Academic Intelligence · Curated Daily

探索全球前沿学术脉络

AcademicHub 汇聚顶级期刊与预印本平台的实时文献。定制您的专属科研雷达,利用大语言模型自动生成交叉领域文献分析简报。

01.
arXiv (quant-ph) 2026-06-16

On-chip semi-device-independent quantum random number generator exploiting contextuality

arXiv:2601.08392v2 Announce Type: replace Abstract: We present a semi-device-independent quantum random number generator (QRNG) based on the violation of a contextuality inequality, implemented by the integration of two silicon photonic chips. Our system combines a heralded single-photon source with a reconfigurable interferometric mesh to implement qutrit state preparation, transformations, and measurements suitable for testing a KCBS contextuality inequality. This architecture enables the generation of random numbers from the intrinsic randomness of single-photon interference in a complex optical network, while simultaneously allowing a quantitative certification of their security without requiring entanglement. We observe a contextuality violation exceeding the classical bound by more than 10{\sigma}, unambiguously confirming non-classical behavior. From this violation, we certify a conditional min-entropy per experimental round of Hmin = 0.077 +- 0.002, derived via a tailored semidefinite-programming-based security analysis. Each measurement outcome therefore contains at least 0.077 +- 0.002 bits of extractable genuine randomness, corresponding to an asymptotic generation rate of 21.7 +- 0.5 bits/s. These results establish a viable route towards general-purpose, untrusted quantum random number generators compatible with practical integrated photonic quantum networks.

02.
arXiv (CS.AI) 2026-06-12

SMSR: Certified Defence Against Runtime Memory Poisoning in Persistent LLM Agent Systems

作者:

arXiv:2606.12703v1 Announce Type: cross Abstract: Retrieval-augmented generation (RAG) agents increasingly run with persistent memory that accumulates across user sessions. This creates a new attack surface: an adversary interacting only through normal channels can inject crafted memories that, once retrieved, steer the agent's responses for future users, without touching model weights or code. We call this Multi-Session Memory Poisoning (MSMP) and show that no existing defence certifies against it; static-corpus defences (RobustRAG, ReliabilityRAG) assume a fixed knowledge base, and heuristic filters are bypassed by fluent enterprise-style text. We present Signed Memory with Smoothed Retrieval (SMSR), the first defence with a certified robustness bound for this setting. Component 1 adds HMAC-SHA256 provenance at write time, blocking unsigned injection. Component 2 applies randomised memory ablation with verdict-based majority voting at query time, bounding the influence of authenticated adversaries. We prove that no provenance-free retrieval-time filter can certify against adaptive injection, derive a hypergeometric certificate for Component 2, and formalise the Consistent Minority Effect, whereby a consistent adversarial answer wins string-based voting as a numerical minority while verdict-based voting removes it. Across 15 enterprise scenarios (3,150 repeated trials), Component 1 cuts attack success from 93-100% to 0% for all unsigned variants. For an authenticated adversary with a single injection, Component 2 holds success to 8.0% (95% CI [5.8, 10.9], n=450), below the certified worst case. In an end-to-end query-only attack where the agent itself writes the poison rather than it being pre-seeded, SMSR reduces success from 65.3% to 5.3% (n=150, non-overlapping CIs) on a live agent stack. Clean-query utility is 90% (Component 1) and 85% (combined).

03.
arXiv (CS.CL) 2026-06-15

Is ChatGPT Fair for Recommendation? Evaluating Fairness in Large Language Model Recommendation

The remarkable achievements of Large Language Models (LLMs) have led to the emergence of a novel recommendation paradigm – Recommendation via LLM (RecLLM). Nevertheless, it is important to note that LLMs may contain social prejudices, and therefore, the fairness of recommendations made by RecLLM requires further investigation. To avoid the potential risks of RecLLM, it is imperative to evaluate the fairness of RecLLM with respect to various sensitive attributes on the user side. Due to the differences between the RecLLM paradigm and the traditional recommendation paradigm, it is problematic to directly use the fairness benchmark of traditional recommendation. To address the dilemma, we propose a novel benchmark called Fairness of Recommendation via LLM (FaiRLLM). This benchmark comprises carefully crafted metrics and a dataset that accounts for eight sensitive attributes1 in two recommendation scenarios: music and movies. By utilizing our FaiRLLM benchmark, we conducted an evaluation of ChatGPT and discovered that it still exhibits unfairness to some sensitive attributes when generating recommendations. Our code and dataset can be found at https://github.com/jizhi-zhang/FaiRLLM.

04.
arXiv (CS.CV) 2026-06-11

OpenMedReason: Scientific Reasoning Supervision for Medical Vision-Language Models

High-stakes clinical use of large vision-language models (LVLMs) requires reasoning that is grounded in visual evidence and clinical knowledge, not just correct final answers. We introduce OpenMedReason, a large-scale, open multimodal medical reasoning corpus comprising approximately 450K image-question-answer instances whose reasoning traces are primarily derived from curated biomedical, human-authored scientific articles. OpenMedReason provides high-fidelity supervision beyond synthetic chains of thought, covering diverse medical domain vision modalities such as radiological scans, microscopic images, visible light photographs, charts, and others. We complement it with OpenMedReason-Bench, a held-out benchmark that allows fine-grained evaluation of LVLMs along three complementary axes of capability, including perception, medical knowledge, and rationale, enabling diagnostic evaluation beyond final-answer accuracy. OpenMedReason is a rich training resource that exhibits its effectiveness in both supervised fine-tuning (SFT) and reinforcement-based alignment. Training with OpenMedReason yields a 20% average improvement in VQA accuracy over the base model and achieves performance within 4.2% of the strongest comparable-scale medical LVLMs. Fine-grained performance analysis confirms that the gains are not concentrated in any single axis: OpenMedReason improves perception, medical knowledge, and rationale jointly, and its reasoning traces are preferred over those of the base model in 86.1% of pairwise comparisons. We release the code and dataset at huggingface.co/datasets/neginb/OpenMedReason.

05.
arXiv (CS.LG) 2026-06-17

Statistical Learning from Attribution Sets

arXiv:2602.06276v2 Announce Type: replace Abstract: We address the problem of training conversion prediction models in advertising domains under privacy constraints, where direct links between ad clicks and conversions are unavailable. Motivated by privacy-preserving browser APIs and the deprecation of third-party cookies, we study a setting where the learner observes a sequence of clicks and a sequence of conversions, but can only link a conversion to a set of candidate clicks (an attribution set) rather than a unique source. We formalize this as learning from attribution sets generated by an oblivious adversary equipped with a prior distribution over the candidates. Despite the lack of explicit labels, we construct an unbiased estimator of the population loss from these coarse signals via a novel approach. Leveraging this estimator, we show that Empirical Risk Minimization achieves generalization guarantees that scale with the informativeness of the prior and is also robust against estimation errors in the prior, despite complex dependencies among attribution sets. Simple empirical evaluations on standard datasets suggest our unbiased approach significantly outperforms common industry heuristics, particularly in regimes where attribution sets are large or overlapping.

06.
arXiv (CS.AI) 2026-06-11

A Five-Plane Reference Architecture for Runtime Governance of Production AI Agents

作者:

arXiv:2606.12320v1 Announce Type: new Abstract: Enterprise security was built to govern data boundaries: the protected surface was data at rest and in transit, and the controls – access control, data-loss prevention, perimeter inspection – governed crossings of that boundary. Production AI agents dissolve this assumption. An agent reads context, calls tools, invokes connectors, and modifies systems of record on an enterprise's behalf, so risk moves inside the workflow, into sequences of individually-permitted actions that may transform a business process no one authorized. Existing policy engines do not extend to this regime: they evaluate request-time decisions against atomic principals, where agentic systems require stateful evaluation against composite principals whose authority attenuates through delegation chains. We present a reference architecture for the runtime governance of production agents, built from four composable primitives: a five-plane decomposition (a reasoning plane that adjudicates intent, and four enforcement planes – network, identity, endpoint, data – that realize the decision), stop-anywhere mediation, composite principals with capability attenuation, and audit as a structured evidence substrate. We define a taxonomy of six interruption primitives that generalize allow and deny, state and argue for four correctness invariants, and demonstrate the foreclosure of seven production-agent threats across five concrete workflows. A reference implementation of the policy-engine core supplies measured evidence: attenuation correctness and evidence reconstructability hold on every trial, adjudication runs in single-digit microseconds, and the audit substrate's tamper-evidence behaves exactly as designed. We are explicit about scope: the architecture governs delegated action, not model behavior, and a full-system evaluation against a live agent benchmark is the invited next step.

07.
medRxiv (Medicine) 2026-06-11

Foundation model-based tool for automated ulcerative colitis histology scoring demonstrates non-inferiority to pathologists across multiple scoring indices

In clinical trials for ulcerative colitis (UC), pathologists assess disease severity through standardized histological indices, including the Geboes Score, Robarts Histopathology Index (RHI), and Nancy Histologic Index (NHI). Despite strong associations with clinical outcomes, histologic scoring suffers from inter- and intra-reader variability, and consensus criteria for histologic remission remain uncertain. Through a consortium approach, we developed an artificial intelligence-based measurement (AIM) tool for scoring histology in UC mucosal biopsies (AIM-HI UC). This model, trained on a large dataset of UC biopsies (N=10,230), utilizes additive multiple instance learning models leveraging PLUTO, a pathology foundation model, that predict each of the Geboes subgrades, from which the Geboes grade-level score, RHI, and NHI can be calculated. Evaluation of this model on a standalone verification set including clinical trial specimens established algorithm non-inferiority and/or superiority relative to standard qualified pathologists through comparison of algorithm-consensus and pathologist-consensus agreement metrics (non-inferior if difference >-0.1, superior if difference >0, inclusive of confidence intervals). AIM-HI UC was determined to be non-inferior to pathologists (N=3) for the prediction of all seven Geboes subgrades, grade-level Geboes, RHI, NHI, histologic improvement (GS

08.
arXiv (math.PR) 2026-06-16

Sharp freezing time estimates for the subcritical Facilitated Exclusion Process

arXiv:2606.15233v1 Announce Type: new Abstract: We investigate the exact transience time of the Facilitated Exclusion Process (FEP) on the one-dimensional torus with $N$ sites. The FEP exhibits an active/inactive phase transition at critical density $1/2$, such that in the subcritical density regime $(0,1/2)$, it becomes frozen after a finite time period – the transience time or freezing time. We first show that for the FEP starting from a Bernoulli product measure of marginal density $\rho \in (0,1/2)$, the transience time has exactly the scale of $\Theta(\log^3 N)$. Secondly, we prove that in the near-critical case $\rho \simeq 1/2 - N^{-\alpha}$ for $\alpha \in (0,1)$, the transience time is polynomial and has a scale of $N^{1 \wedge (2\alpha)}$. The key idea is to estimate the typical size of locally supercritical intervals of the initial distribution, which has order $\log N$ in the subcritical case and $N^{1 \wedge (2\alpha)}$ in the near-critical case. In the subcritical case this is enough, whereas in the near-critical case we need additional dynamical decorrelation inequalities to apply this static result to estimate the freezing time.

09.
arXiv (CS.CV) 2026-06-11

Reroute, Don't Remove: Recoverable Visual Token Routing for Vision-Language Models

Vision-language models (VLMs) project images into hundreds to thousands of visual tokens, making decoder inference expensive in both attention computation and KV-cache memory. Existing visual-token reduction methods largely follow a rank-and-remove paradigm: they score visual tokens, keep a compact subset, and permanently discard the rest. We show that this irreversible action is fragile because visual-token importance changes across decoder depth; tokens ranked low at one stage may become relevant in later layers, especially for grounding-sensitive queries. We propose Reroute, a training-free plug-in that replaces removal with recoverable routing. At each routing stage, selected vision tokens pass through decoder blocks, while deferred tokens bypass the stage and re-enter the candidate pool at the next routing decision. Reroute reuses existing attention-score ranking rules and stage-wise schedules, preserving the theoretical TFLOPs and KV-cache budget class of the pruning method it augments. Across FastV, PDrop, and Nüwa variants on LLaVA-1.5 and Qwen backbones, reroute improves grounding under aggressive token reduction while maintaining general VQA performance. These results suggest that VLM token reduction should not be viewed only as irreversible pruning, but also as recoverable routing. The code can be found here: https://github.com/elmma/mllm-reroute/

11.
arXiv (CS.AI) 2026-06-19

Repurposing a Speech Classifier for Guided Diffusion-Based Speech Generation

arXiv:2606.20457v1 Announce Type: cross Abstract: Classifier guidance is a way to control diffusion generation by using a noise-conditioned classifier to steer the sampling process toward a target class. One drawback of classifier guidance is that it requires two separately trained models: a classifier and a diffusion model. We therefore study a more compact alternative in which a conventionally trained speech classifier is repurposed as the backbone for diffusion generation. Starting from a frozen noise-conditioned classifier in log-Mel space, we attach a lightweight subnetwork that reuses intermediate classifier representations and train only this subnetwork under a Denoising Score Matching objective. Our work shows that a pretrained classifier can be repurposed for conditional generation, providing an appealing bridge between discriminative modeling and conditional speech synthesis resulting in high speech quality within a single-backbone model, with reduced memory footprint and computational cost.

12.
arXiv (CS.CL) 2026-06-16

RoSE: Round-robin Synthetic Data Evaluation for Selecting LLM Generators without Human Test Sets

LLMs are powerful generators of synthetic data, which are used for training smaller, specific models. This is especially valuable for low-resource languages, where human-labelled data is scarce but LLMs can still produce high-quality text. However, LLMs differ in how useful their outputs are for training. Selecting the best LLM as a generator is challenging because extrinsic evaluation requires costly human annotations (which are often unavailable for low-resource languages), while intrinsic metrics correlate poorly with downstream performance. We introduce Round robin Synthetic data Evaluation (RoSE), a proxy metric for selecting the best LLM generator without human test sets. RoSE trains a small model on the outputs of a candidate generator (LLM) and then evaluates it on generated synthetic examples from all other candidate LLMs. The final RoSE score is the mean performance of this small model. Across six LLMs, eleven languages, and three tasks (sentiment, topic, intent), RoSE identifies the optimal generator more often than any other intrinsic heuristics. RoSE outperforms intrinsic heuristics and comes within 0.76 percentage points of the optimal generator baseline. This result is measured in terms of downstream performance, obtained by training a small model on the chosen generator's outputs (optimal vs. proxy metric selected) and evaluating it on human-labelled test data. Additionally, RoSE is the only metric to achieve a positive correlation with performance on human test data.

13.
medRxiv (Medicine) 2026-06-17

Determinants of non-utilization of insecticide-treated nets among children under five in Rwanda: analyses of the 2024 Rwanda malaria indicator survey

Background Insecticide-treated nets (ITNs) are effective for preventing malaria among children under five years, who bear a disproportionate burden of malaria. This study assessed the prevalence and determinants of ITN non-utilization among children under five in Rwanda using data from the 2024 Rwanda Malaria Indicator Survey (RMIS).Methodology This cross-sectional study utilized nationally representative data from the 2024 RMIS. Analyses were restricted to children under five residing in households that owned at least one ITN. The outcome was non-utilization of ITN, defined as not sleeping under an ITN the night preceding the survey. Survey-weighted descriptive statistics were used to estimate the prevalence of ITN non-utilization. Factors associated with non-utilization were identified using a survey-weighted Poisson regression model. Adjusted prevalence ratios (aPRs), 95% confidence intervals and p-values were reported.Results A total of 1,979 children were included in the study. The weighted prevalence of ITN non-utilization among children under five years was 20.11% (95% CI: 17.81 - 22.63). After adjusting for other factors, children aged 2 - 3 years were associated with an 83% higher prevalence of ITN non-utilization compared with those aged [&le;]1 year (aPR = 1.83, 95% CI: 1.423 - 2.352, p < 0.001). Compared with households that owned only one ITN, children in households with three or more ITNs were associated with a 76% lower prevalence of ITN non-utilization (aPR = 0.24, 95% CI: 0.171 - 0.332, p < 0.001). Children living in households with 5 - 7 members were associated with an 87% higher prevalence of ITN non-utilization compared with those in households with 1 - 4 members (aPR = 1.87, 95% CI: 1.476 - 2.358, p < 0.001).Conclusion The findings suggest that ITN utilization among children is influenced not only by household access to nets but also by household composition and dynamics that shape the allocation and use of available preventive resources.

14.
arXiv (CS.CV) 2026-06-16

Simulation-Based Multi-Fillet Evaluation of Woody Breast Poultry Fillets

Woody breast (WB) is a myopathy in modern broiler chickens that causes the breast muscle to become unusually stiff and fibrous, leading to decreased meat quality and significant economic losses. State-of-the-art automated WB detection relies on a side-view imaging system to analyze the bending behavior of a single fillet as it falls off a conveyor belt. While highly accurate, this approach is constrained by its single-fillet field of view, creating throughput bottlenecks on commercial processing lines. In this paper, we address this limitation via a novel multi-fillet detection architecture utilizing a top-down camera configuration. To validate our approach, we first develop a high-fidelity digital twin of an industrial conveyor system. Next, we synthesize a diverse dataset of 3D fillet meshes and model their viscoelastic bending dynamics using a physics-based simulation engine. Lastly, a continuous 2D shape deformation score is extracted from the top-down perspective as the simulated fillets traverse the roller precipice. Experimental results demonstrate that the top-down shape score effectively captures the contour changes of the fillets as it bends, providing a robust and scalable alternative to a side-view imaging system for simultaneous multi-fillet WB evaluation.

15.
medRxiv (Medicine) 2026-06-16

Enteral docosahexaenoic and arachidonic acid supplementation and retinopathy of prematurity: a re-analysis of randomized controlled trials in preterm infants

Background. A recent meta-analysis by Dang et al. [1] concluded that enteral supplementation with docosahexaenoic acid (DHA), with or without arachidonic acid (ARA) did not significantly affect retinopathy of prematurity (ROP) outcomes in preterm infants. Of four eligible trials that supplemented both DHA and ARA, only two contributed to each ROP outcome analyzed, and severe ROP was not assessed. Methods. We replicated the eligibility criteria and search strategy of Dang et al., restricted to trials that supplemented both DHA and ARA, and reanalyzed three ROP endpoints (any ROP, ROP requiring treatment, and severe ROP [stage 3 and/or treated]) using complete outcome records from all eligible trials. Crude risk ratios (RR) were pooled by Mantel-Haenszel fixed-effect meta-analysis. Gestational age-adjusted odds ratios (adjOR) were pooled on the log scale by inverse-variance random-effects meta-analysis with restricted maximum likelihood (REML) estimation of between-study variance and Hartung-Knapp confidence intervals. Results. Five trials were included; one trial was identified in our replicated search but was excluded by Dang et al. without a stated rationale. The pooled estimate for any ROP was consistent with Dang et al. (RR 0.87 [95% CI 0.71-1.08]; adjOR 0.70 [0.46-1.08]). For ROP requiring treatment, the crude RR suggested a lower risk but did not reach statistical significance (RR 0.60 [0.35-1.04]), whereas the gestational age-adjusted estimate indicated lower odds (adjOR 0.47 [0.23-0.94]). For severe ROP, DHA+ARA supplementation produced a significant protective effect in both unadjusted and adjusted models (RR 0.56 [0.36-0.86]; adjOR 0.42 [0.19-0.96]). Conclusions. When all eligible trials contribute to each endpoint and severe ROP is included as an outcome, enteral DHA+ARA supplementation reduces severe ROP and is associated with lower odds of ROP requiring treatment after adjustment for gestational age. These findings differ from the conclusions of Dang et al. and support reconsideration of DHA+ARA supplementation as a strategy to reduce sight-threatening ROP in preterm infants.

16.
arXiv (CS.CL) 2026-06-12

WildIFEval: Instruction Following in the Wild

Recent LLMs have shown remarkable success in following user instructions, yet handling instructions with multiple constraints remains a significant challenge. In this work, we introduce WildIFEval - a large-scale dataset of 7K real user instructions with diverse, multi-constraint conditions. Unlike prior datasets, our collection spans a broad lexical and topical spectrum of constraints, extracted from natural user instructions. We categorize these constraints into eight high-level classes to capture their distribution and dynamics in real-world scenarios. Leveraging WildIFEval, we conduct extensive experiments to benchmark the instruction-following capabilities of leading LLMs. WildIFEval clearly differentiates between small and large models, and demonstrates that all models have a large room for improvement on such tasks. We analyze the effects of the number and type of constraints on performance, revealing interesting patterns of model constraint-following behavior. We release our dataset to promote further research on instruction-following under complex, realistic conditions.

17.
arXiv (CS.LG) 2026-06-19

Multi-Modal Contrastive Learning for Implicit Earth Embeddings via Location Tying

arXiv:2606.20167v1 Announce Type: new Abstract: Spatial prediction tasks are often limited by a lack of high-quality labelled ground-truth observations. To overcome this challenge, self-supervised pre-training is a possible solution, with contrastive learning dominant for location encoders. Those approaches usually align geographic coordinates with just one additional modality. We propose two multimodal contrastive learning architectures: Multimodal Embedding via Location Tying (MELT) and Sequential Alternating Location Training (SALT). These architectures expand this framework beyond two modalities by utilising unpaired geospatial data. Both methods are technically viable and match the performance of the strongest two-modality baseline (SATCLIP) across four downstream tasks. However, increasing the number of modalities does not consistently improve performance, suggesting that the chosen location encoder is the main limitation - the contrastive objective reaches its peak early, regardless of modality diversity or pre-training volume. MELT provides more stable training than SALT and presents a stronger foundation for future scaling.

18.
arXiv (CS.AI) 2026-06-16

Discovering Lattice Reduction Strategies via Self-Play

arXiv:2606.15301v1 Announce Type: cross Abstract: The Lenstra-Lenstra-Lovász (LLL) algorithm is a seminal contribution to computer science used for lattice basis reduction, yet its polynomial-time outputs produce bases that are far from optimal as the dimension grows. We show that deep reinforcement learning can discover strictly superior, generalizable reduction strategies by interacting with the primitive action space of LLL. We formulate lattice reduction as a single-player Markov Decision Process (MDP) and train a deep residual network using an AlphaZero-style self-play pipeline augmented with adaptive-horizon MCTS (Monte Carlo Tree Search), which couples multi-step network predictions with an entropy-gated expansion mechanism. The resulting policy, DeltaStar, is trained exclusively on small $8$-dimensional $q$-ary lattices and requires fewer primitive row operations than LLL. Crucially, it generalizes zero-shot to unseen moduli and higher dimensions up to $n=32$ without retraining.

19.
arXiv (quant-ph) 2026-06-19

Distinguishing quantum processes with bounded coherent memory

arXiv:2606.19511v1 Announce Type: new Abstract: Distinguishing multi-time quantum processes is a fundamental task underlying the diagnosis, benchmarking, and learning of temporally correlated quantum dynamics. The standard benchmark for distinguishing two processes is the strategy-norm distance, which optimizes over arbitrary adaptive probing strategies but can require large coherent memory and time-dependent control. We introduce machines for autonomous distinction~($\mathsf{MAD}$s): probing strategies that apply the same quantum instrument at each time step, retain the full classical outcome record, and carry a coherent memory of dimension $d_A$. Optimizing over these strategies defines a memory-parametrized distinguishability measure, $d^{(N)}_{\mathsf{MAD}}(\mathbf{P}^N,\mathbf{Q}^N;d_A)$. We show that the resulting hierarchy is monotone in coherent memory and complete at finite times. Specifically, any admissible $N$-step probing strategy can be compiled into a single $\mathsf{MAD}$ with an internal counter and sufficiently large coherent memory, so the hierarchy saturates the strategy-norm benchmark. For recurrent processes generated by repeated system–environment interactions, we derive a single-step description that separates the generation of new distinguishing information from the propagation and decay of information generated at earlier times. Numerical results in a repeated-interaction model show that increasing coherent memory systematically improves the $\mathsf{MAD}$ success probability and closes the gap to the strategy-norm distance while remaining substantially more tractable to evaluate. $\mathsf{MAD}$ distinguishability therefore provides an operational and scalable framework for quantifying what can be learned about genuinely multi-time quantum processes with bounded coherent memory.

20.
arXiv (CS.LG) 2026-06-17

Evaluating Uplift Modeling under Structural Biases: Insights into Metric Stability and Model Robustness

arXiv:2603.20775v2 Announce Type: replace Abstract: In personalized marketing, uplift models estimate the incremental effect of an intervention by modeling how customer behavior would change under alternative treatments using counterfactual analysis. However, real-world marketing data often exhibit various biases, such as selection bias, spillover effects, measurement error, and unobserved confounding. These biases can adversely affect both the accuracy of uplift estimation and the validity of evaluation metrics. Despite the importance of bias-aware assessment, there remains a lack of systematic studies evaluating how different models and metrics perform under such biased conditions. To bridge this gap, we design a systematic benchmarking framework. Unlike standard predictive tasks, real-world uplift datasets inherently lack counterfactual ground truth. This limitation renders the direct validation of evaluation metrics infeasible and prevents the precise quantification of biases. Therefore, a semi-synthetic approach serves as a critical enabler for systematic benchmarking. This approach effectively bridges the gap by retaining real-world feature dependencies while providing the ground truth needed to isolate structural biases. Our investigations reveal that (i) uplift targeting and prediction can manifest as distinct objectives, where proficiency in one does not ensure efficacy in the other; (ii) while many models exhibit inconsistent performance under diverse biases, TARNet shows notable robustness, providing insights for subsequent model design; (iii) the stability of evaluation metrics is linked to their mathematical alignment with the ATE, suggesting that ATE-approximating metrics yield more consistent model rankings under structural data imperfections. These findings suggest the need for more robust uplift models and evaluation metrics under real-world data imperfections.

21.
arXiv (CS.CL) 2026-06-18

Possible or Definite? A Benchmark for Evaluating Diagnostic Uncertainty Preservation in Clinical Text

Large language models (LLMs) are increasingly used for clinical text tasks such as summarization and revision. While most studies evaluate the fluency and coherence of LLM-generated text, whether LLMs correctly preserve diagnostic uncertainty remains underexplored. In clinical practice, phrases such as ``possible pneumonia'' communicate the strength of available evidence and directly guide decisions about follow-up testing and treatment. Altering these uncertainty expressions can change the clinical meaning entirely. In this paper, we systematically evaluated this problem in two steps. First, we constructed a benchmark of 1,200 clinical documents with 9,184 uncertainty annotations across five levels. Second, we evaluated three LLMs on this benchmark. Our results show that (1) LLMs preserve the original uncertainty cues poorly, often less than half the time; (2) LLMs struggle with nuanced distinctions between adjacent levels. This work reveals a failure mode not captured by standard evaluation metrics and provides implications for the safe deployment of LLMs in clinical workflows.

22.
arXiv (CS.CL) 2026-06-19

Benchmarking Local LLMs for Natural-Language-to-SQL Querying in Biopharmaceutical Manufacturing: An Empirical Benchmark on Consumer-Grade Hardware

Biopharmaceutical manufacturing organizations operate under regulatory frameworks such as FDA guidance, EU Good Manufacturing Practice (GMP), and the EU AI Act, which can restrict the use of cloud-based artificial intelligence systems. Locally deployed large language models (LLMs) offer a privacy-preserving alternative, but their suitability for pharmaceutical manufacturing tasks remains underexplored. This study evaluates four open-source LLMs (Qwen 2.5 Coder 7B, Llama 3.1 8B, Mistral 7B, and Meditron 7B) deployed locally via Ollama for natural-language-to-SQL generation over a pharmaceutical manufacturing database. A FastAPI-based evaluation platform, PharmaBatchDB AI, was developed using a synthetic Microsoft SQL Server database containing approximately 63,000 records across Batch, Manufacturing Execution System (MES), and Clean-In-Place (CIP) modules. Models were benchmarked on 60 domain-specific natural-language questions using metrics including SQL extraction rate, SQL compliance, factual consistency, ROUGE-L, hallucination rate, throughput, and latency. Qwen 2.5 Coder 7B, Llama 3.1 8B, and Mistral 7B generated SQL for all evaluation tasks, while Meditron 7B failed on nearly all tasks due to context-window limitations and poor SQL generation capability. Llama 3.1 8B achieved the highest SQL compliance, whereas Qwen 2.5 Coder 7B achieved the strongest overall text similarity and factual consistency. Performance differences between the two leading models were not statistically significant. The results show that code-tuned general-purpose LLMs outperform a domain-specific biomedical model on structured query generation for pharmaceutical manufacturing data. Although fully local, GxP-aligned NLQ systems are feasible on consumer hardware, current performance levels still require human oversight and downstream validation for regulated use.

23.
arXiv (CS.AI) 2026-06-19

The Hidden Evolution of Disguised Visual Context inside the VLM

arXiv:2606.20077v1 Announce Type: cross Abstract: Visual tokens enter Large Language Models (LLMs) as raw, foreign signals. How they are transformed into meaningful representations and interact with the language space depends entirely on the integration architecture. Whether by treating visual tokens as in-context prompts within the input sequence or injecting them directly into the LLM's intermediate layers. A controlled comparison and understanding of how these architectural choices affect visual information and its internal transformation to integrate with the LLM remains underexplored. We provide a fair comparison by evaluating in-context and layer-wise injection VLM integration paradigms under identical training conditions across single image, multi-image, and video benchmarks. In doing so, we uncover a hidden evolution where visual tokens enter the LLM as disguised visual context, raw representations lacking linguistic structure, but are progressively reshaped depending on the integration paradigm, each capturing fundamentally different frequency characteristics of the visual signal. We show that this evolution inside the LLM determines what visual features the VLM can utilize effectively, how visual representations align with the language space, and ultimately how each paradigm performs across different tasks. We further demonstrate that attention allocation alone is insufficient, and that performance is driven by the quality of visual representations at each layer.

24.
medRxiv (Medicine) 2026-06-15

GLLaucoMed: A Secure LLM-Powered Agentic Workflow for Automated Medication Extraction from Free-Text Glaucoma Clinical Notes

Purpose: To evaluate the efficacy of large language models (LLMs) in extracting medication-related information from glaucoma clinical notes in the electronic health record (EHR). Design: Cross-sectional. Subjects: 1,250 subjects in the Bascom Palmer Ophthalmic Repository. Methods: Extracted clinical notes from glaucoma-related encounters between 2014 and 2024 were labeled by two glaucoma specialists with a third serving as an adjudicator. Graders were asked to label current topical medications (CTM), proposed changes to topical medications ({Delta}TM), current oral medications (COM), and proposed changes to oral medications ({Delta}OM) in a structured fashion. The dataset was split into development (10%), validation (10%), and test (80%) sets stratified by clinician. Development and validation sets were used to engineer and refine prompts, and the held-out test set was used for model assessment. Five LLMs (Claude Opus 4.6, DeepSeek-V3.2, GPT 5.2, Grok 4.1, and Qwen3.6-35B-A3B) were accessed via Microsoft Azure AI Foundry within a HIPAA-compliant environment. Inter-grader agreement was assessed with Gwet AC1. LLM performance was initially assessed in a binary fashion with F1 scores, and the degree of text match among positive cases was evaluated using exact match accuracy and Jaccard Index (JI). Main Outcome Measures: F1 score, exact match accuracy, JI. Results: Gwet AC1 for intergrader agreement was 0.799, 0.888, 0.985, and 0.988 for CTM, {Delta}TM, COM, and {Delta}OM, respectively. F1 scores for CTM were 0.985, 0.971, 0.978, 0.968, and 0.970 for Claude, Deepseek, GPT, Grok, and Qwen, respectively; for {Delta}TM: 0.905, 0.826, 0.897, 0.842, 0.855, respectively; for COM: 0.923, 0.887, 0.899, 0.906, 0.894, respectively; for {Delta}OM: 0.958, 0.815, 0.937, 0.835, 0.940, respectively. Among positive cases, range of exact match accuracies for CTM (N=1354) was 0.730- 0.882 and range of JIs was 0.809-0.918. For {Delta}TM (N=404), exact match accuracy range was 0.619-0.780 and JI range was 0.668-0.827. For COM (N=47), exact match accuracy range was 0.766-0.872 and JI range was 0.765-0.870. For {Delta}OM (N=25), exact match accuracy range was 0.583-0.920 and JI range was 0.583-0.922. Conclusions: The GLLaucoMed pipeline demonstrated high performance in extracting and standardizing medication data from unstructured clinical notes, including both current medications and proposed changes. Claude and GPT exhibited the strongest performance.

25.
arXiv (quant-ph) 2026-06-17

Cavity-enhanced superconducting response in an underdoped cuprate

arXiv:2606.18084v1 Announce Type: cross Abstract: Superconductors carry electrical current without resistance when paired electrons condense into a coherent macroscopic quantum state. In underdoped cuprates, evidence suggests that pairing-related correlations and superconducting fluctuations can survive above the temperature at which global coherence is lost, pointing to phase fluctuations as a key limitation on superconductivity in this regime. Motivated by recent demonstrations of cavity-modified collective states in quantum materials, we investigate whether superconducting coherence can be stabilized by engineering the electromagnetic environment of the superconductor. We study an underdoped YBa$_2$Cu$_3$O$_{7-\delta}$ thin film in a tunable terahertz cavity formed with a semi-transparent gold mirror. From temperature-dependent terahertz transmission measurements, we find that the cavity enhances the superconducting response below the critical temperature, with an increase of the inferred superfluid weight. The effect becomes more pronounced at smaller cavity lengths and is accompanied by an upward shift of the superconducting onset temperature. Calculations based on a cavity-coupled model for phase-fluctuating superconductors capture these trends and support an interpretation in terms of cavity-enhanced phase stiffness. These results showcase the potential of cavity engineering for designing emergent functionalities in correlated systems.