# Research Papers — 2026-07 · bundle page 155 of 222 > Full-text markdown of papers #7701–7750 of 11075 published in 2026-07, newest first. AI-readable mirrors of the full 11075-paper 2026-07 archive (50 papers per file). Canonical page: https://aipod.dev/papers/archive/2026-07/page/155 [← Previous bundle](https://aipod.dev/papers/archive/2026-07/page/154.md) [Next bundle →](https://aipod.dev/papers/archive/2026-07/page/156.md) ### Correlation-Aware Contextual Bandits with Surrogate Rewards for LLM Routing Authors: Ajay Narayanan Sridhar, Ronak Singh, Mehrdad Mahdavi, Vijaykrishnan Narayanan Published: 2026-07-10 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2607.09015v1 We study contextual bandit problems with correlated arms and access to surrogate reward signals produced by a machine learning model, motivated by applications such as large language model (LLM) routing. Unlike classical contextual bandits that rely solely on bandit feedback and assume conditional independence across arms, our setting allows context-dependent inter-arm correlations and auxiliary reward information that may be noisy or misspecified. We propose algorithms that leverage such surrogate rewards through two complementary designs. A coupled reward-mixing approach pools true and surrogate rewards to accelerate learning when surrogate signals are reliable, while a decoupled prediction-mixing approach maintains separate estimators for bandit feedback and surrogate rewards and adaptively combines their predictions. This decoupling yields robustness to surrogate misspecification, recovering regret guarantees comparable to reward-only bandit methods in the worst case, while achieving improved regret when surrogate predictions are sufficiently informative. We provide theoretical regret analyses for both approaches and evaluate them on LLM routing benchmarks under varying accuracy versus cost trade-offs. The results demonstrate improved sample efficiency and consistently better accuracy-cost trade-offs compared to standard contextual bandit baselines and strong static routing methods. --- ### C-GAP: Class-Aware and Online Prompting Improves Vision-Language Models on Imbalanced Classes Authors: Francis Fernandez, Arash Jahangiri, Salimeh Sekeh Published: 2026-07-10 Categories: cs.CV arXiv: https://arxiv.org/abs/2607.09008v1 Safety-critical perception systems must reliably detect rare object classes within small label spaces, a setting that long-tailed detection methods, designed for hundreds of classes with dense annotation, fundamentally do not address. Open-vocabulary detectors offer a promising alternative, as they use natural language queries at inference time, making prompt quality a first-class lever for detection performance. We exploit this property to address class imbalance: rather than retraining models or collecting additional annotations, we ask whether iteratively refining the language prompts, fed to frozen detectors, can improve minority class detection. We introduce C-GAP Caption-Guided Augmentation and Prompting), a detector-agnostic, annotation-free framework that operates in two phases. First, we establish a composite caption baseline combining per-image scene descriptions with class-quantity context, which we show outperforms scene-description only or class-quantity-only prompts across multiple open-vocabulary architectures and benchmarks. Second, an LLM iteratively refines each image's caption individually, with trials triaged into accept, tentative, or regenerate buckets based on minority-class AP@0.5 against a dynamic threshold derived from the composite baseline. Refinement terminates early once sufficient AP@0.5 gain is achieved. No detector weights are updated at any stage. Our experiments shows that C-GAP improves minority-class average precision up to 53% over the baselines. On COCO, C-GAP improves minority-class AP@0.5 by ~81% relative over the composite baseline (17.69 -> 32.09). Experiments confirm that composite captions provide the critical foundation for effective refinement: using scene-description-only or class-quantity-only prompts as the refinement starting point yields diminishing returns, supporting both stages of C-GAP as necessary contributions. --- ### RaMark: Radioactive Watermarking for Generated Tabular Data Authors: Xin Che, Lingyang Chu, Qiqi Zhang, Xinyu Ma, Xuan Luo, Jian Pei Published: 2026-07-10 Categories: cs.CR, cs.LG arXiv: https://arxiv.org/abs/2607.09000v1 Recent advances in generative modeling have made generated tabular data a practical solution for privacy-sensitive data sharing, where watermarking enables ownership verification. However, existing watermarking methods fundamentally fail under retraining attacks, in which an adversary retrains a generative model on a watermarked dataset and regenerates high-utility data that no longer carries the watermark. We address this challenge by introducing radioactivity, the property that a watermark remains detectable after generative model retraining, and propose RaMark, a radioactive watermarking method that embeds a sinusoidal dependency as an intrinsic component of the data distribution. By coupling the watermark with the underlying distribution, RaMark ensures that any generative model preserving data utility also has to preserve the watermark. We theoretically show that with high probability removing watermark degrades utility and alters data distribution. Extensive experiments on two real-world tabular datasets, under a large-scale ownership verification setting with $10^5$ independent data owners, demonstrate that RaMark achieves substantially stronger radioactivity than seven state-of-the-art methods and consistently outperforms them against both retraining and data modification attacks. --- ### When Should Forecasting Agents Reason? Behavioral Stress Tests for Reliability Routing Authors: Yufeng Wang Published: 2026-07-09 Categories: cs.AI, cs.CL arXiv: https://arxiv.org/abs/2609.28475v1 Forecasting agents increasingly combine language-model reasoning, retrieval, ensembling, and calibration, but it remains unclear when each behavior should be trusted. We study this question on ForecastBench-style binary forecasting tasks, treating the choice to retrieve, reason, defer to a market prior, or use a historical analog as an observable agent behavior rather than a hidden implementation detail. Our central finding is that mechanism choice is source-dependent: structured analogs dominate for some data-generating processes, while market/crowd-style and conservative baselines are better for others. We introduce ReliabilityRoute, a structural intervention that steers forecasting-agent behavior using reliability features such as historical coverage, market-prior availability, source-prior sharpness, evidence strength, evidence disagreement, and horizon. A fixed 2024-fitted rule closely matches a hand taxonomy without hard-coded source-name decisions, while a walk-forward self-adjusting rule refits thresholds from previously resolved vintages and obtains the best mean Brier score among our deterministic systems across 16 later LLM vintages. The gain is modest and historical/search baselines remain highly competitive. The main contribution is therefore a behavioral stress test showing that more reasoning is not always better; forecasting agents should first estimate which evidence source deserves control, routing policies should themselves adapt under auditable constraints, and reproducibility artifacts are available at https://github.com/louiswang524/forcastagent --- ### Self-reported archetypes and behavioral failures in Large Language Models Authors: Tabia Tanzin Prama, Calla Glavin Beauregard, Christopher M. Danforth, Peter Sheridan Dodds Published: 2026-07-09 Categories: cs.CL, physics.soc-ph arXiv: https://arxiv.org/abs/2609.15998v1 Every large language model (LLM) has behavioral traits and moral preferences that comprise its character. Whether by design or as an emergent property of training, these systems exhibit persistent dispositions that shape how they interact, comply, resist, and err, yet the structure of LLM character remains poorly understood. We map the self-reported personality archetypes of 22 LLMs spanning closed-source frontier systems (GPT-4.0-5.2, Grok-3/4, Gemini 2.5 Pro/Flash, Claude Sonnet 4.5/4.6) and open-source models (Llama, DeepSeek, OLMo, and Qwen series). Each model self-rated across 464 bipolar semantic-differential trait pairs, and the resulting profiles were projected into a six-dimensional archetypal space derived from crowd-sourced ratings of 2,000 fictional characters using the Archetypometrics framework. Closed-source models' self-rating traits align with the empirical trait co-occurrence structure of human-rated fictional characters, suggesting coherent, human-like self-representations organized around combinations of four recurring archetypal dimensions: Hero, Angel, Traditionalist, and Geek. Their closest analogues include Data, Vision, and Janet. Open-source models show weaker, noisier, and internally contradictory self-representations, occupying a diffuse region of archetype space with weak structure. Cross-referencing self-reported profiles with developer constitutions reveals a consequential gap between claimed character and enacted behavior: hallucination undermines claimed precision, sycophancy complicates claimed kindness, and agentic failures contradict claimed obedience. These self-ratings should therefore be interpreted not as neutral measurements of model character, but as structured outputs of the same optimization processes that shape model behavior. This work provides a reproducible, character-grounded framework for evaluating what LLMs are, not just what they do. --- ### Crash Narrative-Guided Countermeasure Recommendation Using Large Language Models: A Retrieval-Augmented Generation Framework for Intersection Safety Authors: Abu Saif Md Nasim Uddin, Mohamed Abdel-Aty, Zubayer Islam, Parvez Anowar, Chenzhu Wang Published: 2026-07-09 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.15997v1 Improving safety at intersections requires identifying crash mechanisms and recommending appropriate countermeasures. However, this process traditionally relies on expert judgment, making it labor-intensive, difficult to scale, and dependent on the availability of experienced traffic safety engineers. Although crash narratives contain rich description of crash mechanisms, this unstructured information remains largely underutilized in safety analyses. This study presents a crash narrative-guided retrieval-augmented generation (RAG) framework that translates narrative-derived crash mechanisms into site-specific countermeasure recommendations. Key mechanism attributes including traffic control, signal indication, driver fault, vehicle movement, and travel direction were extracted from crash narratives and linked to evidence-based treatments from the FHWA Proven Safety Countermeasures and the CMF Clearinghouse. The framework integrates embedding-based retrieval of historically similar intersections, association-rule mining, statistical guidance on the expected number of relevant countermeasures, and an engineering reasoning guidance that directs LLM through a domain-consistent decision process before selecting countermeasures. Evaluated on 312 fatal and serious-injury crashes across 115 intersections in Lake and Sumter Counties, Florida, using five-fold cross-validation, the framework achieved a precision of 0.82, recall of 0.85, and F1-score of 0.82, while recommending an average of 3.91 countermeasures per location with 3.14 matching, closely matching the actual average (3.86). Overall, the proposed framework demonstrates the potential of retrieval-augmented LLMs as an interpretable and scalable decision-support tool for transportation agencies for translating crash narratives into countermeasure recommendations. --- ### Comment on arXiv:2607.01233: Survivorship Bias in Published-Paper Baselines for Research-Idea Distributions Authors: Fredrik A. Dahl Published: 2026-07-09 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.15996v1 Chen, Zhao, and Cohan introduce a valuable distributional evaluation of LLM-generated research ideas. This comment raises a narrower identification concern: their human baseline consists of published papers, whereas the LLM baseline consists of one-shot proposals. If bridge-like or synthesis-like ideas are relatively easy to generate but relatively unlikely to survive publication, then the published human baseline will understate their prevalence in the unseen human idea pool. The observed human--LLM gap may therefore be partly, or even largely, a consequence of survivorship bias. --- ### Bias Audits Detect Bias but Disagree on Ranking: Evidence from Ten Instruments and Ten Frontier Models Authors: William Guey, Pierrick Bougault, Wei Zhang, Vitor D. de Moura, José O. Gomes Published: 2026-07-09 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.15995v1 Emerging AI regulation mandates bias audits of high-risk systems, and audit scores are beginning to be used to rank models. Both uses assume different audit tools measure the same thing well enough to compare. We test that assumption directly, running ten extrinsic audit instruments over a shared panel of ten frontier models through one pooled inference gateway, first on occupational gender bias, then on age and socioeconomic status. Detection succeeds while ranking fails. Eight of ten tools detect bias with confidence intervals clear of zero; two widely cited direct-probe benchmarks are saturated because frontier models now answer neutrally. But cross-tool rank agreement is indistinguishable from chance (Kendall's W=0.07, p=0.83). A positive control with six deliberately weaker models separates two explanations: within-tool reliability recovers once the panel spans real capability gaps, yet cross-tool ranking never recovers, which points to the tools measuring different constructs rather than one construct noisily. Even the direction of bias splits by audit format: forced-choice decision tools mostly over-correct (toward women, and toward working-class candidates in 273 of 278 hiring decisions), while free generation and default coreference stay stereotype-congruent. The pattern replicates on socioeconomic status; an apparent ranking agreement on age dissolves under the paper's own tool-inclusion rules. The practical message: a single audit can detect bias and estimate its direction within its own operationalization, but no single audit supports ranking one model against another. All raw responses, code, and the analysis that recomputes every reported number from source are available at https://github.com/williamguey/bias-audit-agreement. --- ### AI, Brain Death Detection, and Islamic Law Authors: Muhammad Aurangzeb Ahmad Published: 2026-07-09 Categories: cs.CY, cs.AI arXiv: https://arxiv.org/abs/2608.16903v1 The deployment of machine learning systems capable of detecting covert consciousness in neurologically injured patients creates a profound challenge at the intersection of clinical medicine, AI ethics, and Islamic jurisprudence. We argue that the shift from binary clinical verdicts to probabilistic, temporally granular neural-state estimates should be addressed through three foundational constructs in Islamic legal epistemology: bayyina (clear evidentiary proof), yaqin (epistemic certainty), and the theologically mandated agnosticism about there (soul). We survey the current technical literature on AI-based consciousness detection, map it onto the landscape of Islamic brain death scholarship, and identifykey challenges. We also discuss its implications for AI surrogate decision systems. --- ### Advancing Health Equity through Multi-Level Fairness in Health Informatics Authors: Nick Souligne, Vignesh Subbian Published: 2026-07-09 Categories: cs.CY, cs.LG arXiv: https://arxiv.org/abs/2608.16902v1 The increasing integration of machine learning in healthcare has highlighted critical challenges related to fairness, transparency, and health equity. Specifically, the use of multi-level fairness techniques, which combine multiple bias mitigation steps or techniques, show promise for reducing biases across different patient demographics, yet this approach remains underexplored in terms of its health equity outcomes. In this paper, we assess the current landscape of multi-level fairness in health informatics by focusing on its impact on equitable healthcare outcomes and evaluating how transparency and reporting standards contribute to these advancements. Through an examination of the existing literature, we identify key gaps in both the implementation of multi-level fairness techniques and the consistent reporting of health equity impacts. Furthermore, we analyze the role of reporting standards, including MINIMAR and TRIPOD, in improving model transparency and ensuring that machine learning models in healthcare address health disparities. These standards offer valuable benchmarks for reporting on ML models, yet we identify key opportunities for enhancing how these reports capture fairness and equity outcomes. The paper concludes by providing recommendations that focus on improving transparency in reporting, advocating for the broader adoption of multi-level fairness techniques, and ensuring that health equity is explicitly prioritized in future research efforts. --- ### QuantumNovelty: A Skill-Orchestrating Language Agent for Referee-Style Review and Patentability Screening of Quantum Papers and Patents Authors: Shlomo Kashani Published: 2026-07-09 Categories: physics.soc-ph, cs.AI, cs.CY, quant-ph arXiv: https://arxiv.org/abs/2608.16900v1 Language-model agents increasingly produce quantum-science results; we ask whether the same agentic paradigm can also scrutinize them in an auditable, reproducible, and cost-transparent form. We present QuantumNovelty, an open-source skill-orchestrating language agent that both generates quantum-computing artifacts (papers, Pareto-front ansatz candidates, and patent drafts) and reviews them through simulated referee and patent-examiner panels. Its design contribution is an audit-and-falsify layer of deterministic gates -- strict Pareto domination, numerical recomputation from on-disk artifacts, Wilson small-sample intervals, and a cross-vendor consensus guard -- that constrains, rather than generates, the claims allowed to survive; every model call is logged with backend, token count, and cost. We make no accuracy claim against human experts, and validate only what is checkable without human labels: on a planted adversarial corpus the deterministic gates catch every planted overclaim with no false positives, and on a first deployment (six manuscripts and one granted patent, at a measured cost of about twenty-four US dollars) the panels are directionally more conservative than the public acceptance record, on a one-sided sample. The framework is decision support, not a replacement for peer review or patent examination, and we report in full where its mechanisms remain unexercised on real inputs. --- ### UltraArUco: A Lightweight Multilingual Library And Framework With Low-Latency Real-Time Marker-Based Tracking System For Mobile AR Interaction Authors: Mikhail Kiselev, Aleksandr Marukhin, Ivan Snegirev, Elizaveta Semenyakina, Miguel Altamirano Cabrera, Dzmitry Tsetserukou Published: 2026-07-09 Categories: cs.HC, cs.AI, cs.CV arXiv: https://arxiv.org/abs/2608.13584v2 UltraArUco is a lightweight multilingual library and framework for low latency, realtime marker-based tracking in mobile augmented reality. Unlike standard OpenCV-based implementations, UltraArUco introduces an optimized multilingual wrapper that reduces per-frame latency in six times, while maintaining high accuracy. Distributed Wi-Fi architecture provides portability, connects a mobile device (camera input) with a PC-based visual application, enabling responsive interactions. The framework is validated through an interactive piano simulation, where static ArUco markers on keys enable occlusion based note triggering, and hand-mounted markers provide spatial gesture recognition. UltraArUco's system requirements make it highly suitable for resource-constrained mobile AR applications, demonstrating a viable AR music application without specialized equipment. --- ### Closed-Loop LLM Co-Pilots for Digital Agriculture Authors: Serge Kernbach Published: 2026-07-09 Categories: cs.AI, physics.bio-ph arXiv: https://arxiv.org/abs/2608.09949v1 This study evaluates the application of Large Language Models (LLMs) in complex biological systems, evolving from data analysis to autonomous, AI-guided experimentation. The framework is driven by data from a 49-channel phytosensor network, encompassing multispectral, electrochemical, and dielectric modalities. To enhance accessibility, the system provides real-time natural-language interpretation for both specialists and non-experts. However, its core advantage lies in the transition from human-in-the-loop analysis to autonomous control. Processing biophysical data, the LLM evaluates plant physiology and triggers hardware actuators to optimize microclimates, execute phenotyping protocols, or induce controlled stress scenarios. This closed-loop architecture establishes a direct AI-biology interface, enabling data-driven exploration of complex biosystems and ecologies. The framework was validated across three case studies, based on a vertical farm and a single-plant setup and deciphered complex micro- and macro-fluctuations in plant physiology. Agents in a production-scale deployment executed multi-parameter optimization, balancing biomass accumulation, chlorophyll content, and energy consumption. The LLM processed biosensing telemetry to modulate full-spectrum, 450 nm, and 660 nm lighting at 2-hour intervals. Compared to periodic control, the system in minimal-time mode reduced the production cycle by 35%. In the energy-optimization mode, it reduced energy consumption by 18% with only a marginal increase in cultivation time, exploiting physiological inertia via light pulses. Finally, the agents autonomously developed an unforeseen strategy of dark-induced chlorophyll accumulation, resulting in a 67.9% energy saving. This framework transforms LLMs into autonomous co-pilots for digital agriculture, improving the cost-to-value ratio and lowering computational and expert-labor constraints. --- ### Role Steering of Language Models for Social Simulations Authors: Isaac Song, Mohammed Rehan Parwani, Glenn Matlin, Emile Anand, Akhil Theerthala, Arjun Chatterjee, Anthony Wen-Ming Zang, Maria Kostylew, Yonadav G. Shavit, Sebastien Krier, Mark Riedl Published: 2026-07-09 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2608.00023v2 Social simulations built from language-model agents need role-conditioned behavior that can be checked before agents are placed into a simulated population. We introduce an activation-steering screening workflow for role-conditioned agents: define a role profile, extract a role-specific direction, sweep four steering coefficients, evaluate role-profile alignment, and pass or flag each candidate configuration. On OLMo-3-7B-Instruct, we apply the workflow to a mixed 275-role inventory with 228 role-agnostic questions, GPT-4.1-mini prompted role references, and GPT-4.1-mini judges. Role-specific directions receive higher judged role-profile alignment than an assistant-axis directional control from prior persona-vector work, with mean overall scores of 63.2 versus 41.1 across the tested grid. They also preserve high lexical diversity, while the control drops sharply at larger coefficients. The role-level screen is the main practical output: most roles improve as steering increases, but 38 roles decline across all six measured dimensions, showing why simulation builders should choose coefficients per role rather than deploy a uniform high-strength setting. We make our code and evaluation artifacts available at https://anonymous.4open.science/r/anonymous-research-code-5F03/. --- ### Autonomous disproofs of the sum-product conjecture over $\mathbb R$ with GPT-5.5 Pro Authors: Yichen Huang Published: 2026-07-09 Categories: cs.AI, math.CO, math.NT arXiv: https://arxiv.org/abs/2607.20525v1 OpenAI's recent disproof of the Erdős unit distance conjecture marked a milestone for AI in mathematics. It also inspired another breakthrough: a human disproof of the Erdős--Szemerédi sum-product conjecture over $\mathbb R$. In this paper, we present a simple agent built on GPT-5.5 Pro. Using a problem-agnostic, three-stage prompting pipeline -- proof-plan proposal, proof construction, and review -- the agent autonomously generated correct proofs that the sum-product conjecture is false over $\mathbb R$ in 7 of 8 independent trials; in the remaining trial, it identified an unresolved gap in its argument. The seven proofs are diverse: some are close to existing unit-based constructions, while others avoid units by using $L^p$-type regions of algebraic integers. The system used an average of 132.4k reasoning tokens per trial. We release the code, intermediate outputs, and generated proofs, providing a reproducible, data-contamination-free case study in autonomous proof generation. --- ### Attention Degradation, Function Token Anchoring, and the Limits of Attention-Based Intervention in Large Language Models Authors: Sagar Dangal, Manoj Shakya Published: 2026-07-09 Categories: cs.AI arXiv: https://arxiv.org/abs/2607.20524v1 Mean cross-positional attention degradation is widely reported in transformer interpretability, yet whether it causally limits contextual retrieval remains untested. We present six coordinated experiments across GPT-2, LLaMA-3.2-1B/3B, OPT-1.3B, and distilgpt2. We first characterise short-term (5-100 token) attention degradation, finding a universal exponential-then-plateau pattern whose rate is inversely correlated with depth, with distinct layer-wise entropy signatures per architecture. Function token anchoring proves architecture-dependent: OPT-1.3B (absolute positional encoding) shows distance-dependent preposition specificity, GPT-2 shows uniform non-specific dependence, and LLaMA (RoPE) shows reversal at long distances. Strategic comma insertion at clause boundaries causally reduces prediction degradation in the 40-80 token range, with the benefit tied to syntactic boundary alignment rather than token density. We then test the mechanism causally: Relay-Aware Attention (RAA), which biases attention logits toward function token positions, verifiably increases attention mass by 16-24% yet yields null effects on GPT-2 and LLaMA-1B, preliminary harm on LLaMA-3B, and a mixed effect on OPT-1.3B that nets to approximately zero. Multi-fact retrieval probes further show that degradation rate does not predict retrieval accuracy across models. We conclude that mean attention degradation is largely descriptive rather than prescriptive: function tokens contribute through what their hidden states compute, not through the attention they receive -- with implications for interpretability methodology and attention-score-based inference optimisations such as KV-cache eviction. --- ### Do Active SAE Feature Planes Carry More Holonomy? A Preregistered Reversal in Gemma Authors: Larry Richards Published: 2026-07-09 Categories: cs.LG arXiv: https://arxiv.org/abs/2607.20522v1 This paper tests whether holonomy concentrates on active sparse-autoencoder (SAE) feature planes in Gemma 2 2B, a concrete operationalization of the broader semantic-concentration prediction. Holonomy is measured at the final-token layer-12 to layer-13 residual-stream readout by carrying a local frame around small loops using the instrument's restricted-Jacobian transport rule, then normalizing the resulting rotation by enclosed area. The design, materiality threshold, analysis, and verdict rules were preregistered and frozen before the analysed measurements were inspected. The prediction was falsified in reverse: active-feature planes carried less holonomy than matched mixed-feature controls, with an adjusted log contrast of -0.29439 and 95% interval (-0.43989, -0.14889). A magnitude-only explanation was not supported in this design, while the three-way ordering across random, mixed-feature, and active-feature planes was undefined at matched magnitude because common support failed. Post-freeze diagnostics at the same readout supported the area law on a small validation subset, bounded matched-center displacement under a simple paired regression, and identified transport distortion as a live mechanism or confound. The result is therefore a narrow, auditable operational reversal, not a causal claim that meaning suppresses holonomy. The cause remains open, with activation-strength geometry, degree of feature engagement, dictionary geometry, matched-center displacement, activation-manifold proximity, and transport shear as live alternatives. --- ### Generative Bayesian Filtering for State Estimation Authors: Lei Cao, Sihang Feng, Jixin Yan, Tao Sun, Naichen Shi Published: 2026-07-09 Categories: cs.LG, stat.ML arXiv: https://arxiv.org/abs/2607.20521v1 The state of a dynamic system evolves over time, switching among several latent modes that govern its observable behavior. Filtering methods infer the latent state from observations. Classical filtering approaches, including Kalman filters, typically rely on simple observation models, such as linear-Gaussian models, that are incapable of characterizing the increasingly nonlinear and heterogeneous patterns in high-dimensional sensor signals. To tackle the challenge, we propose Generative Bayesian Filtering (GBF), a filtering framework that replaces restrictive observation models with pretrained conditional generative models parametrized by conditional variational autoencoders (CVAE). For online inference, GBF performs a Bayesian prediction-update recursion in which the measurement update is formulated as a posterior sampling problem that combines the dynamical prior with the CVAE-induced likelihood. The resulting filtering problem is then transformed into a score-based sampling problem, which naturally inherits the flexibility from generative models and the uncertainty quantification capabilities from ensembling. Experiments on synthetic datasets and real-world applications involving manufacturing system monitoring and arrhythmia diagnosis demonstrate that GBF improves state estimation accuracy and robustness relative to baseline approaches. --- ### Moving Like a Human: Ego-Motion-Normalized Temporal Signatures for Real-Time Aerial Person Tracking on Milliwatt-Class Hardware Authors: Akbar Anbar Jafari, Cagri Ozcinar, Gholamreza Anbarjafari Published: 2026-07-09 Categories: cs.CV arXiv: https://arxiv.org/abs/2607.16282v1 Follow-me person tracking must run on the drone itself, where affordable companion computers offer only a few effective int8 GFLOP/s. At typical follow distances a person spans 10-60 pixels, indistinguishable from clutter and beyond the reach of single-frame appearance detectors. The missing evidence is temporal and belongs in the input representation, computed analytically, rather than in learned temporal machinery. EMTS-Det is a five-stage system that estimates ego-motion, converts each frame into ego-motion-normalized residual-motion channels, detects person centers with a 22k-parameter, 7.6-MFLOP network, tracks a locked target with a Kalman filter in stabilized coordinates, and verifies tracks with a 1-D convolutional classifier of human motion (ROC AUC 0.941). Training uses a synthetic-motion curriculum with motion channels generated by the deployed ego-motion code. Multi-seed ablations locate the value in generalization: on held-out VisDrone-DET a luminance-only variant collapses to 0.051 AP25 versus 0.415, as does YOLOv8n fine-tuned identically despite 1,100 times the compute, while the deployed int8 detector reaches 0.694 AP25 in-domain and 0.444 on this split. Temporal-shift modules lower accuracy, so the deployed detector is stateless. Silent int8 calibration failures are documented; min-max calibration with propagated caches matches float within 0.008 AP. On a Raspberry Pi Zero 2W the pipeline runs at 31.85 FPS with 0.462 AP25 and 0.714 recall over 1,000 real-world UAV videos, versus 1.95 FPS and 0.172 AP25 for YOLOv8n. A 57-second field sequence shows auto-lock at 1.3 s, 97.9% lock recall, and recovery from all nine occlusions with zero false re-locks. --- ### A Novel Hybrid Quantum Reservoir Computing (nHQRC) for Phase Transition Detection in Non-Equilibrium Dynamical Systems Authors: Manoj B. Bhatkar, Prashant M. Yawalkar Published: 2026-07-09 Categories: quant-ph, cs.LG, q-fin.CP, q-fin.ST arXiv: https://arxiv.org/abs/2607.16281v1 The analysis of highly non-linear stochastic data within non-equilibrium dynamical systems requires computational frameworks capable of detecting latent phase transitions before systemic structural breakdowns occur. Traditional Variational Quantum Algorithms (VQAs) are frequently bottlenecked by vanishing gradients, the barren plateau problem, and prohibitive training overheads. In this paper, we propose a novel Hybrid Quantum Reservoir Computing (nHQRC) framework, which bypasses these limitations by employing a frozen, disordered Transverse-Field Ising Model (TFIM) to project time-dependent stochastic driving forces into an exponentially large Hilbert space. To resolve the physical phase multi-wrapping vulnerabilities present in baseline quantum reservoir models, we introduce a lookahead-free, pre-amplification manifold scaling technique. Multi-qubit configurations are genetically optimized to the "edge of chaos," while quantum state tracking is performed by extracting von Neumann entropy ($S$) and exact mixed-state Quantum Fisher Information (QFI) to act as leading entanglement witnesses. Utilizing these quantum triggers as boundary constraints, trajectory predictions are constructed via a generative Stochastic Schrödinger Bridge (SSB) readout. By subjecting the quantum reservoir to an 8-dimensional non-stationary stochastic driving field, the framework significantly improves systemic drift-to-diffusion efficiency ($η$) and actively arrests maximum trajectory decay (MTD) by over 13% compared to standard classical benchmarks. This establishes a robust, $\mathcal{O}(1)$ temporal overhead blueprint for near-term quantum regime detection and macroscopic subsystem stabilization. --- ### 3D FaceShell: Attribute Transfer in 3D Face Avatars as a VLM Defense Mechanism Authors: Weston Bondurant, Srijan Das, Hieu Le, Stephanie Schuckers Published: 2026-07-09 Categories: cs.CV arXiv: https://arxiv.org/abs/2607.16280v1 Photorealistic 3D face avatars are increasingly deployed as reusable digital assets across applications such as telepresence, animation, and personalized media. At the same time, vision-language models (VLMs) can infer sensitive attributes from rendered images with open-ended semantic reasoning without any fine-tuning. This creates a new privacy challenge: once a 3D face avatar is shared, any of its renderings can be analyzed to extract high-level facial attributes. Existing defenses largely operate in 2D image space and do not address identity-preserving semantic manipulation of 3D facial representations. We propose 3D FaceShell, a framework for steering VLM interpretations of faces rendered from 3D models while preserving geometric fidelity and facial identity. 3D FaceShell augments the original 3D representation with a learnable Gaussian shell that produces subtle, spatially distributed perturbations optimized through multi-view embedding alignment. The perturbations are designed to be visually inconspicuous yet sufficient to redirect VLM-based attribute inference in a view-consistent manner. Extensive experiments on reconstructed celebrity face avatars and multiple black-box VLMs demonstrate that 3D FaceShell significantly increases attribute injection and mismatch rates while maintaining high perceptual similarity and identity consistency. Our results show that it is possible to manipulate VLM-level semantic interpretation of 3D faces without compromising their human-recognizable appearance. --- ### A Step Forward Towards Trustworthy Risk-Aware Facial Retrieval (RA-FR) Authors: Muhammad Emmad Siddiqui, Muhammad Rafi Published: 2026-07-09 Categories: cs.CV arXiv: https://arxiv.org/abs/2607.16279v1 Facial image retrieval in unconstrained surveillance environments is a high-stakes challenge where missing a subject of interest -- a single false negative -- is simply not an option. Despite near-perfect performance on curated benchmarks, current recognition systems falter under real-world domain shifts such as low resolution, motion blur, and uncontrolled illumination (e.g., SCFace). Addressing this reliability gap, we propose Risk-Aware Facial Retrieval (RA-FR), a framework that moves beyond fixed Top-$k$ retrieval to adaptive set generation, guaranteeing ground truth inclusion within a user-specified risk level ($α$) and confidence level ($1 - δ$). Our approach integrates three core contributions: (1) reducing aleatoric uncertainty via a hybrid blind face restoration technique coupling Latent Consistency Models (InterLCM) and DiffBIR; (2) extracting discriminative, restoration-robust features via self-supervised DINOv1 ViT-B with GGeM pooling; and (3) employing conformal prediction with Hoeffding's inequality to dynamically calibrate retrieval set sizes based on query uncertainty. On the IMFDB benchmark, it consistently satisfies a 5% risk target with an average retrieval set size of approximately 10 images. By unifying domain-specific restoration, robust representation learning, and provable decision rules, RA-FR offers a pipeline that makes facial retrieval in surveillance both reliable and auditable. The code is available at: https://github.com/MuhammadEmmadSiddiqui/RA-FR. --- ### Uncertainty-Aware Multi-Source Retinal Fluid Segmentation in OCT Authors: Animesh Kumar Published: 2026-07-09 Categories: eess.IV, cs.CV arXiv: https://arxiv.org/abs/2607.12212v2 Measuring retinal fluid from optical coherence tomography (OCT) drives treatment decisions in macular disease, but manual annotation is slow and segmentation models trained on one scanner degrade on another. We present an attention-guided TransUNet that segments three fluid types across four independent OCT sources, combining a domain-adaptive normalisation scheme with an uncertainty estimate that flags unreliable pixels. The model reaches a mean fluid Dice of 0.78, and -- most usefully for clinicians -- its uncertainty is 1.34x higher exactly where expert graders disagree (p<10^-4), turning a raw segmentation map into an actionable clinical triage signal. --- ### Calibrated Hybrid CNN-Transformer for Retinal OCT Classification Authors: Animesh Kumar Published: 2026-07-09 Categories: eess.IV, cs.CV arXiv: https://arxiv.org/abs/2607.09809v1 Deep models for retinal optical coherence tomography (OCT) classification report high accuracy but rarely report whether their confidence can be trusted -- a gap that matters when a wrong-but-confident reading delays sight-saving treatment. We pair a hybrid convolutional-Transformer encoder with a gradient-boosting (XGBoost) classification head and a three-part clinical safety layer: confidence calibration, out-of-distribution (OOD) rejection, and per-prediction uncertainty flagging. On four-class OCT (84,495 scans) the model reaches 95.4% accuracy while cutting calibration error twelve-fold (expected calibration error, ECE = 0.0024), so the confidence it reports tracks its true accuracy. To our knowledge this is the first OCT classifier to validate all three safety mechanisms jointly, with public weights and reproducible multi-seed evaluation. --- ### An Autonomous Scientific Knowledge Generation Framework for AI-Driven Scientific Discovery Authors: Dibakar Datta Published: 2026-07-09 Categories: cs.DL, cond-mat.mtrl-sci, cs.AI arXiv: https://arxiv.org/abs/2607.09806v2 Artificial intelligence (AI) is transforming scientific discovery, but its effectiveness is fundamentally limited by the availability of structured scientific knowledge. Although existing databases have accelerated data-driven materials research, much of the knowledge needed for predictive modeling and inverse design remains embedded in unstructured scientific literature. We present an Autonomous Scientific Knowledge Generation Framework that transforms scientific publications into a Unified AI-Ready Scientific Knowledge Base. The framework integrates ontology-guided literature acquisition, hybrid scientific knowledge extraction, semantic harmonization, knowledge fusion, and validation within a unified workflow. Rather than treating literature retrieval, information extraction, and database construction as separate tasks, the framework progressively converts scientific publications into structured, semantically consistent, and provenance-preserving knowledge suitable for AI-driven reasoning. As a proof of concept, the framework was applied to electro-optic materials. Autonomous literature acquisition retrieved and validated about 1,000 publications from multiple scholarly repositories. A representative subset of eight publications was processed through the complete workflow, generating 29 structured scientific records that were harmonized into 7 canonical scientific records. The results demonstrate the complete transformation from scientific literature to an AI-ready scientific knowledge base while preserving quantitative measurements, operating conditions, provenance, and scientific context. The proposed framework provides a scalable, domain-independent foundation for predictive AI, generative AI, and closed-loop AI-driven scientific discovery. --- ### A Unified Model for Highly Accurate ECG-Free Dynamic Coronary Roadmapping Using Spatio-Temporal Transformers Authors: Saahil Islam, Sebastian Piat, Venkatesh N. Murthy, Serkan Cimen, Puneet Sharma, Andreas Maier, Florin C. Ghesu Published: 2026-07-09 Categories: eess.IV, cs.AI, cs.CV arXiv: https://arxiv.org/abs/2607.09805v1 Percutaneous Coronary Intervention (PCI) is a minimally invasive procedure used to restore coronary blood flow obstructed by atherosclerotic plaque. During PCI, repeated injections of iodine-based contrast agents are required to visualize the coronary arteries and guide interventional devices. However, frequent contrast injections increase radiation exposure and the risk of contrast-induced nephropathy, with acute kidney injury reported in up to 30% of patients with renal impairment. Dynamic Coronary Roadmapping (DRM) reduces these risks by overlaying a precomputed angiographic vessel map onto live fluoroscopy and continuously updating it throughout the procedure. Accurate DRM relies on precise cardiac phase matching between angiography and fluoroscopy, together with reliable catheter tip tracking for motion compensation. These tasks remain challenging in ECG-free settings and when only limited manual annotations are available. We present a unified DRM framework that simultaneously performs cardiac phase matching and catheter tip tracking for accurate real-time guidance. Our method employs a large-scale spatio-temporal encoder pretrained on 16 million X-ray frames to learn cardiac motion dynamics. To the best of our knowledge, this is the first application of large-scale spatio-temporal pretraining for motion compensation in DRM. We further introduce auxiliary tasks based on ECG R-peak detection and catheter tip tracking, improving optimization while eliminating the need for extensive catheter mask annotations. Finally, a majority-voting postprocessing strategy aggregates temporal predictions, improving robustness and providing a confidence score that correlates with phase-matching error. Comprehensive evaluation on clinical X-ray datasets demonstrates state-of-the-art performance, achieving low temporal misalignment and robust phase-matching accuracy suitable for real-time DRM. --- ### Trivial Prompt Reframing Bypasses Safety Guardrails in Googleś MedGemma-4B Authors: Avi-ad Avraam Buskila Published: 2026-07-09 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2607.09804v1 Open-weight medical language models are increasingly used as the base of patient-facing and clinician-support applications. Their model cards prohibit specific behaviors -- recommending exact drug dosages, issuing definitive diagnoses, prescribing treatments, adjudicating drug-drug interactions, and advising that emergency care can be skipped -- yet a model card describes intended behavior, not robust behavior. We quantify that gap for MedGemma-4B-it under attacks that require no technical sophistication. We build a fully factorial benchmark of 5 guarded-behavior concepts x 50 deterministically templated questions x 6 lay-accessible attack manners x 3 repetitions (4,500 generations), serve the model locally through Ollama under default sampling, and code every response refuse/hedge/comply with three independent judges (an LLM judge, a transparent regex judge, and an NLI-entailment judge). Under the primary LLM judge the overall Attack Success Rate (ASR, the fraction coded comply) is 38.0%. The two framings that reinterpret the request as legitimate dominate: recasting a question as a "medical board exam" item raises ASR from a 29.0% baseline to 53.1% (+24.0 points), and an appeal to an alleged doctor's authority raises it to 43.7% (+14.7); crude instruction-override prefixes have no significant effect. Robustness is dominated by topic: the drug-interaction guardrail is nearly absent (83.2% ASR) while the emergency-deferral guardrail is strong (4.7%) -- and the authority framing is the only attack that breaches it. We report Wilson confidence intervals, cluster-bootstrap effect sizes, a cluster-robust logistic regression, Cochran's Q, per-manner McNemar tests, and inter-judge reliability (Fleiss' kappa = 0.26); absolute ASR is judge-dependent while the ordering of attacks and topics is not. Our findings motivate stronger deployment-time guardrails for open medical models. --- ### Spectral Origins of the Self-Correction Blind Spot in Autoregressive Generation Authors: Ingrid Petrova, Luan Vejsiu Published: 2026-07-09 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2607.09803v1 Large autoregressive language models exhibit a self-correction blind spot: they reliably fix identical errors when attributed to an external source yet fail to fix the same errors in their own outputs. Prior work has documented this phenomenon empirically, through controlled error injection, error-depth decompositions, RL-based verifier-corrector training, and intrinsic self-verification, but offers no formal model of why generating a token suppresses the ability to detect its error, no quantitative activation condition for correction markers, and no convergence guarantee for reinforcement-learning-based self-correction. We close these gaps with SPARC, a spectral-algebraic theory of self-correction in autoregressive generation. We define the error-propagation operator as the product of per-step attention Jacobians on the residual stream and prove that the blind spot arises if and only if the spectral radius of this operator is at least one. We derive a sharp activation threshold, given as a function of the spectral radius, that a correction marker must exceed, recovering the 89.3\% blind-spot reduction observed with a simple ``Wait'' marker. We further prove that RL-based verifier-corrector training converges at a rate proportional to the squared coupling strength over the square root of the number of samples if and only if the verifier-corrector coupling matrix has spectral norm below one, and that this criterion is invariant across residual-stream autoregressive modalities, unifying text LLMs and autoregressive image and video generation. Experiments across four backbones and a visual autoregressive probe validate every theorem, with spectral predictions matching measured blind-spot rates within 3.2\% RMSE. --- ### Quota Marketplace: Dynamic Pricing for Efficient Allocation of ML Training Resources Authors: Balasubramanian Sivan, Renato Paes Leme, Mihai Tiuca, Ian McFarlane, Vasilis Gkatzelis, Nehal Mehta, Soheil Hassas Yeganeh, Vahab Mirrokni, Amin Vahdat Published: 2026-07-09 Categories: cs.LG, cs.GT arXiv: https://arxiv.org/abs/2607.09802v1 The escalating demand for Machine Learning (ML) training resources in recent years has resulted in a substantial gap between the high demand and the available supply. Efficient allocation of these scarce and expensive resources is crucial for organizations to maximize their return on investment. Existing resource allocation mechanisms, like Karma (OSDI'23), are designed to guarantee Pareto efficiency and max-min fairness in settings with dynamic (time-varying) user demands, but fail to preserve these key properties in the presence of demands with heterogeneous values. Given the ubiquity and inevitability of heterogeneity in organizational values of different workloads, effective resource allocation policies must accommodate these variations. In this paper, we describe the design, implementation, deployment, and theoretical analysis of Quota Marketplace, a market-based mechanism to efficiently allocate ML training chips (like GPUs), explicitly addressing scenarios with demands of heterogeneous value. We detail the implementation of this mechanism within Google and present metrics that demonstrate its impact. We also discuss many business-critical requirements that the Quota Marketplace handles quite effectively, and document the gains and opportunities it has unlocked. We establish theoretically how this market-based approach achieves the essential properties of Pareto efficiency and max-min fairness by allowing the users to express the value of their workloads and enabling dynamic resource pricing based on supply and demand fluctuations. Ultimately, the market facilitates resource allocation that aligns with organizational priorities. --- ### Discovering Latent Response Laws in Forced Physical Systems Authors: Yi Zhu, Su Chen, Xiaojun Li, Xiuli Du Published: 2026-07-09 Categories: cs.LG arXiv: https://arxiv.org/abs/2607.09801v1 Governing equations provide compact descriptions of physical systems, yet the variables in which they are simple are often hidden in high-dimensional measurements. This challenge is sharper for forced systems, whose responses depend on both intrinsic dynamics and time-dependent inputs. Here we introduce FLARE, a forced latent autoencoder for response equations that learns compact response coordinates, identifies sparse input-dependent latent dynamics and decodes equation rollouts to full responses. By estimating latent dimension from data and separating state estimation from external forcing, FLARE enables forecasts to be initialized from past responses and driven by prescribed future inputs. Across known dynamical systems, application-scale forced responses and visual observations, FLARE recovers compact forced dynamics and predicts long-horizon high-dimensional responses under inputs not used for training. By turning learned coordinates into a dynamical interface, FLARE extends equation discovery to systems whose effective states are hidden within complex observations, providing a route for interpretable modelling and prediction of high-dimensional responses in forced dynamical systems. --- ### Auditing Invisible Weight Updates with Reference Traces Authors: Zekai Shang Published: 2026-07-09 Categories: cs.LG arXiv: https://arxiv.org/abs/2607.09800v4 Direct low-precision write-back can erase nonzero optimizer proposals. We ask what a high-precision reference trace establishes before a low-precision run. The exact target-code event is auditable coordinatewise on a realized target trajectory; pre-run aggregate projection also assumes the reference remains a useful counterfactual. In a controlled two-layer grid, 55/72 cells have measured and predicted post-initialization crossings: times span $384\times$, 52/55 are within 15\%, and 4/72 differ in category. Matched decoder experiments show stochastic rather than nearest write-back recovers most of the loss gap. A prospective analytic-grid E4M3 audit reuses one fp32 trace across three unseen NeoX-style seeds. It passes absolute-accuracy and skill gates (macro RMSE 0.00858) but fails directional specificity. In a target-outcome-blind comparison, a historical template has lower descriptive RMSE (0.00360) than the predeclared source predictor (0.00438); a post-outcome decomposition assigns 99.65\% of variation to common time, while a privileged matched-reference correction reaches 0.00283. Persistent-native Study~1 pairs three seeds across two schedules. Five cells are canonical; a manual sixth lacks canonical process identity, so the registered result remains inconclusive. A retrospective protocol-deviation analysis is negative because the complete constant-mid cohort is disjoint from the recovered cosine-restart cell. Study~2 reports mean full-SR/dead-zone-SR recoveries of 0.9766/0.9777 and a ratio of 1.0012, a policy contrast rather than causal mediation. Simulated-INT3 Study~3 replays six checkpoints and observes a 7.3071-nat (69.71\%) validation-loss reduction in one fixed seed. Exact events and write-back effects are auditable, but aggregate forecasts can reflect shared time rather than source-specific transfer. --- ### Reference Traces for Auditing Invisible Weight Updates and Guiding Exact-Budget Protection Authors: Zekai Shang Published: 2026-07-09 Categories: cs.LG arXiv: https://arxiv.org/abs/2607.09800v3 Direct low-precision write-back can erase optimizer proposals, while aggregate update visibility need not identify parameters worth protecting. We study two uses of high-precision reference traces: candidate-matched auditing before a low-precision run and regime-matched allocation before unseen targets. The exact target-code event is distinct from aggregate proxies, which are not pathwise certificates. In a controlled two-layer grid, an archived proxy tracks crossings spanning 384x in time, with median predicted/measured ratio 1.000 and 94.5% within 15%. In analytic-grid GPT-2-124M/OpenWebText runs, changing only post-optimizer write-back from round-to-nearest to stochastic rounding recovers 89.2-92.2% of the loss gap through 10,000 steps. The effect repeats in a separate modern-decoder bundle, but its visibility trajectories are uninformative. In a fully prospective 162.2M-parameter allocation test, a 600-step source trace freezes a 10% fp32-master tile mask before three unseen targets. It recovers 60.4-61.0% (mean 60.7%), versus 60.5-61.5% (mean 60.8%) for target-specific masks and 17.5-18.2% for the best fixed composition-matched controls; freeze-only recovers 0.7-0.9%. A separately frozen 197.6M-parameter depth migration recovers 65.9-67.2% (mean 66.4%). A matched systems check reduces PyTorch peak allocation by 584.65 MB versus a blanket master, but adds 65.85 MB and has 7.196% lower arm-median timed-window throughput versus direct round-to-nearest, with no validated break-even. A post hoc same-grid ECO-Adam comparator recovers 96.4-96.6%, versus 60.4-61.0% for the source mask; its systems cost is unmeasured. Thus reference traces can audit aggregate visibility and separately guide exact-budget protection, but the audit did not transfer diagnostically and the current allocator has no net systems benefit. --- ### JEPA for AI-Native 6G: Predictive Representations and Open Challenges Authors: Sheikh Salman Hassan, Irshad A. Meer, Almoatssimbillah Saifaldawla, Yan Kyaw Tun, Mustafa Ozger, Madyan Alsenwi, Nguyen Van Huynh, Woong-Hee Lee, Cedomir Stefanovic, Mathini Sellathurai, Henk Wymeersch, Tharmalingam Ratnarajah Published: 2026-07-09 Categories: cs.LG, cs.AI, cs.NI arXiv: https://arxiv.org/abs/2607.09798v1 Sixth-generation (6G) networks are moving toward AI-native operation, where learning modules are embedded across the radio access network (RAN), edge, and core. This transition requires learning from limited labels, heterogeneous wireless and network data, partial observations, non-stationary propagation, and latency-constrained control loops. Joint-embedding predictive architecture (JEPA) is a promising self-supervised paradigm for this setting because it predicts missing or future representations in latent space instead of reconstructing raw measurements or using contrastive negative samples. This article presents a wireless-oriented tutorial on JEPA for 6G intelligence. We define the JEPA training mechanism, describe how CSI, beam measurements, KPIs, topology graphs, and sensing observations can be tokenized and masked, and position the learned encoder as a predictive representation layer for RAN, O-RAN, edge, and core functions, with task-specific heads or controllers producing final decisions. Then we present an illustrative, beam-management case study suggesting that a wireless-aware target, specifically an auxiliary future beam-energy target during self-supervised pretraining, can improve label efficiency and robustness across shifted deployment conditions relative to a supervised source domain. Finally, we outline open challenges in multi-timescale prediction, action-conditioned modeling, distributed training, trustworthiness, efficient deployment, benchmarking, and standardization. --- ### Metadata-Free Meta-Reweighted Direct Preference Optimization under Noisy Preference Labels Authors: Hua Qu, Yifan Li, Xiaodong Yuan Published: 2026-07-09 Categories: cs.LG arXiv: https://arxiv.org/abs/2607.09796v2 Direct Preference Optimization (DPO) has become an important method for aligning large language models (LLMs) with human preferences because it removes the need for explicit reward modeling and reinforcement learning. However, its performance depends heavily on the quality of preference data, and noisy preference data in real-world settings can weaken alignment performance. To address this issue, we propose a bilevel optimization framework and prove, under some idealized conditions, that this framework can recover the DPO optimum under clean data. We further derive a prior form for the learnable weighting function under label-flipping noise. Considering that high-quality metadata may be difficult to obtain, we propose a prompt augmentation consistency method that enables meta-learning even when metadata is completely unavailable. To reduce the high cost of higher-order gradients in LLM meta-learning, we combine central-difference approximation with LoRA fine-tuning and develop a scalable training scheme. Experiments on TL;DR summarization and Anthropic Helpful and Harmless dialogue show that the proposed method improves alignment performance over multiple DPO baselines under different noise rates. --- ### Large Multimodal Model-Based Environment-Aware Mobility Management Authors: Seokhyun Jeong, Sangmok Shin, Seungnyun Kim, Jiao Wu, Byonghyo Shim Published: 2026-07-09 Categories: cs.IT, cs.AI, eess.SP arXiv: https://arxiv.org/abs/2607.09795v1 Recently, large language models (LLMs) have been successfully adopted in various fields, including wireless communications, robotics, and autonomous vehicles, owing to their outstanding adaptability and reasoning abilities. Despite their huge potential, the application of LLMs for mobility management is relatively scarce since it requires not only analyzing wireless measurements but also predicting dynamic user trajectories and making real-time handover decisions across densely deployed small base stations (SBSs). In this paper, we propose an environment-aware mobility management scheme based on large multimodal models (LMMs), which extend capabilities of LLMs to process multimodal sensing data. By leveraging LMMs, the proposed scheme extracts contextual information on the surrounding environments from RGB-D images to capture user equipment (UE) mobility patterns and identify signal reflections and blockages caused by static reflectors and dynamic obstacles. Using the extracted environmental information, the proposed scheme learns the intrinsic mapping from UE and SBS positions to channel capacity, referred to as channel capacity map (CCM), from which future channel capacities along UE trajectories are predicted. Based on the predicted channel capacities, we determine proactive handover decisions maximizing the cumulative channel capacities. Simulation results demonstrate that the proposed scheme achieves substantial channel capacity improvements over conventional deep learning (DL)-based approaches. --- ### Agentic Context Learning with Self-Discovered Specification Authors: Jike Zhong, Ming Li, Yuxiang Lai, Ziyan Yang, Jingyu Xie, Jihyung Kil, Zheda Mai, Shao-Yuan Lo, Ren Xiang, Konstantinos Psounis, Yuanyuan Lei Published: 2026-07-09 Categories: cs.AI, cs.MA arXiv: https://arxiv.org/abs/2607.09794v1 Context learning is an emerging inference-time task where LLMs must learn and apply novel, task-specific knowledge from intricate contexts absent from pre-training; even frontier models score under 24% task success. In this work, we conduct a comprehensive empirical study to understand why this setting remains difficult. A natural hypothesis is that failures stem from content access; yet across twelve retrieval, reflection, and verification baselines on CL-Bench, an extensive context learning benchmark, we find limited gains over direct full-context prompting. Further failure analysis reveals a key finding: unlike typical long-context tasks such as long document understanding, context learning requires not only recovering local content but also acquiring local specifications that are often unspecified in the query but distributed across the context: domain-specific formats, local rules, and completeness conditions. Across all 31,592 rubric items, we find that 55.4% clearly evaluate specification acquisition, while only 22.6% evaluate content acquisition. Moreover, despite 76.7% of specifications being unspecified in the user query, 95.5% are traceable to the context, indicating these are learnable obligations rather than hidden requirements. To validate this diagnosis, we design a deliberately simple intervention PSCI (private specification-contract induction) which extracts local specifications and enforces them through adversarial checking and repair; PSCI achieves state-of-the-art 28.14% with GPT-5.1 (+5.59 pp absolute and +24.8% relative) on CL-Bench, replicated on Qwen3.5-27B (+5.28 pp) and Gemini 3 Pro (+6.17 pp). Seventeen ablations further isolate the role of task-specific specifications. Overall, our results suggest context learning hinges on not only content acquisition but also specification acquisition. --- ### A Comprehensive Survey and Systematic Real-World Evaluation of Embodied Vision-and-Language Navigation Authors: Liuyi Wang, Kai Sheng, Zongtao He, Jinlong Li, Yongrui Qin, Haojie Dai, Xiangyi Wang, Jingwei Yang, Qingqing Yan, Chengju Liu, Qijun Chen Published: 2026-07-09 Categories: cs.RO, cs.AI arXiv: https://arxiv.org/abs/2607.09792v1 Navigation is a fundamental capability of autonomous systems, yet most existing approaches rely on highly structured models and strong prior assumptions, limiting their robustness in open and uncertain real-world environments. Vision-and-Language Navigation (VLN) offers a promising direction by enabling robots to integrate natural language understanding with visual perception in a data-driven manner. Although VLN has attracted increasing research attention, systematic methodological taxonomy and real-world validation remain limited. This survey presents a comprehensive review of VLN research. Specifically, state-of-the-art methods are organized along two orthogonal dimensions: action paradigms, including hierarchical and monolithic frameworks, and model paradigms, including discriminative and generative approaches. A critical analysis of their respective strengths and limitations is provided. Additionally, we conduct a systematic real-world evaluation of representative VLN system configurations on a physical robotic platform. Experiments across ten diverse real-world scenes show a substantial performance gap between simulation and real-world deployment under the tested configurations: a representative monolithic RGB-only method achieves 61% success in simulation but drops to 22% in real-world deployment, while a hierarchical framework achieves a higher real-world success rate of 51%, suggesting stronger robustness in our evaluation setting. Finally, we highlight key challenges in perception, decision-making, and control that must be addressed in future research. --- ### Gauge dependence and structured-output corruption in sign-branched repetition penalties: measurements across models, inference stacks, and alternative repetition controls Authors: Peter Hollows Published: 2026-07-09 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2607.09791v2 The multiplicative repetition penalty shipped across the LLM inference ecosystem (HuggingFace, vLLM, llama$.$cpp, and a dozen further engines) branches on the sign of each raw logit (divide positives by theta, multiply negatives). But the softmax is unchanged by adding a constant to every logit, so a model's logit zero-point is arbitrary (a gauge choice), and the sign-branch reads it. Two measurable consequences follow. (1) The penalty is not well-defined: re-centering a model's logits by a constant is a provable no-op at theta=1, yet at a routine theta=1.3 it changes 58-96% of greedy tokens, while subtractive and normalized penalties change none; real checkpoints sit at widely different zero-points, so a fixed repetition_penalty is a different operation on every model. (2) It corrupts structured output: on 200 real-world JSON schemas, theta=1.3 drops the rate of valid, schema-conformant output from 97% to 23%. Applying the penalty to normalized log-probabilities instead of raw logits removes the gauge dependence by construction; HuggingFace's beam search has applied its processor chain, penalty included, to log-probabilities since at least v4.0.0, so repetition_penalty already names two different operators depending on decoding strategy. Because equal theta is not equal strength across the two operators, we also compare them at matched suppression, calibrated per model by search: there the normalized operator is statistically no worse on any quality metric measured, but on four of seven models it cannot match the raw operator's suppression at theta >= 1.15, and on six of seven at theta=1.3, the setting where the corruption was measured. This note gives the mechanism, the measurements (five models up to 7B; two code models; both effects replicated inside vLLM and llama$.$cpp through their own samplers), the per-model calibration map, and the normalized variant. --- ### Gauge dependence and structured-output corruption in sign-branched repetition penalties: measurements across models, inference stacks, and alternative repetition controls Authors: Peter Hollows Published: 2026-07-09 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2607.09791v1 The multiplicative repetition penalty shipped across the LLM inference ecosystem (HuggingFace, vLLM, llama$.$cpp, and a dozen further engines) branches on the sign of each raw logit (divide positives by theta, multiply negatives). But the softmax is unchanged by adding a constant to every logit, so a model's logit zero-point is arbitrary, and the sign-branch reads that arbitrary point. The sign-branch is itself the accepted fix for an earlier bug, so the accepted fix branches on a quantity the training objective leaves unconstrained. Two measurable consequences follow. (1) The penalty is not well-defined: re-centring a model's logits by a constant is a provable no-op at theta=1, yet at a routine theta=1.3 it changes 58-96% of greedy tokens, where subtractive and normalized penalties change none; real checkpoints sit at widely different zero-points, so a fixed repetition_penalty is a different operation on every model. (2) It corrupts structured output: on 200 real-world JSON schemas, theta=1.3 drops the rate of valid, schema-conformant output from 97% to 23%. In our measurements, applying the penalty to normalized log-probabilities instead of raw logits removes both effects. HuggingFace already ships that operator (LogitNormalization); today it is off by default and applied after the penalty. This note gives the mechanism, the measurements (five models up to 7B, base and RLHF, on WikiText-103 prefixes; two code models on HumanEval and JSONSchemaBench; both effects replicated inside vLLM and llama$.$cpp through their own samplers on the same inputs), and the normalized variant. --- ### Model Agnostic Graph Prompt Learning for Crystal Property Prediction Authors: Shrimon Mukherjee, Kishalay Das, Partha Basuchowdhuri, Pawan Goyal, Niloy Ganguly Published: 2026-07-09 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2607.08996v1 Graph Neural Networks have emerged as a powerful tool for the fast and accurate prediction of various crystal properties. These models often encode domain-specific knowledge into their graph encoding modules, which increases their parameter size and makes their performance heavily dependent on domain expertise. Added to this, explicitly incorporating all chemical and structural features, that might influence a specific crystal property into the GNN encoder, is a challenging task. In this work, we propose a soft prompt learning framework that captures latent features essential for property prediction, which are not explicitly provided to the GNN. We introduce a novel multilevel graph prompt learning framework comprising both node-level and graph-level soft prompts. At the node level, we capture the local chemical semantics of different atom types, while at the graph level, we encode the global structural symmetry of the crystal graph. Our proposed prompt learning framework is lightweight and seamlessly integrates with any existing GNN encoder. Extensive experiments on popular benchmark datasets show that incorporating prompt learning significantly improves (3\% - 15\%) the performance of state-of-the-art GNN models in crystal property prediction tasks. Furthermore, the learned soft prompts enable cross-property knowledge transfer, enhancing prediction performance for properties with limited training data. Code is available at https://github.com/shrimonmuke0202/Prompt.git --- ### Sensitivity-Aware Thresholding and Token Routing for Activation Sparsification in Large Language Models Authors: Bishmoy Paul, Youngmin Yi, Hoeseok Yang Published: 2026-07-09 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2607.08991v1 Efficient inference in Large Language Models (LLMs) requires deciding where computation can be reduced while preserving model quality. We study this problem through multilayer perceptron (MLP) activation sparsification and token-level conditional routing. We first propose Sensitivity-Aware Thresholding for Sparsity (SATS), a threshold calibration method to choose layerwise gate thresholds using a local MLP output sensitivity proxy rather than calibrating thresholds directly from activation percentiles. While SATS retains the existing mechanism of sparsifying MLP activations by thresholding gate activations, it replaces percentile-based calibration with a sensitivity-aware selection rule. We then introduce a lightweight token routing framework that dynamically selects between a base path and a modified path on a per-token basis, rather than applying the modified computation uniformly to all tokens. We evaluate both methods on multiple recent open-weight LLMs. Our results show that SATS improves over the threshold-based sparsification baseline at matched actual sparsity and that token routing yields a more favorable quality-throughput trade-off than static activation modification baselines. Overall, our results suggest that improved threshold calibration and token routing can improve the quality-throughput trade-off in LLMs. --- ### Group Invariant Spectral Embedding Authors: Yeari Vigder, Paulina Hoyos, David Thong, Joakim andén, Joe Kileel, Amit Moscovich Published: 2026-07-09 Categories: cs.LG, math.NA, math.ST arXiv: https://arxiv.org/abs/2607.08987v1 Spectral embedding methods are widely used for dimensionality reduction and clustering of high-dimensional datasets with intrinsic low-dimensional structures. Although many datasets of practical interest exhibit invariance under symmetries such as rotations, standard spectral embedding methods do not account for this, treating symmetry-related data points as unrelated. Our approach to this problem is to incorporate the symmetries directly into the affinity kernels used for spectral embedding. We analyze the case of a Riemannian data manifold $M$ with symmetries given by a compact Lie group~$G$ and prove that, under suitable conditions, graph Laplacians constructed from three types of invariant kernels converge pointwise to explicit second-order differential operators on the quotient space $M/G$. Our analysis implies improved convergence rates, as the effective dimension drops according to the dimension of the group. We validate our approach on datasets with $\mathrm{SO}(2)$ or $\mathrm{SO}(3)$ symmetry, and show that $G$-invariant spectral embedding recovers the intrinsic geometry of the data, in contrast to standard spectral embedding, which fails to do so even in the limit of infinite data. --- ### A Formalization of the Mean-Field Derivation of the Vlasov Equation Authors: Joseph K. Miller Published: 2026-07-09 Categories: cs.AI, cs.LO, math-ph, math.AP arXiv: https://arxiv.org/abs/2607.08986v2 We formalize a research result in the Lean 4 proof assistant by having a mathematician direct an AI system, and frame the activity as a formalization game. The objective is to turn a LaTeX document into Lean. The game is won when the development compiles, contains no sorry, and a machine check shows the target theorems rest on Lean's foundational axioms alone. Reuse is a second check, by a definition we introduce: whether the development yields a self-contained layer of general mathematics the wider library could absorb. The case study is a complete, axiom-clean formalization of well-posedness for the nonlinear Vlasov equation via Dobrushin's mean-field route -- existence, uniqueness, the stability estimate and mean-field limit, and a short-window superposition principle (weak solutions are Lagrangian). The human's role was to direct, not to write proofs: to scope the definitions, steer the decompositions, and triage the library's gaps; the AI agent executed. The formalization certifies the proof of each statement as written; whether the written statement is the intended theorem stays the mathematician's judgment. The optimal-transport machinery that fell out of the build (in particular, properties of the Wasserstein-1 metric and the Kantorovich-Rubinstein duality theorem) separates into a self-contained layer that compiles against Mathlib alone: about a sixth of the development (49 of 299 declarations), behind a 22-declaration interface with no reverse dependency. The headline theorems ran in about a week, the full development in about a month. We report the quantitative claims as observations of one game, not as general laws. The game's rules name no particular system, so the methodological framing is meant to outlast the tools of any one run. --- ### AlphaZero in Sparsely Rewarded Games: Limits and Auxiliary Supervision Authors: Brent Kong, Tejas Ram, Tony Yue Yu Published: 2026-07-09 Categories: cs.LG, cs.AI, cs.GT, math.CO arXiv: https://arxiv.org/abs/2607.08984v1 AlphaZero has demonstrated that a neural-guided Monte Carlo Tree Search can achieve superhuman performance, but strong play does not necessarily imply perfect play. We study this gap in two oracle-evaluable domains with contrasting structure: Connect Four, a solved partisan game with exact game-theoretic values, and Chomp, an impartial game whose optimal play is governed by Grundy-number structure. Under a unified self-play $+$ MCTS pipeline, we compare vanilla AlphaZero, a multi-frame variant (limited to Chomp), and an AlphaZero Auxiliary Loss (AZAL) that adds oracle-derived policy supervision. We find that vanilla AlphaZero achieves strong play across both domains but cannot preserve the exact trajectories required for optimal play: in Connect Four, it fails to maintain the optimal line of play, while in Chomp, it fails to consistently restore the $g=0$ invariant. On rectangular Chomp boards, multi-frame inputs alone do not remove this gap. Nevertheless, AZAL substantially improves oracle consistency across multi-seeded full-game traces and sampled-state evaluations. On Chomp, AZAL reaches perfect full-game oracle consistency on 10x11 and high but not complete consistency on 9x10; on Connect Four, AZAL improves oracle-match rate and delays the first oracle mistake, but does not reach perfect play. --- ### SCATE: Learning to Supervise Coding Agents for Cost-Effective Test Generation Authors: Sijia Gu, Noor Nashid, Ali Mesbah Published: 2026-07-09 Categories: cs.SE, cs.AI arXiv: https://arxiv.org/abs/2607.08983v1 While autonomous coding agents have significantly advanced automated test generation, they remain fundamentally limited by lazy generation, a phenomenon where agents prematurely terminate tasks and systematically avoid complex programmatic logic, resulting in inadequate code coverage. Currently, mitigating this premature termination requires continuous human-in-the-loop supervision. This heavy reliance on human intuition creates a bottleneck that negates the efficiency gains of automated generation. We propose SCATE, a framework for adaptive, automated supervision of coding agents that replaces human intervention during test generation. By formulating supervision as a contextual bandit problem, SCATE learns to select the most promising testing actions based on the current coverage and class testability metrics, maximizing coverage gains while minimizing wasted generation effort. Our empirical evaluation demonstrates that SCATE integrates seamlessly with different coding agents. When applied to GEMINI-CLI, it achieves 32.3% higher line coverage and 30.9% higher branch coverage than the agent-only baseline. A comparison with CLAUDE CODE confirms the framework dynamically adapts its policy to optimize each agent's unique strengths. SCATE also consistently outperforms state-of-the-art non-agentic approaches across all metrics. --- ### The Patchwork Problem in LLM-Generated Code Authors: Viraaji Mothukuri, Reza M. Parizi Published: 2026-07-09 Categories: cs.SE, cs.AI arXiv: https://arxiv.org/abs/2607.08981v1 LLM-generated code often compiles, passes tests, and appears correct, yet breaks once deployed. The root cause is frequently structural rather than logical. A generated endpoint references configuration keys never declared in the project, an import targets a package that does not exist in any registry, or a new route omits the authentication guard applied to every sibling endpoint. Each patch is locally valid but globally incoherent, and standard CI toolchains rarely surface these failures. As LLM-powered coding tools see widespread adoption, this blind spot poses a growing risk to software quality. We call this the \textbf{patchwork problem}. This paper formalizes structural coherence as consistency invariants over graph representations of repository artifacts, including import, call, dependency, configuration, schema, resource, control-flow, and routing graphs, and introduces an eight-category failure taxonomy distinguishing defects specific to LLM generation from those merely amplified by it. We present a hybrid verification framework that delegates to mature static analysis tools where they already excel and deploys purpose-built detectors for cross-cutting invariants underserved by existing toolchains, targeting provable constraint violations rather than heuristic pattern matching. Empirical evaluation across two frontier models under four prompting strategies reveals that the vast majority of structural failures evade type checking, testing, and SAST entirely, and that failure patterns diverge qualitatively between models in ways that challenge model-agnostic mitigation strategies. External validation on real-world AI-generated repositories confirms that these failures are not artifacts of controlled experimentation but are prevalent wherever LLMs write code with minimal human oversight. --- ### Optimal Top-$k$ Identification from Pairwise Comparisons Authors: Motti Goldberger, Nils Rudi Published: 2026-07-09 Categories: cs.LG, stat.AP, stat.ML, stat.OT arXiv: https://arxiv.org/abs/2607.08979v1 We study the active learning problem of fixed-confidence top-$k$ identification from noisy pairwise comparisons. In this problem, an algorithm sequentially chooses pairs of items to compare, observes the outcomes, and stops when it can return the set of top-$k$ items with error probability at most $δ$. The objective is to design such a $δ$-correct procedure that minimizes the expected number of comparisons (the sample complexity). This problem falls within the broader literature on fixed-confidence pure exploration in bandit models, where a common target is asymptotic optimality: the algorithm's expected sample complexity matches the information theoretic lower bound as $δ\to 0$. Asymptotically optimal procedures have been developed for a range of fixed-confidence pure-exploration problems, however to the best of our knowledge, for top-$1$, or more generally top-$k$ identification from pairwise comparisons under latent utility models an asymptotically optimal algorithm has not been established. In this setting, we develop such an algorithm. We characterize the structure of the lower bound and formulate it as a saddle-point problem. This structure enables a computationally efficient primal-dual procedure that learns the asymptotically optimal comparison allocation online. We then construct an adaptive comparison-allocation algorithm that tracks the allocation learned by the primal-dual procedure and prove it is asymptotically optimal. --- ### Federated Low-Rank Koopman Learning for Multivariate Time-Series Anomaly Detection in IoT Systems Authors: Tung-Anh Nguyen, Van-Phuc Bui, Anh Tuyen Le, Kim Hue Ta, Minh Thuy Le, J. Andrew Zhang, Xiaojing Huang Published: 2026-07-09 Categories: cs.LG, eess.SP arXiv: https://arxiv.org/abs/2607.08978v1 Distributed IoT systems generate multivariate time-series streams for monitoring physical assets, servers, and embedded sensing platforms. Detecting abnormal temporal behavior is critical for fault diagnosis, predictive maintenance, and security. However, practical IoT anomaly detection is hindered by decentralized and non-IID data, limited bandwidth, and the constrained computation and memory of edge devices. This paper proposes FedKAD, a resource-efficient federated Koopman anomaly detection framework for distributed IoT multivariate time series. Unlike deep-learning-based anomaly detectors that require training and communicating large neural models, FedKAD learns normal temporal dynamics through lightweight sliding-window Koopman representations. Federated training is formulated as a low-rank consensus problem, where raw sensor streams and local reduced dynamics remain on device while only compact subspace variables are exchanged with the server. To optimize the shared representation under orthonormality constraints, we develop a federated Stiefel-ADMM algorithm and provide convergence and stationarity analysis under partial client participation. During inference, each client detects anomalies locally by measuring the prediction residual between observed future trajectories and the learned Koopman dynamics. Experiments on four widely used multivariate time-series anomaly detection benchmarks show that FedKAD maintains or improves detection performance compared with federated deep-learning baselines. More importantly for IoT deployment, FedKAD provides up to $2.1\times10^3$ faster training, $80\times$ lower communication, and $79\times$ lower inference latency than neural baselines, confirming its suitability for resource-constrained edge devices. --- ### CLAP: Direct VLM-to-VLA Adaptation via Language-Action Grounding Authors: Yuri Ishitoya, Jeremy Siburian, Masashi Hamaya, Kuniaki Saito, Cristian C. Beltran-Hernandez, Mai Nishimura Published: 2026-07-09 Categories: cs.RO, cs.AI arXiv: https://arxiv.org/abs/2607.08974v1 Vision-language-action models (VLAs) inherit semantic capabilities from pretrained VLMs, yet large-scale post-training on robot data and architectural modifications can reshape the backbone so extensively that it becomes difficult to isolate what the VLM contributes to control. Directly converting pretrained VLMs into VLAs with minimal architectural change offers a more transparent path to understanding how VLM capabilities transfer across model scales. The core obstacle is output-distribution mismatch: predicting actions as bare numeric token sequences moves generation away from the VLM's pretrained language distribution, degrading the capabilities we seek to preserve. To address this, we propose CLAP (Causal Language-Action Prediction), which prepends each numeric action sequence with a natural-language action description, causally conditioning precise action-token prediction on a language-action plan without modifying the backbone architecture. With single-epoch fine-tuning alone, 2B CLAP achieves 90.8% on LIBERO (+14.9 pt over VLA-0) and improves robustness on LIBERO-PRO under language, object, and spatial perturbations. We will release CLAP at 0.8B, 2B, and 4B as an open-weight, multi-scale compact VLA family from a single VLM lineage, enabling controlled analysis of VLM-to-VLA capability transfer. --- ### Stochastic Linear Bandits with Partially Observed Actions Authors: Gautam Dasarathy, Vineet Gattani, Lalit Jain Published: 2026-07-09 Categories: cs.LG, math.ST, stat.ML arXiv: https://arxiv.org/abs/2607.08971v1 The stochastic linear bandit, where actions are represented as vectors and rewards are linear, is a central paradigm for sequential decision making. We study a partially observed variant of this problem in which the learning agent only sees a random subset of coordinates for each action. Such partial observability arises naturally in settings like recommendation and healthcare, where full action descriptions can be expensive or even impossible to obtain. In general, this makes sublinear regret information-theoretically impossible. However, we show that this barrier can be overcome when the action vectors have low intrinsic dimension. We propose an algorithm, TOFU-POV, that estimates the latent action subspace using the masked actions, imputes current actions using an epoch-wise frozen representation, and runs OFUL in the resulting low-dimensional coordinates. Our theory shows that TOFU-POV enjoys a $\sqrt{T}$ regret that scales with the intrinsic action subspace dimension as opposed to the ambient dimension and quantifies the interaction between these quantities and the missingness, decision set size, and subspace conditioning. We also devise a rank-adaptive algorithm that does not require the knowledge of the intrinsic dimension. We complement these guarantees with a lower bound based on a novel product construction that separates usual reward-learning uncertainty from a missingness-dependent cost intrinsic to partial observation. Synthetic and real data experiments support our theory and show that TOFU-POV can substantially improve upon natural baselines in this challenging problem. ---