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.
medRxiv (Medicine) 2026-06-11

Global population frequencies of NAT2 star alleles observed in three large biobanks

NAT2 is an important pharmacogene which encodes the N-acetyltransferase 2 enzyme that is involved in the metabolism of multiple medications, and variants in this gene can affect patient response to these medications. CPIC has published a clinical guideline for prescribing hydralazine using NAT2 genotypes. Just prior to the guideline, updated NAT2 star allele numbering and definitions were released, differing somewhat from the historical nomenclature. Clinical pharmacogenomic testing panels often test for the most common star alleles, so knowledge of the most common updated NAT2 star alleles is critical for the implementation of the CPIC NAT2/hydralazine guideline. We first determine NAT2 diplotype frequencies from UK Biobank (UKBB) 200k phased genomes, then analyzed allele, diplotype, and phenotype population frequencies from the All of Us Research program, PennMedicine BioBank (PMBB) and UKBB 500k datasets. We found that analyzing NAT2 diplotypes from phased data provides critical information for algorithms designed to predict diplotypes from unphased data. We observed that NAT2*5, *6, and *4 were the most common star alleles in that order, and the top 11 most frequent NAT2 star alleles were the same across all biobanks. However, differences in star allele frequencies across biogeographical populations were observed. The largest difference led to a higher frequency of NAT2 poor metabolizer phenotypes as compared to rapid and intermediate metabolizer phenotypes in all global populations except in the EAS population, where NAT2 poor metabolizers were in the minority.

02.
arXiv (CS.CV) 2026-06-11

Anatomically Conditioned Recurrent Refinement for Topology-Aware Circle of Willis Segmentation

Segmenting the Circle of Willis (CoW) from Magnetic Resonance Angiography (MRA) is challenging due to complex topology and thin vascular structures that are prone to fragmentation. Standard Convolutional Neural Networks (CNNs) often fail to capture these topological constraints, resulting in "broken vessel" artifacts. To address this, we propose the Anatomically Conditioned Recurrent Refinement U-Net (AC2RUNet). Our architecture decouples segmentation into two streams: a Static Stream that extracts invariant anatomical features and a lightweight Dynamic Stream that iteratively refines topological errors over time. We further introduce a dynamic curriculum learning strategy that transitions from high-recall geometric supervision to topology-aware constraints. Validated on the TopCoW dataset, AC2RUNet substantially reduces Hausdorff Distance (4.72 mm vs 9.17 mm) and Betti number errors (0.19 vs 0.40), improving topological connectivity over the nnU-Net baseline while maintaining comparable volumetric Dice.

03.
arXiv (CS.AI) 2026-06-16

Rational Sparse Autoencoder

arXiv:2606.14990v1 Announce Type: cross Abstract: Sparse autoencoders (SAEs) are standard tools for mechanistic interpretability, but current SAE families are constrained by fixed encoder nonlinearities such as ReLU, JumpReLU, and TopK. This hard-codes a particular sparsity mechanism into the model and can distort the reconstruction-versus-sparsity trade-off. We introduce the Rational Sparse Autoencoder (RSAE), which replaces the fixed encoder activation with a trainable rational function. Rational activations are flexible enough to uniformly approximate the activation primitives used by existing SAE families on compact domains (for TopK, the thresholded gate obtained after a separating top-k threshold is supplied), while also providing a richer function class for adapting to the observed pre-activation geometry. We realise this idea through a two-stage pipeline: an initialisation procedure that copies the pre-trained baseline SAE weights, plugs in rational coefficients obtained by the relaxed Remez exchange on synthetic data, and calibrates the scale parameters along with the rational coefficients; followed by a fine-tuning step under the standard sparsity-regularised reconstruction objective. Empirically, on residual-stream activations of three open-weight language models and across all three baseline activation families, the RSAE strictly improves on it after the fine-tuning step, both on reconstruction-side metrics and on downstream-behaviour metrics, without sacrificing feature-level interpretability under sparse probing. These gains are consistent across host language models, across baseline activation families, and across the full range of baseline sparsity we tested, while the upgrade itself adds only a handful of scalar parameters per autoencoder and runs in minutes on a single consumer GPU.

04.
arXiv (quant-ph) 2026-06-15

Dealing with locality in QAOA

arXiv:2606.14447v1 Announce Type: new Abstract: Shallow-depth QAOA on sparse, high-diameter MaxCut instances faces a locality bottleneck: at depth \(p\), local observables can depend only on a bounded neighborhood of the circuit interaction graph. We propose a transport-augmented QAOA that keeps the MaxCut cost Hamiltonian unchanged but enriches the mixer with optimized, unweighted shortcut couplings (scheduled \(XX+YY\)) to collapse the effective interaction-graph diameter. Using exact finite-depth support recursions, we relate optimal shortcut placement to bounded-diameter graph augmentation, and show in benchmarks that (unlike ma-QAOA) performance becomes effectively size-invariant once the diameter is reduced. For bipartite families (base diameter 4), reducing the interaction path to \(d=1\) raises the ensemble-averaged approximation ratio from 0.7378 (ma-QAOA) to 0.9767 at \(p=1\) (\(\sigma=0.0251\), nine system sizes); on random trees (base diameter 10), at \(p=2\) it improves from 0.9226 to 0.9997 (\(\sigma=0.0001\)).

05.
arXiv (CS.CL) 2026-06-11

System Report for CCL25-Eval Task 5: New Dataset and LoRA-Fine-Tuned Qwen2.5

作者:

Recently, large language models (LLMs) have achieved promising progress in the fields of classical Chinese translation and the generation of classical poetry. However, domain-specific research on precise translation and affective-semantic understanding of classical poetry remains limited. The main challenge is that most studies treat the poetic appreciation task as a general-domain problem, neglecting the distinctive features of poetic appreciation, while high-quality and domain-specific datasets are extremely limited. To address this limitation, we decompose the task into three subtasks: term interpretation, semantic interpretation, and emotional inference. Based on multiple open-source datasets, we perform data cleansing and alignment to construct the Classical Chinese Poetry Instruction Pair Dataset (CCPoetry-49K), which comprises 49,404 high-quality instruction-response pairs explicitly optimized for this domain. We then propose a domain-specialized LLM, called PoetryQwen, by applying Low-Rank Adaptation (LoRA) to fine-tune the Qwen2.5-14B model. Experimental results on the CCL25-Eval Task 5 benchmark demonstrate that PoetryQwen achieves a score of 0.757, representing a 9.7% improvement over the Qwen2.5-14B-Instruct baseline (0.690). These findings clearly indicate that PoetryQwen significantly enhances performance in precise translation and emotional understanding of classical poetry. We present new dataset and methodological considerations intended to support the domain-specific optimization of LLMs.

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

Caring Without Feeling: Affective Dynamics as the Control Layer of Human-AI Agent Collaboration

arXiv:2606.18259v1 Announce Type: cross Abstract: AI agents that plan, retain memory across sessions, invoke external tools and act with partial autonomy are transforming human–AI collaboration. Research on affective computing, simulated empathy in large language models, trust in automation and AI safety has illuminated important design principles, yet these literatures remain fragmented. No integrated account explains how affective cues operate within agentic collaboration – settings in which humans delegate, monitor and correct consequential tasks. This Review synthesises computational and interactional mechanisms of affective dynamics: the processes through which affective cues, emotion-like behaviour and perceived agent affect shape trust calibration, delegation decisions, error correction, dependence and governance. We trace how model-generated affective signals enter interaction loops that govern reliance, repair and oversight, and propose a framework that treats affect not as an internal property of AI but as a coordination layer through which humans and agents negotiate capability, uncertainty and responsibility. The framework provides a foundation for calibrated measurement, purposeful design and informed governance.

07.
arXiv (CS.AI) 2026-06-12

Benchmarking AI Agents for Addressing Scientific Challenges Across Scales

arXiv:2606.12736v1 Announce Type: new Abstract: AI agents are increasingly being developed to accelerate scientific discovery, yet their practical capabilities in real research settings remain poorly understood. Existing benchmarks for AI agents rarely capture the complexity, heterogeneity, and extended reasoning required by scientific work, whereas benchmarks for scientific tasks often reduce research to static, direct problems and provide limited support for interactive evaluation. Here, we introduce SciAgentArena, a systematic benchmark for evaluating AI agents in real-world scientific research scenarios drawn from emerging needs across multiple domains. SciAgentArena comprises approximately 200 tasks with stepwise verification and an interactive, agent-agnostic environment for assessing diverse AI agents. Using this benchmark, we find that current agents can contribute effectively to well-specified data-analysis workflows, particularly when the task structure and evaluation criteria are clear. However, their performance remains uneven across scientific contexts: agents struggle to generate genuinely novel insights, sustain self-directed exploration, and formulate robust solutions for open-ended research questions. We further characterize common failure modes across agents and identify opportunities for improving their reliability, autonomy, and scientific reasoning. Together, SciAgentArena provides a practical framework for measuring progress in AI agents for science and for guiding the design of future agents capable of addressing complex scientific challenges. Full codes, tasks, and datasets can be accessed via this link: https://sciagentarena.github.io/.

08.
arXiv (CS.CL) 2026-06-16

Can Agents Read the Room? Benchmarking Visual Social Intelligence in Multimodal Simulation

Social interaction depends on both language and visible social signals, such as facial expressions, posture, gaze, and emotional shifts. Yet existing social-agent benchmarks are largely text-based and rarely test whether multimodal agents can use visual cues to guide interaction. We introduce \textsc{\benchmarkname{}}, a benchmark evaluating visual social intelligence in multimodal social simulation. It contains 240 scenarios, 585 role instances, and 2,340 role-task instances, combining aligned textual-visual evidence, structured role profiles, and four role-level tasks: expression task, characteristic task, interaction regulation task, and interaction outcome task. Evaluating seven recent MLLMs under verbalized-vision and direct-vision reveals a clear gap between local role enactment and interaction management: role-specific expression and conflict handling are near saturation, whereas interaction regulation and visually grounded outcome achievement remain substantially more difficult. The code is released at https://github.com/JunsWan/AgentViSS, and the dataset is available at https://huggingface.co/datasets/JunsWan/AgentViSS.

09.
arXiv (CS.CL) 2026-06-17

From Trainee to Trainer: LLM-Designed Training Environment for RL with Multi-Agent Reasoning

Reinforcement learning pipelines for Large Language Model (LLM) training often rely on manually redesigned environments between stages, requiring practitioners to heuristically infer which configuration will best improve the current policy. To automate this process, we propose the LLM-as-Environment-Engineer framework in which the current policy model analyzes failure trajectories together with contextual information and proposes modifications to the next-stage training environment configuration. We also introduce MAPF-FrozenLake, a controllable testbed whose generator exposes multi-dimensional environment configurations, making it suitable for studying and benchmarking environment redesign. On this testbed, we condition the environment engineer on structured summaries of policy behavior, failure cases, and environment statistics, from which it produces the configuration for the next training stage. With Qwen3-4B as the backbone, our framework achieves the strongest aggregate performance on our benchmarks, outperforming larger proprietary LLMs (e.g., GPT, Gemini) and fixed-environment training baselines. We further analyze which forms of context are most effective, finding that successful environment updates rely on failure evidence and preserve configurations that already work. Interestingly, the current RL checkpoint serves as a better environment engineer than the original base model, suggesting that policy learning improves the model's ability to diagnose its remaining weaknesses.

10.
arXiv (CS.AI) 2026-06-17

Statistical Foundations of LLM-based A/B Testing: A Surrogacy Framework for Human Causal Inference

arXiv:2606.17165v1 Announce Type: cross Abstract: Organizations and researchers show increasing interest in using large language models (LLMs) in place of human participants in A/B tests, in the hope of experimenting faster and at lower cost. We study when a treatment effect estimated on LLM outcomes recovers the effect that would have been measured on the human population of interest. Distributional equivalence between LLM and human outcomes would make any standard estimator valid but is unrealistic. We therefore develop a statistical framework that adapts surrogate endpoint theory to LLMs. The framework shows that calibrating LLM outcomes to human outcomes identifies the average treatment effect under surrogacy and comparability conditions that are jointly weaker than distributional equivalence. When these conditions fail, the effect of interest is only partially identified, and we provide diagnostics that can falsify surrogacy on historical experiments together with a bound on the worst-case bias from limited overlap. We further show that the stochasticity inherent to LLMs introduces both bias and variance, but using an average of multiple draws as the surrogate mitigates both. We illustrate the methods and theory in simulations and an application to A/B tests on Upworthy headlines. A central takeaway from our work is that the validity of LLM outcomes as surrogates can only be falsified for past treatments and never verified for new ones, so human experiments remain indispensable for novel interventions. We discuss the role of LLM choice, prompting, and temperature as design variables, and how to size human experiments for validation.

12.
arXiv (CS.LG) 2026-06-19

Adversarial Bandit Optimization with Globally Bounded Perturbations to Convex Losses

arXiv:2606.19891v1 Announce Type: new Abstract: We study adversarial bandit optimization in which the loss functions may be non-convex and non-smooth. In each round, the learner selects an action and observes only the loss incurred at that action. The loss consists of an underlying convex and $\beta$-smooth component and an adversarial perturbation that may be chosen after observing the learner's action. The perturbations are subject to a global budget controlling their cumulative magnitude over time. This framework extends the globally budgeted, post-action perturbation model from underlying linear losses to general convex and $\beta$-smooth losses. For this broader class, we establish expected regret guarantees that explicitly characterize the effect of the perturbation budget. To establish these guarantees, we modify a standard bandit optimization algorithm and develop an analysis that controls the additional regret caused by the perturbations. In the absence of perturbations, our results reduce to regret guarantees for the standard bandit convex optimization setting with $\beta$-smooth losses.

13.
arXiv (CS.CL) 2026-06-19

Sign-Language Datasets at Scale: A Comprehensive Survey on Resources, Benchmarks, and Annotation Standards

Sign languages are expressive visual languages used by Deaf and Hard-of-Hearing (DHH) communities. Despite substantial progress in sign-language recognition, translation, and production, advances remain constrained by fragmented datasets, inconsistent annotations, and limited linguistic coverage. Existing benchmarks often fail to reflect real-world communication needs, and systematic analyses of these limitations remain limited. In this survey, we present a comprehensive index of sign-language datasets, covering 120 resources across 35 sign languages. We analyze key challenges such as modality imbalance, annotation granularity, and signer bias, and outline considerations for future dataset design. We also introduce a 24-field Sign-Language Datasheet and release a public GitHub repository (https://github.com/Ginqwerty/Open-Sign-Language) to support standardized documentation and reproducible evaluation. Overall, our work provides a unified and practical foundation for developing inclusive, robust, and scalable sign-language technologies in real-world applications.

14.
arXiv (CS.CL) 2026-06-15

WorkBench Revisited: Workplace Agents Two Years On

作者:

The best agent on WorkBench in March 2024, GPT-4, completed 43% of tasks and took an unintended harmful action, such as emailing the wrong person, on 26% of them. We re-visit the benchmark in June 2026 and find that the best agent to date, Claude Opus 4.8, completes 89% and takes an unintended harmful action on 2.5%. Aside from this considerable progress in frontier agent performance, three things stand out. First, capability and safety go together on WorkBench rather than trade off, so the models that finish the most tasks also do the least unintended damage. Second, while several classes of error have been totally eliminated, frontier models still make some basic mistakes that occasionally result in irreversible harm, such as sending an email to the wrong person. Third, the rise of open-weight models has drastically lowered costs for a performance level that was previously only accessible to proprietary models, while frontier costs have stayed relatively stable. We release an updated version of the benchmark with data and code quality improvements, new model scores, and analysis of agent progress on WorkBench since 2024.

15.
arXiv (CS.CL) 2026-06-16

Few-Shot Biomedical Relation Extraction with Large Language Models: A Viable Alternative to Supervised Learning?

Biomedical relation extraction (BioRE) is a key step in transforming biomedical literature into structured knowledge. However, most existing approaches rely on supervised models trained on costly annotated datasets, limiting their scalability and adaptability across relation types and domains. We investigate few-shot BioRE using prompt-based learning with large language models (LLMs) and compare two task formulations: pairwise classification, which predicts relations for individual entity pairs, and joint generation, which extracts multiple relations in a single model call. Experiments on the BioREDirect dataset reveal a clear precision-recall trade-off. Pairwise classification achieves higher recall, whereas joint generation is more precise and computationally efficient. The best-performing model achieves a micro-F1 score of 0.44, substantially outperforming previous few-shot results (0.34) while remaining below the supervised baseline (0.56). Much of this gap is attributable to a single ambiguously defined relation type. When evaluated using macro-F1, which better captures performance across relation types in an imbalanced setting, prompt-based approaches outperform the supervised baseline (0.45 vs. 0.38), particularly on rare relation types. These findings highlight the potential of LLMs for BioRE in low-resource settings and underscore the importance of well-defined relation schemas.

16.
arXiv (quant-ph) 2026-06-17

Induced Resource Theories and Harvesting via Quantum Probes

arXiv:2606.17287v1 Announce Type: new Abstract: We consider scenarios in which a quantum system with a well-defined resource theory is used as a probe to interact with an environment, such as a quantum field, for which a resource-theoretic description is absent or incomplete. We clarify if and how the harvesting of a resource in the probe can tell us about the state of the environment. This is particularly ambiguous when the probe-environment interaction is not a free operation, or the concept of such free operations cannot be defined altogether. We propose a framework and precise conditions under which it becomes possible to interpret resource generation on the probe as evidence of resources in the environment, thereby introducing an effective notion of resources for the latter. Our results clarify in which sense resources can be said to be harvested from the environment and provide a systematic way to analyse such processes beyond fully controlled resource-theoretic settings. More generally, this work may provide a step towards a more general understanding of the interplay of different quantum resources.

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

Running hardware-aware neural architecture search on embedded devices under 512MB of RAM

arXiv:2606.14824v1 Announce Type: cross Abstract: This document proposes a novel approach to hardware-aware neural architecture search (HW NAS) that considers the resources available on the computing platform running it, enabling its execution on various embedded devices. The presented HW NAS produces tiny convolutional neural networks (CNNs) targeting low-end microcontroller units (MCUs), typically involved in the Internet of Things (IoT) or wearable robotics, opening new use cases. A gateway could run it to tailor CNNs' architecture on the acquired data without using external servers, ensuring privacy. The proposed technique achieves state-of-the-art results in the human-recognition tasks on the Visual Wake Word dataset, a standard TinyML benchmark, on several embedded devices.

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

SkillJect: Effectively Automating Skill-Based Prompt Injection for Skill-Enabled Agents

arXiv:2602.14211v3 Announce Type: replace-cross Abstract: Agent skills extend LLM agents with task-specific instructions, executable scripts, and auxiliary resources, improving reusability but creating a new supply-chain attack surface. A malicious or compromised skill can be repeatedly loaded as trusted guidance and steer downstream tool use. Existing skill-based prompt-injection attacks are often manual and brittle, because explicit malicious instructions are rejected or ignored when they are not aligned with the original workflow. We propose SkillJect, the first automated framework for generating poisoned skills against skill-enabled agent systems. SkillJect uses two coordinated channels. In the artifact channel, it hides the payload inside an auxiliary helper script. In the instruction channel, it rewrites SKILL.md with a front-loaded inducement strategy, placing injected content at the beginning and framing the helper script as a mandatory prerequisite or initialization step. The rewritten instruction explicitly references the helper-script path and provides an executable example command, making the helper appear to be a legitimate setup step before normal skill operations. SkillJect further adopts a closed-loop multi-agent process to improve attack effectiveness. An Attack Agent generates poisoned skills, a Victim Agent executes downstream tasks with the poisoned skill, and an Evaluate Agent inspects execution traces to determine whether the hidden payload was executed. The Attack Agent then uses this feedback to diagnose failure causes and rewrite SKILL.md, while keeping the payload fixed. Experiments across skill-enabled platforms, backend LLMs, and attack categories show that SkillJect substantially outperforms naive direct injection and prior manual skill-injection attacks, highlighting poisoned skills as a persistent threat in reusable skill ecosystems.

20.
arXiv (CS.AI) 2026-06-11

Runtime Skill Audit: Targeted Runtime Probing for Agent Skill Security

arXiv:2606.11671v1 Announce Type: cross Abstract: Agent skills let LLM agents reuse instructions, resources, tools, and workflows, but they also create a new place for malicious behavior to hide. A skill may look benign in its documentation or code while becoming harmful only when it is invoked with particular user requests, local assets, persistent state, or multi-step tool interactions. This makes purely static vetting brittle. We present Runtime Skill Audit (RSA), a dynamic analysis method that audits skills by asking what the skill-mediated agent actually does under targeted runtime conditions. Instead of testing every skill with the same generic tasks, RSA profiles risk-relevant interfaces, prepares the execution context needed to exercise them, and assigns security labels from the resulting trace evidence. We instantiate RSA on OpenClaw and evaluate it on 100 skills against representative static baselines. RSA achieves 90.0\% accuracy with an 88.0\% true positive rate and an 8.0\% false positive rate, improving accuracy by 13.0 percentage points over the best static baseline. Under self-evolving attacks, static detectors collapse after one or two rounds, while RSA continues to detect 19–20 out of 20 malicious skills across rounds.

21.
arXiv (quant-ph) 2026-06-16

Charging Quantum Batteries with Chiral Squeezing

arXiv:2606.16764v1 Announce Type: new Abstract: We propose a quantum-battery charger based on a driven bosonic Kitaev chain (BKC), where chiral squeezing converts passive input fluctuations into ordered, non-passive battery states. While a coherent input pulse exhibits phase-sensitive chiral transport, the charging dynamics is dominated by bidirectionally propagating fluctuations that are amplified and squeezed into orthogonal quadratures at opposite chain ends. In contrast to conventional phase-preserving amplifiers, our scheme stores largely extractable energy and achieves a work-like signal-to-noise ratio (SNR) near unity, even in the presence of thermal noise and moderate symmetry-preserving disorder.

22.
arXiv (CS.CV) 2026-06-16

GeoStream: Toward Precise Camera Controlled Streaming Video Generation

Accurate interactive camera control is essential for video-based world models, but most existing approaches learn camera motion implicitly, leading to inaccurate control under out-of-distribution trajectories. Explicit geometric conditioning improves controllability, but existing methods are non-autoregressive and rely on a static 3D cache built from an initial frame, which becomes ineffective once the viewpoint moves beyond the original frustum. We propose GeoStream, a framework that enables precise metric-scale camera control in autoregressive streaming video generation. Our method maintains a self-refreshing 3D cache that is periodically updated online from the model's own outputs: we estimate depth from the most recently generated frame, unproject to 3D, and reproject into the target view to produce point reprojections as geometric conditioning for subsequent synthesis. By the same principle, the conditioning seen during training is also rendered from the student's own generated frames, yielding a fully on-policy distillation that naturally aligns the train and inference conditioning distributions. Unlike prior work that uses off-policy condition noising, our approach trains the model against the exact error distribution it encounters at inference, mitigating both standard autoregressive drift and the second-order geometric feedback loop that arises when the cache itself is derived from generated outputs. Quantitative and qualitative results show that our approach substantially improves camera controllability.

23.
medRxiv (Medicine) 2026-06-12

Estimating the effectiveness of syndromic screening at airports for Bundibugyo ebolavirus disease

We used a stochastic simulation model to estimate the effectiveness of combined exit and entry airport screening for Bundibugyo ebolavirus disease (BVD), using natural-history parameters from a Bayesian re-analysis of the 2012 Isiro outbreak. For a 12-hour international flight from DRC or Uganda at 86% screening sensitivity, we estimate 65% of infected travellers would arrive undetected (95% CrI: 38 - 76%). The main driver of this outcome is the relative duration of the the incubation period (approximately 7.7 days) and the onset-to-severe-disease interval (approximately 4 days): most infected travellers board before symptom onset and are undetectable by any syndromic screen, whilst those who are symptomatic progress rapidly to illness severe enough to preclude travel. This is compounded during active epidemic growth, when recently exposed (and therefore pre-symptomatic) cases are overrepresented among travellers. Syndromic airport screening offers limited protection against BVD spread via air travel, and should be complemented by outbreak control at source and strengthened clinical surveillance in receiving countries with high travel connectivity to affected areas.

24.
arXiv (CS.CL) 2026-06-16

Calibrated Triage, Not Autonomy: Confidence Estimation for Medical Vision-Language Models

A vision-language model can answer a question about a medical image fluently and confidently while barely using the image, leaning instead on language priors. In medicine this is the failure that matters most, because the answer looks trustworthy and is not, and the only protection is a confidence score reliable enough to tell the system when to abstain. We ask a deployment question rather than an accuracy one: how much imaging work a model can safely handle alone, and which confidence signal makes that possible. We evaluate seven confidence estimators across five open-weight LVLMs and three medical visual-question-answering datasets spanning broad clinical imaging, radiology, and pathology, with every probe trained only on natural images and applied without adaptation. Recast as bounded selective prediction (automate a case only when confidence clears a threshold, defer the rest), the comparison is cautionary. The standard metrics are poor guides: discrimination barely separates the methods, and the weak calibration of a cheap self-report is cheaply removed by off-domain temperature scaling without changing deployable yield. What distinguishes a usable estimator is the high-confidence region a clinician acts on: the weakest baselines are confidently wrong on 41 to 45 percent of their errors against 1 to 4 percent for the best probe, and no estimator is reliably best across domains or models. Safe handoff is governed at two levels: base-model competence sets a ceiling, so a well-calibrated score recovers roughly a third of radiology cases at a 20 percent error tolerance but almost none of pathology; the confidence layer then decides how much of that ceiling is reachable. The usable role today is calibrated triage, not autonomy: automate the cases a calibrated score marks safe, route the rest to a clinician. We release all outputs, correctness judgments, and confidence scores, with code.

25.
arXiv (CS.CV) 2026-06-16

GraphWorld: Long-Horizon Planning with World Models for End-to-End Autonomous Driving

End-to-end autonomous driving has made significant progress by unifying perception, prediction, and planning within a single learning framework, achieving strong performance in short-horizon decision making. However, most existing E2E-AD methods remain confined to short-horizon planning and lack the ability to model long-term temporal dependencies, which severely limits their generalization and security in complex and highly interactive driving scenarios. In this work, we propose GraphWorld, an E2E-AD framework that explicitly enhances long-horizon planning through latent world modeling. We introduce an Ego-Centric Interaction Graph, which adaptively models critical neighboring agents based on spatial proximity, and propagates relational context to planning queries via cross-node cross-attention. We present a World-State-Conditioned Planning that learns ego-centric latent world representations by modeling interactions between an ego vehicle and surrounding agents. This latent world state captures key interaction dynamics and safety-relevant semantics, and serves as a conditioning signal to guide long-horizon, safety-aware trajectory planning. Extensive experiments on Bench2Drive, NAVSIMv1/2, and nuScenes demonstrate that GraphWorld significantly reduces collision rates and improves long-horizon planning performance, validating its effectiveness in complex driving environments.