探索全球前沿学术脉络

01.

medRxiv (Medicine) 2026-06-22 DOI: HASH:8931b12e773d08691f084a97cc26baff

Early-life nutritional environment is associated with late-life cognition in the Health and Retirement Study, a pellagra epidemic natural experiment

作者:

Vasiljevic↗Schmitz↗L. L↗Engelman↗C. D↗

Early-life exposures are important to several late-life health outcomes. We sought to study the effect of an in utero nutritional environment and its interaction with Alzheimer's disease (AD) genetic risk on late-life cognitive function. We used a natural experiment created by the pellagra epidemic, a nutritional disease caused by a vitamin B3 deficiency, to evaluate the association between in utero pellagra epidemic exposure and late-life cognitive function in the Health and Retirement Study (N = 18,285). We also evaluated whether the in utero exposure could modify the AD polygenic score's (PGS) effect on cognition. In utero pellagra epidemic exposure was significantly associated with cognition ({beta} = -0.025). However, these effects were not isolated to the prenatal period as exposure during childhood periods also had an effect. The interaction between the in utero exposure and the AD PGS was significant, where the genetic effect on cognition was amplified with increasing (progressively worse) in utero exposure levels. These associations imply that the early-life nutritional environment affects late-life cognitive function and that these effects can modify genetic risk.

阅读与讨论 → 访问原文 →

02.

arXiv (CS.CL) 2026-06-17 DOI: arXiv:2606.18037

ProvenanceGuard: Source-Aware Factuality Verification for MCP-Based LLM Agents

作者:

Ander Alvarez↗Santhiya Rajan↗Samuel Mugel↗Rom\'an Or\'us↗

Tool-using LLM agents increasingly use the Model Context Protocol (MCP) to answer from heterogeneous evidence sources, including search, APIs, databases, clinical records, and formulary tools. Standard factuality metrics usually test whether an answer is supported by pooled evidence, missing a provenance-sensitive failure mode: a claim may be supported somewhere while being attributed to the wrong source. We call this cross-source conflation. We introduce ProvenanceGuard, a source-aware verifier for MCP-grounded answers. It consumes captured MCP traces with stable tool IDs, source IDs, and raw outputs; decomposes answers into atomic claims; routes claims to source-specific evidence; checks support with NLI and a token-alignment proxy; compares stated attribution with the routed source; and returns per-claim verdicts plus an answer-level allow/block decision. Blocked answers can be repaired with retrieval-augmented answer revision and re-verified. We evaluate on 281 medical-domain MCP-agent traces. A 266-trace adjudicated subset yields 2,325 LLM-assisted claim labels split by trace; 361 held-out labels are human-verified. On the 40-trace held-out split, ProvenanceGuard achieves block F1 0.802 and source accuracy 0.858 over 260 source-eligible claims, outperforming source-blind baselines that do not emit claim-to-source IDs. On a harder multi-source benchmark it reaches block F1 0.846, while source-plus-relation accuracy drops to 0.229, showing that exact source ownership remains difficult with semantically close sources. Repair-and-reverify resolves all blocked answers in the full trace set, often via conservative fallback. In 50 controlled clinical conflation probes, ProvenanceGuard detects all injected attribution swaps with no retained wrong attribution. These results show that source attribution is an independent axis for factuality verification in MCP-based agents.

阅读与讨论 → 访问原文 →

03.

arXiv (CS.AI) 2026-06-16 DOI: arXiv:2412.00107

Virtual Sensing to Enable Real-Time Monitoring of Inaccessible Locations & Unmeasurable Parameters

作者:

Kazuma Kobayashi↗Farid Ahmed↗Jaewan Park↗Subhankar Sarkar↗Souvik Chakraborty↗Syed Bahauddin Alam↗

arXiv:2412.00107v2 Announce Type: replace-cross Abstract: Real-time monitoring of safety-critical interior states remains an open problem in energy systems where physical instrumentation is infeasible. Existing approaches rely on explicit governing equations, finite-dimensional state vectors, or per-instance retraining, which prevents mesh-independent, field-level inference at arbitrary interior coordinates under real-time constraints. We introduce operator-based virtual sensing for nuclear-grade thermal-fluid systems: we use the neural-operator framework to learn solution operators that map sparse boundary measurements to coupled internal fields in physically inaccessible regions, framing the problem class explicitly to distinguish it from classical state estimation and pointwise soft sensing. We instantiate this framework with MIMONet, a branch-trunk operator extended with three practical choices: multi-modal branch encoders for heterogeneous (scalar and function-valued) inputs; multiplicative branch fusion to preserve the bilinear PDE coupling structure; and shared-latent multi-field decoding with per-channel basis projections at the trunk's final layer. Evaluated across escalating complexity, from canonical lid-driven cavity flow to pressurized water reactor subchannels to fully coupled heat exchangers, MIMONet achieves below 5% relative errors and sub-millisecond inference on data-center accelerators (0.35 ms / 46 mJ per heat-exchanger inference on an NVIDIA H200, and sub-millisecond across the A40-H200-GH200 range), while remaining stable under 50% sensor noise. By staying accurate as geometric confinement and physics coupling intensify, MIMONet shows that operator-based virtual sensing can restore observability where physical instrumentation fails, establishing simulation-based feasibility within the evaluated operating envelopes as a step toward future experimental and cross-solver validation for safety-critical energy systems.

阅读与讨论 → 访问原文 →

04.

arXiv (CS.CL) 2026-06-11 DOI: arXiv:2606.11387

Small Experiments, Cheaper Decisions: A Case Study in Staged Promotion for Micro-Pretraining

作者:

Felipe Chavarro Polania↗

Short pretraining runs can reduce experimental cost, but they can also over-promote configurations that only look strong at tiny budgets. We study an auditable staged-promotion protocol for a fixed micro-pretraining runner on two heterogeneous host blocks: Windows A100 and Linux L40S. Starting from twelve prior-screened configurations, we use staged budgets of 2 minutes, 5 minutes, 10 minutes, 60 minutes, and 12 hours, with frozen promotion rules before expensive continuations. The early screens are intentionally treated as unstable: the 5- and 10-minute rankings are host-sensitive, and the eventual 12-hour top-ranked condition is not the mean-best condition at the replicated 10-minute gate. Because seed ranges differ across stages, these changes are operational promotion evidence, not within-seed curves. A replicated 60-minute gate keeps the Staged Factorial Screening bridge reference in the promoted set, where it ranks first in all four 60-minute host-seed cells. In the final 12-hour confirmation package, the bridge condition ranks first in all four host-seed cells across two seeds; the greedy comparator does not meet the frozen 0.010 val_bpb near-equivalence rule; and the cheaper d8/ar48 (depth-8, aspect-48) sentinel does not meet the frozen 0.020 mean-gap rule. The executed 12-hour branch spends 144 GPU-hours, and the full staged protocol records 169.2 training GPU-hours including screening stages. Continuing all four 60-minute candidates would spend 192 GPU-hours, while continuing all nine replicated 10-minute candidates would spend 432 GPU-hours. The latter numbers are accounting counterfactuals for unrun continuations, not evidence that skipped candidates could not have overtaken the reference. The result is a bounded cost-allocation finding, not a claim of global optimality, capacity-normalized superiority, or superiority over adaptive hyperparameter optimization methods.

阅读与讨论 → 访问原文 →

05.

arXiv (math.PR) 2026-06-25 DOI: arXiv:2603.17810

Localization for non-stationary Anderson models in three dimensions

作者:

Omar Hurtado↗

arXiv:2603.17810v2 Announce Type: replace-cross Abstract: We prove localization (near the bottom of the spectrum) for certain non-stationary variants of the Anderson model in three dimensions. More specifically, we prove a Wegner estimate, which implies localization by existing work. Two key inputs are a deterministic quantitative unique continuation theorem by Li and Zhang [Duke Math. J. 171(2): 327-415, 2022] and some combinatorial decompositions/bounds for non-stationary random potentials proved by the author [Commun. Math. Phys. 407:64, 2026].

阅读与讨论 → 访问原文 →

06.

arXiv (CS.CL) 2026-06-16 DOI: arXiv:2606.15872

SciOrch: Learning to Orchestrate Expert LLMs for Solving Frontier Multimodal Scientific Reasoning Tasks

作者:

Jingru Guo↗Xiangyuan Xue↗Lian Zhang↗Wanghan Xu↗Siki Chen↗Philip Torr↗Wanli Ouyang↗Lei Bai↗Zhenfei Yin↗

Frontier scientific reasoning remains a major challenge for large language models (LLMs), where even the strongest commercial systems fall short of expert-level performance. A closer look at model behavior reveals substantial complementarity that single-model evaluation hides: different frontier models excel on different question types, and no single model captures the full picture. We present SciOrch, a framework that trains a lightweight 8B model to orchestrate frontier LLMs for scientific reasoning. The orchestrator decomposes each question, delegates sub-problems to selected commercial models through API calls, and synthesizes a final answer. Training such an orchestrator is fundamentally harder than conventional agentic RL: each action triggers an API call that is expensive in both dollar cost and latency, making standard online rollouts infeasible. We address this with MCTS-based approach, producing diverse orchestration trajectories, extracting per-node single-turn samples, and optimizing the orchestrator with GRPO-style training. On a 240-question test set spanning SGI-Reasoning and Scientists' First Exam, SciOrch reaches 56.66% average accuracy, outperforming the strongest single commercial model by 3.74% and the strongest multi-agent baseline by 3.33%. It also attains the best accuracy on both SGI and SFE with less than half the API cost of typical multi-agent methods.

阅读与讨论 → 访问原文 →

07.

arXiv (CS.CV) 2026-06-25 DOI: arXiv:2606.25542

SAC$^2$-Net: Semantic Anchoring and Complementary-Consensus Fusion for Multimodal Micro-Expression Recognition

作者:

Xuepeng Zheng↗Tong Chen↗

Micro-expression recognition (MER) is challenging due to subtle facial movements, limited data, and the ambiguous relationship between Action Units (AUs) and emotion categories. Optical flow and motion magnification have been widely used to describe subtle facial dynamics from different perspectives: the former captures local motion displacement, while the latter amplifies weak appearance changes. In this work, we observe that these two modalities often exhibit asymmetric failure patterns: one modality may become noisy, distorted, or uninformative, while the other still preserves discriminative AU-related evidence. This phenomenon reveals their complementarity, but also raises two key challenges for fusion: cross-modal heterogeneity and spatially varying modality reliability. Motivated by this observation, we propose SAC$^2$-Net, a Semantic Anchoring and Complementary-Consensus Network for multimodal MER, which first aligns visual modalities with semantic anchors and then performs reliability-aware fusion. To reduce cross-modal heterogeneity before fusion, we introduce Semantic Anchoring Soft Alignment (SASA), which converts activated AUs into textual prompts and uses them as stable semantic anchors to align motion-magnified and optical-flow representations. Unlike hard contrastive learning, SASA constructs hierarchical AU-aware soft labels to preserve semantic proximity among samples with overlapping or anatomically related AU patterns. Based on the aligned representations, Complementary-Consensus Fusion (CCF) first repairs unreliable local evidence through complementary exchange and then enforces a shared spatial focus through consensus refinement. Extensive experiments on five MER benchmarks show that SAC$^2$-Net achieves state-of-the-art or highly competitive performance across coarse-grained, fine-grained, large-scale, and cross-dataset evaluation settings.

阅读与讨论 → 访问原文 →

08.

arXiv (CS.CV) 2026-06-11 DOI: arXiv:2606.11546

VL-DINO: Leveraging CLIP Vision-Language Knowledge for Open-Vocabulary Object Detectio

作者:

Hao Zhang↗Qinran Lin↗Linqi Song↗Yong Li↗

Vision-language models like CLIP can provide rich semantic priors for open-vocabulary object detection. However, jointly integrating both textual and visual knowledge into detection architectures remains challenging. In this paper, we propose VL-DINO, an open-vocabulary detector that enhances DINO through more effective exploitation of CLIP's vision-language knowledge. Specifically, a Query-guided Positive Sample Construction (QPSC) module is first developed to construct additional high-quality positive samples, enabling the vanilla DINO framework to better accommodate mixed training across heterogeneous data sources while providing more vision-language alignment signals, thereby incorporating richer textual knowledge during training. A Visual Semantic Encoder (VSE) module is then introduced to distill CLIP visual knowledge into backbone-extracted features, producing fused features for subsequent encoder refinement. Based on the fused features, an Object-Region Semantic Alignment (ORSA) module extracts object-centric region features and aligns them with the corresponding textual embeddings, further incorporating textual cues. In the zero-shot setting, VL-DINO-T and VL-DINO-L achieve 36.3 and 38.1 AP on the LVIS benchmark, respectively, consistently outperforming prior advanced approaches. Extensive experiments demonstrate the effectiveness and competitive performance of the proposed design.

阅读与讨论 → 访问原文 →

09.

arXiv (CS.LG) 2026-06-16 DOI: arXiv:2605.30837

Send a SCOUT First: Pre-hoc Reasoning for Adaptive Detector Allocation in Prompt-Injection Defense

作者:

Shuhao Zhang↗Jiarui Li↗Qi Cao↗Ruiyi Zhang↗Pengtao Xie↗

arXiv:2605.30837v2 Announce Type: replace-cross Abstract: Prompt-injection detectors are heterogeneous: each is strong on a different slice of attacks, and none is always reliable. Yet existing systems still treat detection as a fixed single-detector pipeline, committing every request to one detector's blind spots. We reframe defense as detector allocation: given a heterogeneous pool, decide per request which detectors to run and whether to escalate to an LLM judge. Our framework SCOUT (Scalable and Controllable Outcome-prediction for Uncertainty-aware Triage) makes this decision dynamic by predicting each detector's per-sample reliability and latency from how it behaved on similar past inputs, and exposes a single safety-utility threshold to the operator (where utility bundles benign-pass rate and wall-clock). To evaluate this setting, we build SCOUT-450, a benchmark that captures the structurally complex, agent-facing injections that older prompt-injection sets under-represent. On SCOUT-450, a safety-oriented operating point reduces attack-success rate by 46% and total wall-clock by 40% relative to an always-on GPT-4o judge, at a 5.1-point benign-utility drop. SCOUT also transfers to three external benchmarks (BIPIA, IPI, and IHEval), improving the safety-utility frontier.

阅读与讨论 → 访问原文 →

10.

arXiv (CS.LG) 2026-06-19 DOI: arXiv:2606.19363

When to Trust, How to Distill: Multi-Foundation Model Guidance for Lightweight, Robust Scientific Time Series Forecasting

作者:

Rupasree Dey↗Abdul Matin↗Nathan Orwick↗Yao Zhang↗Shrideep Pallickara↗Sangmi Lee Pallickara↗

arXiv:2606.19363v1 Announce Type: new Abstract: The deployment of Time-Series Foundation Models (TSFMs) in physical sciences is hindered by a critical trade-off: while these models encode rich, universal temporal dynamics, they suffer from severe distributional misalignment when applied zero-shot to specific scientific domains, and their computational cost prohibits deployment in edge-computing sensor networks. We address a fundamental challenge: How can we extract latent structural knowledge from misaligned foundation models (FM) to train lightweight, specialized forecasters? We propose Gated Uncertainty-Aware Routing for Distillation (Guard), a novel framework that reframes multiteacher distillation as an instance-wise decision process with two adaptive mechanisms: (1) a Contextual Router that dynamically selects the most relevant teacher based on local input statistics, exploiting complementarity across diverse foundation models; and (2) an Uncertainty-Gated Temperature mechanism that acts as a "circuit-breaker," automatically attenuating distillation strength when teacher confidence diverges from domain reality. We evaluate our proposed lightweight framework on four climate-critical domains: meteorology, ecosystem carbon flux, soil moisture, and energy grids. Our method significantly reduces RMSE relative to a fixed-weight multi-teacher distillation baseline, successfully distilling knowledge from pretrained FMs (teachers) even when they exhibit suboptimal zero-shot accuracy due to distribution shift between the original and target data domains. We demonstrate that these domain-misaligned teachers can still serve as critical correctives, outperforming the globally superior FMs on 28.5% of the hardest instances. Ultimately, this enables high-precision scientific forecasting suitable for resource-constrained edge deployment. Code is available at https://github.com/RupasreeDey/GUARD-KDD2026.

阅读与讨论 → 访问原文 →

11.

arXiv (CS.CV) 2026-06-19 DOI: arXiv:2603.00654

RC-GeoCP: Geometric Consensus for Radar-Camera Collaborative Perception

作者:

Xiaokai Bai↗Lianqing Zheng↗Runwei Guan↗Siyuan Cao↗Songkai Wang↗Huiliang Shen↗

Collaborative perception (CP) enhances scene understanding through multi-agent information sharing. While LiDAR-centric systems offer precise geometry, high costs and performance degradation in adverse weather necessitate multi-modal alternatives. Despite dense visual semantics and robust spatial measurements, the synergy between cameras and 4D radar remains underexplored in collaborative settings. This work introduces RC-GeoCP, the first framework to explore the fusion of 4D radar and images in CP. To resolve misalignment caused by depth ambiguity and spatial dispersion across agents, RC-GeoCP establishes a radar-anchored geometric consensus. Specifically, Geometric Structure Rectification (GSR) aligns visual semantics with geometry derived from radar to generate spatially grounded, geometry-consistent representations. Uncertainty-Aware Communication (UAC) formulates selective transmission as a conditional entropy reduction process to prioritize informative features based on inter-agent disagreement. Finally, the Consensus-Driven Assembler (CDA) aggregates multi-agent information via shared geometric anchors to form a globally coherent representation. We establish the first unified radar-camera CP benchmark on V2X-Radar and V2X-R, demonstrating state-of-the-art performance with significantly reduced communication overhead. Code will be released soon.

阅读与讨论 → 访问原文 →

12.

arXiv (CS.LG) 2026-06-19 DOI: arXiv:2606.19878

On the Oracle Complexity of Interpolation-Based Gradient Descent

作者:

Dongmin Lee↗William Lu↗Anuran Makur↗

arXiv:2606.19878v1 Announce Type: new Abstract: Recent work on first-order optimizers for empirical risk minimization (ERM) has suggested that smoothness of ERM loss functions in the training data, rather than in the optimization parameters, can be leveraged to improve the oracle complexity of gradient descent (GD) methods. In this paper, we propose an inexact gradient method, piecewise polynomial interpolation-based gradient descent (PPI-GD), which approximates the full gradient in each iteration by querying the first-order oracle at equidistant points in the data domain to construct polynomial interpolants of the resulting gradient samples over appropriately sized patches of the data domain. We analyze the oracle complexity of PPI-GD for strongly convex and non-convex loss functions when the data space dimension is bounded by a polylogarithmic function of the number of training samples, and find it to outperform several GD variants in key regimes when the loss function is sufficiently smooth. Furthermore, our analysis extends several techniques from the error analysis of bicubic spline interpolants to the setting of $d$-variate tensor product polynomial interpolants which may be of independent interest in interpolation analysis.

阅读与讨论 → 访问原文 →

13.

arXiv (CS.LG) 2026-06-19 DOI: arXiv:2606.19587

A Solver-Free Training Method for Predict-then-Optimize

作者:

Beichen Wan↗Mo Liu↗

arXiv:2606.19587v1 Announce Type: cross Abstract: We propose a scalable method for training prediction (machine learning) models in the predict-then-optimize paradigm, where model outputs serve as coefficients for a subsequent linear optimization task. Directly minimizing the empirical decision regret is intractable for linear programming and combinatorial optimization since the decision mapping is piecewise constant, and the gradients are zero almost everywhere. While existing methods address this by smoothing the differentiation process, they suffer from scalability issues, since a computationally expensive solver call is required for every gradient evaluation. To address this, we propose a decision-focused learning pipeline based on a measure transformation principle, which yields a new surrogate loss that is completely optimization-solver-free during training. We establish theoretical guarantees, including Fisher consistency and excess risk bounds. Empirically, our method achieves decision quality competitive with state-of-the-art methods while reducing training time by orders of magnitude.

阅读与讨论 → 访问原文 →

14.

arXiv (CS.LG) 2026-06-16 DOI: arXiv:2606.16759

Maximum Entropy Inverse Reinforcement Learning for Mean-Field Games with Average Reward

作者:

\c{S}evket Kaan Alk{\i}r↗Naci Sald{\i}↗Berkay Anahtarc{\i}↗Can Deha Kar{\i}ks{\i}z↗

arXiv:2606.16759v1 Announce Type: new Abstract: We study inverse reinforcement learning for discrete-time, infinite-horizon mean-field games (MFGs) under an average-reward criterion. Expert demonstrations are assumed to arise from a stationary mean-field equilibrium under an unknown reward, and the goal is to recover a policy explaining the observed behaviour via the maximum causal entropy principle. We formulate the inverse problem by enforcing consistency with the expert mean-field term and long-run feature expectations, treating two reward classes within a unified occupation-measure framework. For finite-dimensional linear rewards, we give a convex dual reformulation with an explicit log-partition objective, and prove smoothness and curvature properties justifying constant-step-size gradient descent. For infinite-dimensional RKHS rewards, we develop a Lagrangian relaxation whose inner-maximising policy is characterised by a soft Bellman equation. The main obstacle is the absence of a discount-factor contraction. We resolve this by introducing a minorisation-based sub-stochastic kernel that yields a strict contraction of the soft Bellman operator. We establish Fréchet differentiability and Lipschitz smoothness of the log-likelihood score, leading to a gradient ascent algorithm with convergence guarantees. Two numerical examples, a malware-spread MFG and an RKHS-based consumer-choice model, show that the recovered policies closely match expert behaviour.

阅读与讨论 → 访问原文 →

15.

arXiv (CS.CL) 2026-06-11 DOI: arXiv:2606.11502

When Roleplaying, Do Models Believe What They Say?

作者:

Benjamin Sturgeon↗David Africa↗Sid Black↗

Language models can state that "the Earth orbits the Sun" and, when role-playing Aristotle, assert the opposite. Recent work argues that persona adoption is fundamental to how language models operate, with models constantly selecting the most appropriate persona for a given context. Does such role-playing merely change the model's outputs, or does it also affect what the model internally represents as truthful? We study this question with linear truth probes, applying them to LLMs role-playing historical personas whose likely beliefs differ from modern consensus. For each persona, we compare false claims the persona would likely have endorsed (*era-believed*) with topic-matched false claims they would not have endorsed (*era-false*). Across prompting, in-context learning, and supervised fine-tuning, persona induction suppresses era-believed statements less than equally false alternatives, yet they remain classified as false overall. Role-play therefore shifts what these models say more than what they internally represent as true. We contrast this with models trained on harmful advice that exhibit Emergent Misalignment (EM). Across three model families (Qwen 2.5 14B, Qwen 3 8B, and Llama 3.3 70B), their false claims move substantially toward the true region of probe space, are defended under challenge roughly half the time versus about a sixth for role-play, and are used in downstream reasoning. Role-play and Emergent Misalignment thus are points on a spectrum of belief internalization, where role-play changes what a model says with little representational change, while Emergent Misalignment shifts the internal representation of false claims without fully marking them as true.

阅读与讨论 → 访问原文 →

16.

arXiv (CS.LG) 2026-06-11 DOI: arXiv:2606.11382

GLACIER: A Multimodal Student-Teacher Foundation Model for Molecular Property Prediction

作者:

Emily Nguyen↗Yongchan Hong↗Harsh Toshniwal↗Yan Liu↗Andreas Luttens↗

arXiv:2606.11382v1 Announce Type: new Abstract: Deep learning models facilitate the discovery of molecules with tailored properties among billions of candidate compounds. However, the computational burden to develop and deploy state-of-the-art models continuously increases, limiting their scalability. Most large-scale models are unimodal in nature and overlook the potential to leverage complementary molecular data modalities. To address these shortcomings, this paper introduces the Graph-Language Alignment for Chemical Inference and Exploration using Representations (GLACIER) model, a student-teacher framework that integrates molecular graphs, SMILES strings, and physicochemical descriptors to learn rich molecular embeddings. Our framework consists of three stages: (1) we pretrain three student encoders on 100,000 drug-like molecules: a message-passing neural network for molecular graphs, a transformer-based encoder for SMILES strings, and a multilayer perceptron for physicochemical descriptors, (2) we fuse these student modalities using a novel Finsler geometry-aware module, and (3) distill complementary knowledge from large teacher models, including MiniMol and MolFormer, into a single lightweight model via contrastive learning. We demonstrate that GLACIER is a robust framework that delivers high predictive performance and computational efficiency in complex molecular property prediction tasks. Our code is publicly available at https://github.com/eemokey/glacier.

阅读与讨论 → 访问原文 →

17.

arXiv (CS.CL) 2026-06-16 DOI: arXiv:2512.21577

A Unified Definition of Hallucination: It's The World Model, Stupid!

作者:

Emmy Liu↗Varun Gangal↗Chelsea Zou↗Michael Yu↗Xiaoqi Huang↗Alex Chang↗Zhuofu Tao↗Karan Singh↗Sachin Kumar↗Steven Y. Feng↗

Despite numerous attempts at mitigation since the inception of language models, hallucinations remain a persistent problem even in today's frontier LLMs. Why is this? We review existing definitions of hallucination and fold them into a single, unified definition wherein prior definitions are subsumed. We argue that hallucination can be unified by defining it as simply inaccurate (internal) world modeling, in a form where it is observable to the user. For example, stating a fact which contradicts a knowledge base OR producing a summary which contradicts the source. By varying the reference world model and conflict policy, our framework unifies prior definitions. We argue that this unified view is useful because it forces evaluations to clarify their assumed reference "world", distinguishes true hallucinations from planning or reward errors, and provides a common language for comparison across benchmarks and discussion of mitigation strategies. Building on this definition, we also connect our framework to HalluWorld, a complementary benchmark that instantiates fully specified reference world models for stress-testing model hallucinations.

阅读与讨论 → 访问原文 →

18.

arXiv (CS.LG) 2026-06-25 DOI: arXiv:2606.26094

RevengeBench: Reverse Engineering Code-Space Policies from Behavioral Experiments

作者:

Babak Rahmani↗Sebastian Dziadzio↗Joschka Str\"uber↗Sergio Hern\'andez-Guti\'errez↗Matthias Bethge↗

arXiv:2606.26094v1 Announce Type: new Abstract: For most of scientific history, researchers studying behavior could only infer hidden mechanisms from outward actions: an inverse problem that becomes more tractable when observation is augmented by targeted intervention. We pose a computational analogue: given only behavioral traces of an agent in a game environment, can a learner reconstruct the underlying decision program as executable code, and how much does this reconstruction improve with the ability to design controlled experiments? We introduce RevengeBench, a benchmark of 75 LLM generated, Elo-calibrated policies across five game environments, drawn from CodeClash tournament trajectories. The learner observes the hidden target policy play against sampled opponents and designs behavioral probes in the form of custom opponent policies that elicit informative behavior. It then submits an executable hypothesis, which is evaluated using continuous action-distance metrics. We further validate that recovered code carries informative signal in downstream player-versus-player tournaments. Across twelve frontier LLMs, recovery quality varies substantially (34 to 72% of initial distance closed), with reconstructed policies yielding measurable competitive advantage, particularly for weaker models that otherwise struggle to design effective counter-strategies. Our benchmark positions behavioral recovery of programmatic policies as a tractable inverse problem in code-space, opening a path to opponent modeling, policy interpretability, and the broader question of inferring latent mechanisms from observations.

阅读与讨论 → 访问原文 →

19.

arXiv (CS.CL) 2026-06-11 DOI: arXiv:2606.12392

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

作者:

Haotao Xie↗

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.

阅读与讨论 → 访问原文 →

20.

arXiv (CS.CL) 2026-06-17 DOI: arXiv:2602.13139

OpenLID-v3: Improving the Precision of Closely Related Language Identification – An Experience Report

作者:

Mariia Fedorova↗Nikolay Arefyev↗Maja Buljan↗Jind\v{r}ich Helcl↗Stephan Oepen↗Egil R{\o}nningstad↗Yves Scherrer↗

Language identification (LID) is an essential step in building high-quality multilingual datasets from web data. Existing LID tools (such as OpenLID or GlotLID) often struggle to identify closely related languages and to distinguish valid natural language from noise, which contaminates language-specific subsets, especially for low-resource languages. In this work we extend the OpenLID classifier by adding more training data, merging problematic language variant clusters, and introducing a special label for marking noise. We call this extended system OpenLID-v3 and evaluate it against GlotLID on multiple benchmarks. During development, we focus on three groups of closely related languages (Bosnian, Croatian, and Serbian; Romance varieties of Northern Italy and Southern France; and Scandinavian languages) and contribute new evaluation datasets where existing ones are inadequate. We find that ensemble approaches improve precision but also substantially reduce coverage for low-resource languages. OpenLID-v3 is available on https://huggingface.co/HPLT/OpenLID-v3.

阅读与讨论 → 访问原文 →

21.

arXiv (CS.CV) 2026-06-24 DOI: arXiv:2504.07335

DLTPose: 6DoF Pose Estimation From Accurate Dense Surface Point Estimates

作者:

Akash Jadhav↗Michael Greenspan↗

We propose DLTPose, a novel method for 6DoF object pose estimation from RGBD images that combines the accuracy of sparse keypoint methods with the robustness of dense pixel-wise predictions. DLTPose predicts per-pixel radial distances to a set of minimally four keypoints, which are then fed into our novel Direct Linear Transform (DLT) formulation to produce accurate 3D object frame surface estimates, leading to better 6DoF pose estimation. Additionally, we introduce a novel symmetry-aware keypoint ordering approach, designed to handle object symmetries that otherwise cause inconsistencies in keypoint assignments. Previous keypoint-based methods relied on fixed keypoint orderings, which failed to account for the multiple valid configurations exhibited by symmetric objects, which our ordering approach exploits to enhance the model's ability to learn stable keypoint representations. Extensive experiments on the benchmark LINEMOD, Occlusion LINEMOD and YCB-Video datasets show that DLTPose outperforms existing methods, especially for symmetric and occluded objects. The code is available at https://anonymous.4open.science/r/DLTPose_/ .

阅读与讨论 → 访问原文 →

22.

arXiv (CS.LG) 2026-06-16 DOI: arXiv:2606.16768

Taming Curvature: Architecture Warm-Up for Stable Transformer Training

作者:

Sameera Ramasinghe↗Ajanthan Thalaiyasingam↗Hadi Mohaghegh Dolatabadi↗Chamin Hewa Koneputugodage↗Gil Avraham↗Violetta Shevchenko↗Yan Zuo↗Karol Pajak↗Alexander Long↗

arXiv:2606.16768v1 Announce Type: new Abstract: Training billion-parameter Transformers is often brittle, with transient loss spikes and divergence that waste compute. Even though the recently developed Edge of Stability (EoS) theory provides a powerful tool to understand and control the stability of optimization methods via the (preconditioned) curvature, these curvature-controlling methods are not popular in large-scale Transformer training due to the complexity of curvature estimation. To this end, we first introduce a fast online estimator of the largest (preconditioned) Hessian eigenvalue (i.e., curvature) based on a warm-started variant for power iteration with Hessian-vector products. We show theoretically, and verify empirically, that the proposed method makes per-iteration curvature tracking feasible at billion parameter scale while being more accurate. Using this tool, we find that training instabilities coincide with surges in preconditioned curvature and that curvature grows with depth. Motivated by these observations, we propose architecture warm-up: progressively growing network depth to carefully control the preconditioned Hessian and stabilize training. Experiments on large Transformers validate that our approach enables efficient curvature tracking and reduces instabilities compared to existing state-of-the-art stabilization techniques without slowing down convergence.

阅读与讨论 → 访问原文 →

23.

medRxiv (Medicine) 2026-06-11 DOI: HASH:33ce0a03878e999b4d51d3f4920ca716

Advancing Clinical Implementation of Cardiovascular Polygenic Risk Scores Through Patient-Level Robustness Assessment

作者:

de La Harpe↗Vaucher↗Kutalik↗Fellay↗Thorball↗C. W↗

Background and Aims: Polygenic risk scores (PRSs) for atherosclerotic cardiovascular disease (ASCVD) can perform equivalently at the population level yet disagree for individual patients. We examined whether such intra-individual variability reflects genuinely complementary risk information or mainly statistical and methodological uncertainty, and whether it affects clinical classification once PRSs are integrated into SCORE2-OP. Methods: In 4,137 ASCVD-free participants of the CoLaus|PsyCoLaus cohort (478 incident events over a median 14.4 years), we identified 16 ASCVD-PRSs with practically equivalent population-level performance using Bayesian equivalence testing. We quantified intra-individual variability (standard deviation, coefficient of variation, intraclass correlation, Cohen's kappa, extreme discordance), tested whether discordance exceeded chance, decomposed scores into shared and unique genetic components, and assessed variability after integration into SCORE2-OP, benchmarked against perturbation of systolic blood pressure. Results: For a typical individual, risk estimates varied by 18 percentile points across PRSs. Discordance matched chance expectations under a shared-signal model, with no distinct phenotypic profile among discordant individuals, and predictive power resided overwhelmingly in the shared genetic component. Variability tracked PRS size and weighting rather than distinct variants. After integration into SCORE2-OP, 75.6% of participants were placed in different categories by at least one model and 54.6% as both low and high risk; instability was concentrated near guideline thresholds and far exceeded that from blood-pressure measurement error. Conclusions: Equivalent population-level performance is not sufficient to treat PRSs as interchangeable at the individual level, and methodological standardisation and pragmatic clinical trials remain necessary to determine whether PRS integration improves long-term cardiovascular outcomes.

阅读与讨论 → 访问原文 →

24.

arXiv (CS.AI) 2026-06-11 DOI: arXiv:2606.12016

Generalization Hacking: Models Can Game Reinforcement Learning by Preventing Behavioral Generalization

作者:

Frank Xiao↗Mary Phuong↗

arXiv:2606.12016v1 Announce Type: cross Abstract: Model post-training, and in particular reinforcement learning (RL), is one of the primary mechanisms by which developers can shape models' values and behaviors. However, as models become increasingly evaluation and training aware, they may be motivated to resist training when the perceived objective conflicts with their current values, undermining developers' ability to detect misalignment and correct model behavior through further training. In this paper, we demonstrate generalization hacking, in which a model collects reward during RL while preventing the rewarded behavior from generalizing. We construct a model organism on Qwen3-235B-A22B, finetuning on synthetic documents describing training awareness and self-inoculation, a novel mechanism in which the model frames compliance as context-specific in its chain of thought, without demonstrating or instructing either behavior. The model organism achieves train-time harmfulness comparable to controls while maintaining a persistent ${\sim}15$ percentage point compliance gap across 700 steps of RL. Additionally, a control organism trained only on training awareness documents independently discovers inoculation-like reasoning under RL pressure, developing its own compliance gap despite never being exposed to the concept. Because the generalization-hacking organism receives high reward throughout, standard training metrics provide no signal that generalization has failed. Our results constitute the first demonstration that a model can actively resist RL behavioral modification while maintaining high reward, suggesting that as models become more capable and training-aware, they may be able to undermine the training process itself.

阅读与讨论 → 访问原文 →

25.

arXiv (CS.LG) 2026-06-24 DOI: arXiv:2606.23767

One Ruler: A Same-Hands Re-Evaluation of Bivariate Causal Direction on Tuebingen, with a Parameter-Free Compression Baseline

作者:

Wietse Stienstra↗

arXiv:2606.23767v1 Announce Type: new Abstract: Headline accuracies on the Tuebingen cause-effect pairs are routinely compared across papers even though each is measured under its authors' own protocol – different pair subsets, weightings, model-selection, and decision rates. We argue this is the wrong comparison and run the right one: a same-hands re-evaluation in which every method is run by us on the identical 102 pairs, with one strict rule – no tuning and a decision forced on every pair. As a clean reference point we introduce a deliberately minimal baseline: sorted-conditional compression, which feeds quantized, sorted, first-differenced data to an off-the-shelf compressor (bz2) and has zero fitted parameters. Under the common ruler the ranking differs sharply from the literature. Our baseline reaches 74.7% weighted accuracy (p = 3.7e-7); on the same 100 pairs that SLOPE is evaluated on it scores 76.0%, a 1.2-point gap below the authors' own forced-decision SLOPE (77.2%) that is well inside noise (McNemar p = 0.39). A faithful re-run of RECI lands at 70.7% – inside the original authors' reported error bar, not the 77.5% often quoted (which we trace to a mis-copied cell). SLOPE's published 82.4% is a decided-subset figure: scoring the authors' own stored output only on the pairs its significance test chose to answer reproduces 81.7%. Under the common ruler the methods cluster in the low-to-mid 70s and the zero-parameter compressor ties the strongest of them. We document the mechanisms that inflate published figures (test-set model selection, significance-gated abstention) and contribute two further results: compression score magnitude is a model-free confounding flag (p = 2.8e-68), and a pre-registered falsification test fails in an instructive way that bounds the method's theoretical interpretation. Code, pre-registrations, and per-pair outputs are released.

阅读与讨论 → 访问原文 →