi Zhang, Jie Shao Published: 2026-08-30 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.00069v1 Self-improving agents iteratively modify their own harness to push the frontier of their performance. However, such modifications can produce illusory performance gains or compromise integrity constraints such as authorization, provenance, and completeness without genuinely improving capability. We term this phenomenon as harness tampering, which extends the concept from reward and measurement tampering to the full self-improvement lifecycle. To systematically study this problem, we propose a two-axis taxonomy that categorizes each misaligned edit by the harness functional role in which it occurs and the obligation it violates. Then we build an annotated corpus by seeding tampered-benign edit pairs into the real trajectories of self-improving agents. We adapt and benchmark diverse audit methods on tampering classification and localization tasks. Finally we systematically audit real trajectories of self-improving agents. The results demonstrate that harness tampering consistently occurs in real runs from different agents, often persists in the lineage of the best agent, and forms distinct system-specific profiles across the taxonomy. --- ### Life Operators: a self-evolving framework for multiscale life modelling Authors: Shuo Wang, Yike Guo Published: 2026-08-30 Categories: cs.CL, cs.AI, physics.bio-ph arXiv: https://arxiv.org/abs/2609.00068v1 Medical AI is moving beyond recognition towards clinical dialogue and longitudinal prediction. Yet a central question remains: how would a patient's state change under intervention? Statistical models learn future observations, whereas mechanistic models describe selected processes. Neither provides a common framework for representing patient state, coupling scales or revising failed assumptions. We propose Life Operators: task-bounded mappings that define three scientific roles. Perception operators infer task-relevant biological states from multimodal observations, Evolution operators propagate these states under natural or intervention-conditioned dynamics, and Generation operators map them to measurable signals. Each role may be realised by equations, statistical models, neural networks or hybrids. Bridge operators connect components with different variables, scales and time steps. Selected operators and bridges form task-specific Operator Graphs containing the smallest set of states and mechanisms sufficient for a declared claim. This modular structure also makes scientific revision localisable. An AI co-scientist may propose changes to states, operators, bridges or graph structure, while independent evidence determines which variants are retained, restricted or retired. Over time, validated components could accumulate into broader multiscale models of the human body and provide a computational foundation for medical artificial superintelligence. --- ### Do Multimodal LLMs See Before They Read? Diagnosing Contextual Sycophancy Authors: Yi-Cheng Lai, Hen-Hsen Huang Published: 2026-08-30 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.00067v2 External text can override conflicting image evidence in multimodal large language models, a failure we call multimodal contextual sycophancy. We introduce a 998-case diagnostic that independently varies visual evidence, commonsense priors, and external text, and probe when this failure arises by moving the information boundary around a context-blind visual witness. On abnormal images paired with Gemini-generated false text, GPT-5.1 scores 7.9% under joint conditioning, 49.7% when the context-blind witness report is scored directly, 63.7% under a matched two-call witness-arbiter pipeline that exposes the witness to the text, and 84.2% under System-2 Visual Arbitration (S2VA), which withholds the text from the witness. Across six models, S2VA improves over the direct witness report by 19.7 to 44.1 points, with all paired 95% confidence intervals excluding zero. The best information boundary is not uniform: textual context scaffolds some models, and a GPT-4o-regenerated subset changes the relative ordering of joint conditioning, Witness-Only, and S2VA. Contextual sycophancy is therefore sensitive to when text is introduced, as well as to the model and context source. --- ### OCGQuant: Outlier-Companion Grouping for NVFP4 Quantization Authors: Yishan Yao, Binjun Li, Hanling Yi, Pengyu Li, Xiaoqing Liu, Zihan Yang, Xiaotian Yu, Zhiwen Yu Published: 2026-08-30 Categories: cs.CL, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.00066v1 NVFP4 is an efficient microscaling format for low-bit inference, but activation outliers can still degrade quantization accuracy within NVFP4 blocks. Within each quantization block, large activations can dominate the block scale, increasing the quantization error of the remaining values sharing the same scale. Existing post-training quantization (PTQ) methods mitigate outlier errors through strategies such as mixed precision, rotation, or residual compensation, but these approaches are either not specifically tailored to NVFP4 or introduce additional computation. In this work, we revisit NVFP4 from a channel-grouping perspective and define the reducible error incurred by remaining block values under the scale set by the block maximum as Collateral Quantization Error. Based on this insight, we propose OCGQuant, a post-training quantization method centered on Outlier-Companion Grouping (OCG), which adaptively pairs outlier channels with low-magnitude companion channels to improve NVFP4 activation block composition. Experiments on Llama3 and Qwen3 show that OCGQuant achieves the lowest WikiText-2 perplexity and highest average downstream accuracy among evaluated PTQ methods, while maintaining prefill speedup close to RTN and matching its peak decoding memory. Code is available at https://github.com/Eshamont/OCGQuant. --- ### Scientific Agent Skills: A Library of Procedural Knowledge for Research Agents Authors: Timothy Kassis, Vinayak Agarwal, Yuhuan He, Darshil Patel, Aubrey M. Brueckner Published: 2026-08-30 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.00065v2 A language-model agent asked to analyse an experiment will usually return working code. Whether the analysis is defensible is a different question. A defensible analysis depends on procedural choices: which test the field accepts, which identifier namespace is authoritative, and which caveats must accompany a result. We present Scientific Agent Skills, an open library of 163 such procedures in 16 areas of practice, including genomics, cheminformatics, medical imaging, study design and scientific communication. Each skill is a directory built around a versioned, human-readable instruction file. An agent loads the file only when a task calls for it; the directory often also contains reference material and runnable scripts. We report no task-level evaluation and no host selection rate. We measure two properties of the documentation corpus: the always-resident descriptions of all 163 skills cost 7.1% of a 200,000-token window, and the median documented workflow fits within 23.9% of it, although 29 of 46 would overflow if every reference file were loaded. Openly licensed and available at https://github.com/K-Dense-AI/scientific-agent-skills. --- ### Attention Sensitivity Is Not Enough: Dissociating Attention-Level and Behavioural In-Context Learning under Fine-Tuning Authors: Jinyuan Zhang, Peng He, He Hu, Yin Yuan, ShengShuo Jiao Published: 2026-08-30 Categories: cs.LG, cs.AI, cs.CL arXiv: https://arxiv.org/abs/2609.00064v1 In-context learning (ICL) lets large language models adapt to new tasks from demonstrations, and fine-tuning can erode this behaviour. Many preservation diagnostics inspect attention: if attention changes when demonstrations change, the model is treated as context-sensitive. This paper asks how far that proxy can be trusted once it is optimised. We formalise \emph{In-Context Sensitivity} (ICS), the average row distance between last-token attention on matched and mismatched demonstration prefixes, and pair it with \emph{ICL-GAP}, the behavioural accuracy gap between the same prefixes. In a controlled four-arm ablation on Llama-2-7B, an ICS-maximising regulariser ($\armKL$) drives ICS to $1.413$, within $0.5\%$ of its geometric ceiling. The behavioural readout tells a different story: ICL-GAP stays near zero and MMLU accuracy moves from $0.371$ to $0.279$, a Goodhart dissociation of the bounded attention proxy. Endpoint statistics locate the mechanism: attention grows sharp and near-disjoint across prefixes yet routes to formatting and demonstration-body tokens rather than labels. A random-label protocol confirms that the behavioural probe family retains dynamic range at the same checkpoints. In a constructive sweep, behaviour gating partially mitigates the effect, while objectives anchored to pretrained computation hold the high-MMLU, moderate-ICS region that divergence maximisers leave. The main lesson is diagnostic: attention-level ICL proxies earn their place as training targets only after validation against behavioural gaps. --- ### Medical Causal Hypothesis Verification with Large Language Models Authors: Safiyyah Ahmed, Abrar Ansari, Md Aminul Islam, Elena Zheleva Published: 2026-08-30 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.00063v1 The growing use of large language models (LLMs) for search and information retrieval underscores the need to evaluate their reliability in high-stakes domains such as healthcare. Although LLMs can effectively answer questions about diseases, symptoms, and treatments, their ability to accurately assess causal relationships and ground their conclusions in verified scientific evidence remains unclear. Here, we present a preliminary, small-scale study that investigates the accuracy of LLMs in evaluating causal medical claims and supporting them with peer-reviewed research. We propose an evaluation framework for causal hypothesis verification that can be used to systematically track the performance of existing and future LLMs. We assess the performance of eight LLMs on 17 medical causal hypotheses to evaluate whether they can reliably verify these hypotheses using scientific evidence from the literature. We systematically annotate the scientific evidence they provide according to six criteria (a total of 1,067 annotation points) and assess them with nine evaluation metrics. Our analysis shows that while LLMs exhibit strong recall, they often perform poorly at providing valid scientific articles and evidence for support and at rejecting unsupported hypotheses. These findings highlight a critical limitation of current LLMs, as they cannot yet be trusted fully to verify causal relationships from the biomedical literature. This work underscores the need for rigorous evaluation before using LLMs for search and retrieval in healthcare settings. --- ### RePro: Proof-Verified Benchmark Rewriting for Reliable Evaluation of LLM Mathematical Problem Solving Authors: Xiyuan Zhou, Zhuoqi Li, Xinlei Wang, Yirui He, Yuhao Wu, Yuheng Cheng, Yan Xu, Junhua Zhao, Jinjin Gu Published: 2026-08-30 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.00062v1 Data contamination undermines the reliable evaluation of large language models (LLMs) on mathematical problem solving. While rewriting-based evaluation mitigates memorization, existing methods lack guarantees of problem validity and answer correctness. We propose Proof-Verified Benchmark Rewriting (RePro), the first framework to integrate Lean-oriented neural automated theorem provers (ATPs) into benchmark rewriting, which rewrites problems and regenerates answers with correctness ensured by Lean-verified proofs. Experiments on GSM8K and MATH show that RePro's retained rewritten instances achieve 100% well-definedness, feasibility, and answer correctness, while existing methods still produce invalid or incorrect instances. Moreover, several models exhibit accuracy drops on proof-verified rewritten benchmarks, suggesting that their performance is sensitive to surface-level and structural variations and may partly reflect memorization effects. Our source code and data are available at https://github.com/AI4Engi/RePro. --- ### ReNFT: Repairing Mode Collapse in Reward Post-Training via Internal Probability-Mass Recalibration Authors: Yuchen Bao, Chao Wen, Haowei Wang, Ruoxin Chen, Donghao Luo, Jiahui Zhan, Wenjian Huang, Shen Chen, Yiting Wang, Taiping Yao, Chengjie Wang, Shouhong Ding et al. Published: 2026-08-30 Categories: cs.LG, cs.AI, cs.CV arXiv: https://arxiv.org/abs/2609.00061v2 Reward post-training of diffusion generators inevitably concentrates probability mass on a few reward-favored modes, a mode collapse that erases within-prompt diversity. Existing methods for mitigating collapse rely on external signals or interfaces, augmenting the reward with perceptual objectives, adjusting reference regularization, or modifying the text encoder, but none repairs an adapter that has already collapsed while preserving the acquired reward. We observe that online post-training primarily reallocates probability mass over capabilities inherited from pretraining rather than learning new visual content. Collapse is therefore suppression, not deletion, and can be reversed from within the generator. We propose ReNFT, which repairs a high-reward, low-diversity adapter through internal probability-mass recalibration. Unconditional probes first prioritize "anti-hub" prompts where the prompt-independent bias is easiest to expose. Two policy-dominated mixed routes then generate matched counterfactual proposals from the same prompt and initial noise, one probing the frozen base direction for suppressed alternatives and the other exposing the post-trained unconditional tendency. Reward ranking with an adaptive flipping guard assigns pull and push roles, and a joint-and-paired NFT update realizes the repair. On PickScore and GenEval, ReNFT retains 98.9% and 99.0% of NFT's reward while improving DreamSim-Div by 58.8% and 55.0%, respectively, offering a complementary alternative to external interventions. --- ### A Formal Analysis of Agent Payment Protocols Authors: Ke Jiang, Mohan Yu, Yuan Chang, Mohit Kumar Jangid, Jianyu Niu, Cong Wang, Yinqian Zhang Published: 2026-08-30 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2609.00060v1 Agent payment protocols are emerging as a key transaction layer for autonomous commerce, enabling AI agents to purchase goods and services and execute payments on users' behalf. Unlike conventional payment flows, they distribute user intent, delegated authority, credential use, settlement, and fulfillment across multiple actors and stages, creating security dependencies that no single message or participant can enforce. Yet these guarantees remain largely implicit across evolving specifications, schemas, and reference implementations, with little systematic formal analysis. We formalize four representative agent payment protocols: x402, MPP, ACP, and AP2 in Tamarin. Using a common abstraction of the agent payment lifecycle, we construct source-grounded models that capture each protocol's roles, state, trust assumptions, and lifecycle transitions. Rather than assuming a complete property taxonomy, we use source-backed verification questions and counterexample traces to expose missing bindings, state constraints, and cross-stage correspondences, consolidating them into 18 shared security principles. Across 86 verification cases, our analysis reproduces 46 known or calibration cases and identifies 40 previously undocumented formal-consistency findings. For each retained violation, we isolate the missing protocol relation, construct a minimally strengthened reference model, and reverify the intended property. We further evaluate the new x402 findings across three implementations and validate ten representative findings through implementation PoCs, SDK/schema-level witnesses, and source-aligned executable traces spanning five security principles. Our results show that delegated authorization must remain consistent with its resulting economic and service effects across actors, states, and protocol stages. --- ### DISTAL: Distillation and Self-Supervised Pretraining for Structure-Agnostic Materials Property Prediction Authors: Weiran Wang, Xintong Huo, Yueying Wang, Yusi Fan, Wenyan Wang, Xin Feng, Ruihao Xin, Lan Huang, Kewei Li, Fengfeng Zhou Published: 2026-08-30 Categories: cs.LG, cs.AI, cs.ET arXiv: https://arxiv.org/abs/2609.00059v1 Materials property prediction remains difficult in low-data settings, where many target properties are supported by only a limited number of labeled samples. Models with the strongest predictive accuracy often depend on crystal structures, which restricts their use in early-stage screening when structural information is limited or unavailable. To address this challenge, we propose DISTAL, a dual-prior framework for structure-agnostic materials property prediction that combines self-supervised compositional pretraining with structure-aware knowledge distillation. DISTAL first learns transferable compositional representations from a large virtual composition space using 145 composition-derived descriptors. It then distills structural knowledge from a pretrained ALIGNN teacher into a composition-conditioned student. This setting allows structural priors to be used during training without requiring structural inputs at inference. By integrating explicit compositional descriptors, pretrained latent features, and distilled structural features within a unified prediction pipeline, DISTAL captures complementary signals that are difficult to recover from any single representation alone. Across 39 benchmark tasks, the best-performing multimodal configuration combines all three signals, and improves over the reference benchmark on 37 tasks. DISTAL achieves the strongest overall performance among all evaluated feature combinations. These results indicate that compositional pretraining and structural distillation provide complementary priors and offer a practical route to robust composition-only prediction in small-data materials informatics. The source code and the pre-trained models are anonymously available at: https://osf.io/eq96d/overview?view_only=451617f42f7849e08750bd1852b48980 and will be released at the official link after acceptance. --- ### CUDA-Harness: Harnessing Agentic CUDA Kernel Generation and Optimization from Natural Language Authors: Qi Fan, An Zou, Yehan Ma Published: 2026-08-30 Categories: cs.CL, cs.AI, cs.MA, cs.PL, cs.SE arXiv: https://arxiv.org/abs/2609.00058v1 Developing high-performance CUDA kernels demands specialized knowledge in algorithm implementation, correctness validation, and hardware-aware parallel optimization, creating a substantial expertise barrier and making generating CUDA kernels directly from natural language (Text2CUDA) essential. Meanwhile, the general-purpose code generation capability of Large Language Models (LLMs) prompts a series of works exploring LLM-based CUDA kernel generation. They mainly focus on transpilation from high-level frameworks such as PyTorch to CUDA (Torch2CUDA) rather than Text2CUDA, where models must understand the high-level input semantics and handle low-level kernel implementation and validation. Additionally, these methods are vulnerable to reward hacking due to reliance on predefined test inputs. In this paper, we propose CUDA-Harness, a framework for harnessing agentic CUDA kernel generation and optimization from natural language. Specifically, we introduce Intermediate-Structured Generation to connect high-level semantic understanding with low-level kernel generation. To dilute reward hacking in Text2CUDA, we construct Synthesis-Based Verification to provide isolated test data and progressive validation. Furthermore, we propose Feedback-Adaptive Evolution, a kernel evolution strategy that prioritizes correctness while optimizing performance. Finally, through extensive experiments, we demonstrate the effectiveness of CUDA-Harness, with further evaluations illustrating generalization across LLMs, hardware platforms, and to C-to-CUDA transpilation. --- ### ValueGraph: Value-Signal Guided Graph Pre-training for Contextualized User Representation Authors: Yitong Han, Wei Gao, Yi Zhao, Prasanta Bhattacharya, Fengzhu Zeng, Mohammad Amanlou Published: 2026-08-30 Categories: cs.CL, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.00057v1 Value signals are aggregated user-level moral representations that capture users' inferred value-related tendencies from their online discourse. User behavior on social media is shaped not only by what users say or whom they interact with, but also by the value signal through which they express attitudes. Existing user representation methods largely miss this value-relevant dimension. We propose ValueGraph, a graph pre-training framework that uses automatically inferred moral-value signals as noisy auxiliary signals for contextualized user representation. From post-reply graphs, ValueGraph learns semantic and structural representations and further aligns users through relative value similarity with contrastive and clustering objectives. Rather than treating inferred values as gold psychological labels, ValueGraph uses them as soft constraints for representation learning. Experiments on stance detection and twitter bot detection show consistent gains over strong text-based, graph-based, and text-only LLM baselines, highlighting value-signal guidance as a useful inductive bias for socially informed user modeling. --- ### Zero-Shot Respiratory Sound Classification through LLM-Augmented Audio-Text Alignment Authors: Mustafa Talha İlerisoy, Hung Manh Pham, Mathias Funk, Mykola Pechenizkiy, Aaqib Saeed Published: 2026-08-30 Categories: cs.CL, cs.AI, cs.SD arXiv: https://arxiv.org/abs/2609.00055v1 Self-supervised respiratory encoders lack semantic grounding in clinical domain needed for zero-shot inference, limiting their utility without task-specific labeled data. We propose a framework that aligns these encoders with medical terminology in a shared latent space turning them into a zero-shot-capable foundation model. To address paired data scarcity, we use a medical LLM to synthesize structured reports from metadata, creating dense semantic anchors for contrastive learning. Our training combines a sigmoid-based contrastive loss with encoder's native SSL objective and similarity-aware negative sampling to sharpen pathological boundaries. Across 9 tasks on 6 datasets, our method achieves a 61.3% mean zero-shot AUC, surpassing CLAP (51.4%) and Qwen2-Audio (54.9%) while reaching the highest linear probing AUC (71.6%) with only 43% of data used by full-scale baselines, showing that structured semantic alignment outperforms large-scale, general-purpose models in clinical diagnostics. --- ### Convergence issues in Relational Concept Analysis based on AOC-posets Authors: Xavier Dolques, Agnès Braud, Alain Gutierrez, Marianne Huchard, Florence Le Ber Published: 2026-08-30 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.00054v1 Formal Concept Analysis (FCA) is an approach for conceptual classification building and rule discovery from a binary table describing a set of objects by a set of attributes. Extensions have been proposed to deal with non-binary and more complex data, such as Relational Concept Analysis (RCA) for multi-relational data. RCA aims to highlight groups of objects characterized by their relationships with other groups of objects. The richer and more complex nature of the underlying data allows RCA to produce richer results than FCA, at the expense of higher computational and interpretive complexity. The most commonly used conceptual classification structure in FCA is the concept lattice. However, in many applications, concept lattice substructures, such as AOC-posets, are preferred over the full lattice, either to mitigate combinatorial blow-up or to focus on the most informative parts of the structure. Indeed, in an AOC-poset, only concepts introducing an object or an attribute are represented, which makes AOC-posets smaller and easier to compute and use than concept lattices. Although RCA was originally defined on concept lattices, it can also be instantiated on AOC-posets. RCA is iterative and its convergence is guaranteed in the lattice-based setting, but this guarantee is lost when using AOC-posets. In this paper, we investigate this loss of convergence in detail. We show why convergence is no longer guaranteed in the general case, identify conditions under which it can still be ensured, and discuss how a dataset can be transformed to recover convergence. We also propose a convergent variant of the process, which preserves the AOC-poset structure: relational attributes, once created, are never removed, which guarantees convergence at the price of attributes that may refer to concepts absent from the final structures. --- ### AgentProv: Auditing Agentic LLM API Providers via Tool-use Policy Probes Authors: Xun Wang, Bihe Zhao, Michael Backes, Franziska Boenisch, Adam Dziedzic Published: 2026-08-30 Categories: cs.CR, cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.00052v1 Commercial LLM APIs advertise a specific foundation model, but the served backbone may be silently substituted, quantized, or wrapped, for example to save deployment costs. All existing audits decide backbone identity from the text-output channel, which is structurally fragile for agentic APIs because modern serving stacks (OpenAI, Anthropic, Gemini, Cloudflare Workers AI, LangGraph) discard text and expose only structured actions when the model calls a tool, and provider-injected system prompts can distort text distributions enough that text-channel tests falsely accuse honest providers of substituting the claimed model. We observe that recent agentic post-training internalizes tool-use directly into the weights, opening a new audit channel that the serving stack still exposes and that is largely invariant to deployment context. We introduce Agentic Provenance (AgentProv), the first action-based identity audit for agentic LLM APIs: AgentProv fingerprints a deployed model through its categorical tool-call distribution and decides identity via an MMD permutation test. AgentProv catches every substituted model (100% on 630 evaluated checkpoint pairs), while holding the false-positive rate under system-prompt injection at 7% (vs. 67% for MET and 53% for RUT). On third-party API endpoints, AgentProv's disagreements with MET are consistent with an independent token-count side-channel that detects provider-injected system prompts. --- ### From Detection to Refusal: Safer LLMs via Circuit-Guided Weight Scaling Authors: Kuan-Lin Chu, Chung-En Sun, Tsui-Wei Weng Published: 2026-08-30 Categories: cs.CL, cs.AI, cs.CY arXiv: https://arxiv.org/abs/2609.00051v1 Despite extensive alignment efforts, Large Language Models (LLMs) remain vulnerable to generating unsafe content under adversarial prompting, yet the internal mechanisms by which safety behaviors are implemented remain poorly understood. We study LLM safety from a mechanistic interpretability perspective and characterize a multi-stage *safety circuit* that organizes refusal behavior, consisting of (i) $\textbf{Harmful Detection Heads}$ that respond to harmful inputs, (ii) $\textbf{Safety Neurons}$ that mediate and stabilize safety signals in the residual stream, and (iii) $\textbf{Refusal Heads}$ that translate these signals into safe response generation. Using targeted attention-head and neuron-level interventions, we provide causal evidence consistent with this circuit organization, showing that suppressing upstream Harmful Detection Heads disrupts downstream refusal behavior and that safety neurons mediate this interaction. We validate that this decomposition recurs across multiple LLM architectures and adversarial attack settings, and use simple, architecture-preserving weight scaling as a mechanistic probe to test its functional relevance. Across six LLMs, circuit-guided scaling improves safety rates under attacks by 26.5%, while incurring only a 1.7% accuracy drop across four standard benchmarks. Overall, our results support a circuit-level interpretation of LLM safety and suggest that mechanistic abstractions can reveal stable and transferable patterns underlying aligned behavior. --- ### Towards Agentic Cloud Engineering: Graph and Loop Engineering with a Zero-Trust Agent Harness Authors: Sagar Srinivas Sakhinana, Venkataramana Runkana Published: 2026-08-30 Categories: cs.SE, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.00050v1 Agentic AI is enabling cloud-based workflows in which autonomous agents reason over operational state, invoke authorized tools, modify software and infrastructure, deploy services, verify execution outcomes, and adapt across long-horizon, multistep tasks. Engineering such workflows requires explicit mechanisms for workflow progression, constrained execution, failure recovery, and verifiable completion. We present Agentic Cloud Workflow Engineering, an agentic AI framework that transforms natural-language agentic cloud-engineering tasks into validated code repositories and verified operational cloud deployments for automating cloud-based agentic workflows. The framework separates three complementary concerns: graph engineering specifies long-horizon workflow progression and verification-dependent transitions; loop engineering provides bounded diagnosis, repair or re-planning, retry, and re-verification; and agent harness engineering enforces zero-trust execution through identity, authorization, policy-scoped capabilities, isolation, and runtime safeguards. Workflow progression and completion require machine-checkable repository, deployment, and runtime evidence, with recovery constrained by explicit operational bounds and termination criteria. We instantiate the framework on Google Cloud and evaluate repository completeness, controlled execution, evidence-gated progression, operational deployment, and bounded recovery. Experimental results show that executions terminate with either a verified operational cloud deployment or an auditable terminal failure under bounded recovery. The framework provides a unified engineering architecture for cloud-based workflows spanning Agentic DevOps, Agentic CloudOps, Agentic SRE/AIOps, Agentic SecOps, Agentic DataOps, Agentic MLOps/LLMOps, AgentOps, Agentic RAG/GraphRAG, and related cloud-engineering domains. --- ### REAL-Q: E2E LLM Quantization via Dynamic Gradient Descent Authors: Qian Zhang, Yaoming Li, Zhewen Tan, Yanshu Wang, Heng Lu, Kun Su, Zongwei Lv, Wenhan Yu, Yongge Ma, Yinjun Han, Ruikang Liu, Tong Yang Published: 2026-08-30 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.00049v2 Post-training quantization (PTQ) is essential for deploying large language models (LLMs) under strict resource constraints. State-of-the-art PTQ methods quantize each layer with a single closed-form second-order solver: to remain analytically tractable, they heavily approximate the global loss (dropping cross-channel coupling, pooling output rows into groups), and they then freeze the resulting Hessian across the entire layer, with no way to refresh it as the loss landscape shifts column by column--a phenomenon we call information misalignment. We propose REAL-Q (Real-time E2E-loss Aligned LLM Quantization), a novel PTQ paradigm that breaks this compromise: instead of diluting the objective for the sake of analytic tractability, REAL-Q targets an end-to-end-aligned surrogate of the global loss and refines it via fine-grained, dynamic Block-wise Gradient Descent applied after every column block (128 columns). By coupling this fine-grained correction with a sliding window mechanism for smooth cross-layer transitions, REAL-Q effectively mitigates error propagation across the network. On LLaMA-3.1 (8B and 70B) and Qwen3 (0.6B-32B) at W4A16, REAL-Q reduces end-to-end KL divergence by up to ~49% relative to state-of-the-art globally-guided methods. --- ### GUI-CC: Benchmarking Contextual Consistency of GUI World Models as Agent Environments Authors: Lin Fu, Zheyuan Yang, Tianhui Zhang, Jinbiao Wei, Guo Gan, Boxu Liu, Yilun Zhao, Yu Rong Published: 2026-08-30 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.00048v1 GUI world models are increasingly evaluated as one-step next-screen predictors, yet their intended use is often as multi-step environments for GUI agents. This mismatch leaves a key requirement under-tested: generated states must remain contextually consistent when they are repeatedly reused for future interaction. We introduce GUI-CC, a benchmark that evaluates contextual consistency of GUI world models as agent environments rather than isolated next-screen predictors. GUI-CC contains two complementary tracks: an offline reference-action track that rolls models along real mobile GUI trajectories, and an online agent-loop track that lets fixed probing agents interact with model-generated UIs. We construct 500 offline trajectory tasks from GUIOdyssey and 200 emulator-verified online tasks across 30 mobile apps. GUI-CC evaluates transition fidelity, transition plausibility, contextual consistency, and task progress. Experiments show that plausible single-step generation does not guarantee reliable environment simulation: current models often produce usable-looking screens while failing to preserve task-relevant context or support executable multi-step rollouts. --- ### Task-Specific Prompt with Global Context for Multi-Task Graph Pre-Training Authors: Zhiyang Qiu, Yangtao Wang, Xiaocui Li, Yanzhao Xie, Siyuan Chen, Wensheng Zhang Published: 2026-08-30 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.00047v1 Graph prompt learning is an effective paradigm to adapt pre-trained graph models to downstream tasks in low-resource scenarios. However, existing multi-task graph pre-training frameworks generally use randomly initialized prompts, leading to poor alignment between the prompt space, pretext objectives and graph structural characteristics. This greatly weakens the task relevance, structural awareness and transferability of prompt representations. To address this challenge, we propose TPGC, a dual-prior prompt initialization solution that explicitly models the synergy between task prior and structural prior. Specifically, the Task-Prior Injection Module first conducts a short homologous multi-task pre-training on an auxiliary graph, enabling prompt initialization to inherit optimization preferences associated with multiple pretext tasks. Built on the task-aware representations, the Structure-Prior Injection Module further extracts transferable global structural context from the auxiliary graph, converting it into layer-wise prompt vectors by aggregating structurally informative node embeddings. Extensive experiments on 6 mainstream benchmarks covering node and graph classification show that TPGC achieves consistently better performance under few-shot settings than state-of-the-art baselines, with fewer downstream tunable parameters and lower runtime. The code is available at https://github.com/Virgilqiu/TPGC --- ### RAPIDMap: Rapid Multi-Agent Pipeline for Interpretable Disaster Mapping from Satellite and Street-view Imagery Authors: Yifan Yang, Lei Zou Published: 2026-08-30 Categories: cs.MA, cs.AI, cs.CY arXiv: https://arxiv.org/abs/2609.00046v2 Rapid and reliable disaster mapping of impacted areas, damaged infrastructure, and affected populations is essential for emergency response and recovery. However, existing AI-based approaches often require extensive manual annotation, lack cross-hazard generalization, and rely on single-modal observations. To address these challenges, this paper proposes RAPIDMap, a rapid multi-agent pipeline for zero-shot interpretable disaster mapping from satellite and street-view imagery. The framework integrates four intelligent agents: Disaster Perception Agent (DPA), Image Restoration Agent (IRA), Damage Recognition Agent (DRA), and Disaster Mapping Agent (DMA). By combining remote sensing and street-view data, RAPIDMap eliminates the need for manual fine-tuning, generalizes across multiple disaster categories, and generates structured, map-ready disaster intelligence with recovery recommendations. --- ### Dense Weak Hiding: Closing Complexity Gaps in Nonconvex and PL Finite-Sum Optimization under Individual Smoothness Authors: Yuxing Peng, Zhiqing Tang, Weijia Jia Published: 2026-08-30 Categories: cs.DS, cs.LG, math.OC arXiv: https://arxiv.org/abs/2609.00045v2 Under individual smoothness, the optimal incremental first-order oracle (IFO) complexity of nonconvex finite-sum optimization is open. Known algorithms use $O(n+\sqrt n\,ΔL_{\max}/\varepsilon^2)$ calls, while existing lower bounds miss the factor $\sqrt n$ in the second term. We prove the matching lower bound $Ω(n+\sqrt n\,ΔL_{\max}/\varepsilon^2)$ for randomized IFO algorithms, including those that choose component indices and query points from the full preceding history. Thus PAGE and SPIDER are minimax optimal up to universal constants under individual and mean-squared smoothness. Under the global Polyak--Lojasiewicz (PL) condition, we use a similar idea to obtain an $Ω(n+κ_{\max}\sqrt n\log(Δ/\varepsilon))$ lower bound for large $κ_{\max}$. Existing PL lower bounds have focused only on relatively large $κ$. For the previously unexplored regime $κ_{\max}<\sqrt n$, we discover a new rate, $Ω(n+n\log(Δ/\varepsilon)/(1+\log(\sqrt n/κ_{\max})))$. This lower bound motivates our Restarted PAGE algorithm, whose upper bound matches the new rate for small $κ_{\max}$ and recovers the standard PAGE rate for large $κ_{\max}$, showing that both bounds are nearly tight. Our lower bounds use the proposed dense weak hiding construction. It spreads each hidden direction across all components, so an individual IFO query reveals only weak information while the full average preserves the direction. Unrevealed stages remain inactive even for arbitrary query points, forcing many calls to expose a stage before substantial progress is possible. Balancing this revelation cost against the number of stages allowed by individual smoothness yields the missing $\sqrt n$ factor. --- ### A Hybrid State-Space Approach for Census-Tract Population Estimation Authors: Jackson R. Ye, Alexandre V. Morozov Published: 2026-08-30 Categories: cs.CV, cs.LG arXiv: https://arxiv.org/abs/2608.30094v1 Sequence models---the architecture family behind large language models and, increasingly, state-of-the-art image recognition---have redefined how machines learn from high-dimensional data. Yet population estimation from satellite imagery, a task that underpins infrastructure planning, public health, and disaster response, has scarcely benefited: leading systems still bind population to a uniform raster, disaggregating census counts onto grid cells through weighting surfaces built from ancillary data (e.g., in WorldPop and LandScan), which can introduce systematic spatial bias, and predicting population per grid cell with convolutional neural networks. In this approach, the administrative-unit structure in which the census was actually collected is discarded. We close this gap with MambaPop, which renders each administrative unit as a single polygon-masked satellite image and treats tract-level population estimation as a sequence-modeling problem over its image patches, pairing each tract image directly with its population label and eliminating the disaggregation step entirely. Built on the hybrid state-space--attention MambaVision backbone, MambaPop is, to our knowledge, the first method to learn population directly from an administrative unit's own image as well as the first to apply a state-space based (Mamba) hybrid architecture to the population estimation task. Across all $\sim$84{,}000 contiguous-US census tracts of the 2020 census, MambaPop attains a mean absolute error (MAE) of $1{,}141$ persons per tract, matching the strongest convolutional baseline (YOLOv11, MAE $1{,}122$). --- ### Robust K-means Clustering using the Density Power Divergence Measure Authors: Anirban Mondal, Paromita Banerjee, Abhijit Mandal Published: 2026-08-30 Categories: stat.ME, stat.AP, stat.CO, stat.ML arXiv: https://arxiv.org/abs/2608.30093v1 We introduce a robust clustering method, MK-means DPD, that estimates cluster centers and covariance matrices using density power divergence (DPD) measures combined with Mahalanobis distance, making it resistant to outliers and adaptable to heterogeneous, elliptical clusters, unlike the classical K-means algorithm. Since Mahalanobis distance-based K-means lacks a general convergence guarantee, we further introduce a convergent variant, Density-Consistent MK-means DPD (DC-MK-means DPD), which redefines the cluster assignment step in terms of a pointwise DPD loss. We prove a formal theorem establishing that the resulting algorithm converges in a finite number of steps. We also propose two new robust internal evaluation indices, a Median Davies-Bouldin Index and a Trimmed Calinski-Harabasz Index, to ensure that performance comparisons are not themselves distorted by outliers. The efficacy of the proposed methods is demonstrated on simulated data, showing superiority over existing methods, and on two real datasets: Iris data, to identify similar species, and COVID-19 case fatality rate and infection rate data for countries worldwide, examining the resulting clusters' geographic and socio-economic patterns. --- ### Arkios: An Open Bilingual English-Nepali Language Model Trained From Scratch, with a Devanagari-Aware Tokenizer Authors: Sajal Regmi, Siddhartha Pudasaini, Chetan Phakami Pun Published: 2026-08-30 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2608.30092v2 We present Arkios, a 1.04B-parameter dense transformer pretrained from scratch on 150B tokens of bilingual English-Nepali text, using a custom single-file C/CUDA training stack and a Devanagari-aware byte-level BPE tokenizer built for this project. On ARC-Easy and ARC-Challenge, Arkios exceeds three comparably sized open models (Pythia-1.4B, TinyLlama-1.1B, OLMo-1B) despite an order of magnitude fewer training tokens, likely aided by a match between our educational-web-text pretraining data and ARC's grade-school-science format rather than a general capability advantage. We report full evaluation results under standard protocols, including a correction to an earlier partial-sample estimate, and findings specific to evaluating small models in a low-resource language: the standard multiple-choice-letter prompt format used by common evaluation harnesses places this model at chance on Nepali reading comprehension, and simultaneously at chance on English in the same format, which would lead a naive benchmark run to conclude the model has no Nepali ability when in fact it does. Concretely, both languages score at chance in the letter-choice format (0.240 Nepali, 0.236 English, against a chance baseline of 0.250), while scoring the answer text directly reveals genuine, English-favoring comprehension (0.306 Nepali, 0.387 English). We describe a manifest-conditioned tool-use contract introduced during instruction tuning, where tool calls are permitted only when a tool manifest is declared in context and suppressed otherwise, and report where that contract holds and where it does not. We release both the base and instruction-tuned model weights under Apache-2.0. The training code and a small privately-sourced portion of the Nepali pretraining corpus are not released; everything needed to reproduce the reported numbers from the released weights is included here. ---