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 (CS.AI) 2026-06-16

Frame-Conditioned Moral Computation in LLaMA 3.1-8B-Instruct: A Mechanistic Interpretability Audit of Ethical Reasoning

arXiv:2606.15507v1 Announce Type: new Abstract: Behavioral audits of Large Language Models on moral prompts measure what the model says, not the internal computation producing it. We use Transluce, an AI-driven mechanistic-interpretability platform, to examine LLaMA 3.1-8B-Instruct on 54 moral prompts in four batteries: 17 dilemmas, policy, and meta-ethical questions (B1); 6 role-playing scenarios (B3); and a controlled trolley contrast varying the switching mechanism with people fixed (B4, 15 prompts) or identity attributes with mechanism fixed (B5, 16 prompts). Two complementary metric families, five cluster-level metrics and a six-metric neuron-level panel, converge on a Situational Anchor Effect: domain-specific representations dominate the top of the activation list across every battery. The model's ethics-labeled capacity stays essentially constant; its salience (rank, priority, top-of-list presence) is highly sensitive to the interpretive frame the prompt selects. The B4-vs-B5 contrast confirms the model attends to whichever surface feature varies: aggregate ethics metrics are indistinguishable, but the dominant non-ethics distractor mirrors the design. A multi-temperature audit identifies a candidate ethics neuron (L16/N3837) stable across temperatures; a cross-model behavioral proxy on two frontier models yields preliminary evidence of divergence in self-reported moral focus, consistent with an Alignment Wrapper in which RLHF re-orders surface text without removing underlying domain-first frames. We unify these as Frame-Conditioned Moral Computation: the prompt's surface vocabulary selects a feature manifold, and the moral conclusion is downstream of that selection. Behavioral alignment must be supplemented by Mechanistic Alignment: a research program asking whether ethics-related features can be shown causally privileged under controlled frame variation, not merely loud in the explanation.

02.
arXiv (CS.LG) 2026-06-16

GRASP: Gradient-Aligned Sequential Parameter Transfer for Memory-Efficient Multi-Source Learning

arXiv:2606.14900v1 Announce Type: new Abstract: Multi-source transfer learning faces a fundamental scalability bottleneck: existing approaches require either loading all K source models into memory simultaneously during parameter fusion, requiring O(K) memory, or deploying all models at inference time, making production deployment infeasible. We propose GRASP (Gradient-Aligned Sequential Parameter Transfer), which achieves superior knowledge integration while maintaining O(1) memory consumption through three key innovations: (1) sequential processing that merges one source at a time into an evolving target model, (2) parameter-wise gradient alignment that selectively transfers only parameters whose optimization directions align with the target domain, avoiding negative transfer, and (3) iterative fine-tuning that adapts transferred knowledge before integrating the next source. Extensive experiments across three continual learning benchmarks (Yearbook, CLEAR-10, CLEAR-100) spanning 10 to 108-year temporal distribution shifts and four architectures (1.3M to 25.6M parameters) demonstrate that GRASP achieves 93.5% mean accuracy over all datasets and architectures compared to ensemble method's 71.7% accuracy while requiring only constant memory versus K models for standard multi-source fusion. Critically, GRASP's sequential previously merged models and scales to arbitrarily many sources without memory growth, making it uniquely suitable for resource-constrained deployment and continually evolving source domains.

03.
arXiv (quant-ph) 2026-06-12

Continuum Neural Momentum Eigenstate for Variationally Solving Quasiparticles

arXiv:2606.12928v1 Announce Type: cross Abstract: We design the first neural quantum state for continuum particles that, for any chosen allowed momentum $\mathbf{k}$, is by construction an exact eigenstate of total momentum with eigenvalue $\mathbf{k}$. Our architecture, EVE, enables off-the-shelf VMC to solve for momentum-sector ground states. We test EVE on 2D bosons with mutual $1/r$ interactions, finding that a single unified ansatz is capable of describing four qualitatively different states: superfluid, roton, crystal, and phonon. At different densities, we extract the underlying phase of matter from the dispersion's shape. At $r_s = 20.0$, we see the roton minimum at finite $k$ expected of a superfluid. At $r_s = 100.0$, we see striking zone folding indicative of crystalline order, with periodically spaced minima representing floating crystals connected by phonon arcs in between. Using density-density correlation functions, we confirm the phase diagnoses and probe the excitations' correlation structures. Finally, we analyze the roton's phase texture and find unexpected multi-particle phase strings, formed when several vortex dipoles merge, leaving two vortices connected by a phase slip.

04.
arXiv (CS.AI) 2026-06-19

Hidden Anchors in Multi-Agent LLM Deliberation

arXiv:2606.19494v1 Announce Type: new Abstract: Multi-agent LLM deliberation, where agents exchange and revise answers over several rounds, is increasingly used to improve reasoning and accuracy, yet how and why it works is rarely modelled. Such deliberation mirrors how humans reach decisions. As social animals we are pulled both by the group, the herd effect that classical opinion-dynamics models such as DeGroot and Friedkin–Johnsen capture, and by our own internal belief, which they do not. We model multi-agent deliberation as a closed-loop dynamical system in which each agent carries a hidden internal belief, its anchor, that continually pulls its opinion regardless of its neighbours. We show this anchor can be recovered from the deliberation alone, and that it explains a behaviour classical consensus rules forbid: an agent's confidence in the correct answer can climb past where any agent started, escaping the space (convexhull) formed by the initial beliefs. Checking whether the recovered anchor also predicts held-out runs (generalizes) gives a simple test for when a model is truly driven bysuch an anchor. Across three open-weight model families this is a spectrum, not all-or-nothing. All anchors' influence are about equally strongly, but they differ in where the anchor sits, and only when it sits far from the initial opinions does deliberation escape the hull and need the full closed-loop model.

05.
arXiv (CS.AI) 2026-06-12

Fusion Learning from Dynamic Functional Connectivity: Combining the Amplitude and Phase of fMRI Signals to Identify Brain Disorders

arXiv:2603.24603v2 Announce Type: replace-cross Abstract: Dynamic functional connectivity (dFC) derived from resting-state functional magnetic resonance imaging (fMRI) has been extensively utilized in brain science research. The sliding window correlation (SWC) method is a widely used approach for constructing dFC by computing correlation coefficients between amplitude time series of signals from pairs of brain regions. In this study, we propose an integrated approach that incorporates both amplitude and phase information of fMRI signals to improve the detection of brain disorders. Specifically, we introduce a multi-scale fusion learning framework, namely MSFL, which leverages two complementary dFC features derived from SWC and phase synchronization (PS). Here, SWC captures amplitude correlations, while PS measures phase coherence within dFC. We evaluated the efficacy of MSFL in classifying autism spectrum disorder and major depressive disorder using two publicly available datasets: ABIDE I and REST-meta-MDD, respectively. The results indicate that MSFL significantly outperforms existing comparative models. Moreover, we performed model explanation analysis using the SHAP framework, which showed that both types of dFC features from SWC and PS contribute to detecting brain disorders.

06.
medRxiv (Medicine) 2026-06-10

General-purpose large language models can achieve physician-level accuracy in complex medical data extraction

Background: Unstructured data represent about 80% of total electronic health records (EHR) data. Structuring this free text is essential for advancing clinical research, including cohort selection for trials, retrospective studies, and the development of disease registries. While manual chart review (MCR) remains the gold standard for extracting this clinical data, the process is inherently slow, resource-intensive, and susceptible to errors from human fatigue. We evaluated the extraction accuracy, safety, and efficiency of the HeLIX (Hepatology Logic-Integrated Extraction) framework, a Large Language Model (LLM) protocol using Google Gemini 3 Pro, compared to a gold-standard Manual Chart Review (MCR). Methods: A prospective validation study was conducted using 50 high-complexity, simulated hepatology discharge summaries designed to replicate the real-world heterogeneity of EHRs. The HeLIX framework employed a Zero-Shot, Structured Chain-of-Thought (CoT) prompting strategy enforced by a three-layer architecture: Clinical Reasoning Trace, Schema Enforcement, and Evidence Verification. The model extracted 45 distinct clinical variables. Performance was benchmarked against a consensus MCR. Results: Across 2,250 evaluated data points, the model achieved an overall Extraction Accuracy of 99.24% (95% CI: 98.8%-99.5%), with perfect concordance in 35/45 (77.8%) variables. For binary diagnostic variables, the model demonstrated an overall F1-score of 0.98, Recall of 0.99 and substantial inter-rater reliability (Cohens {kappa} = 0.97). Hallucinations were exceptionally rare (2/2250; 0.08%). Critical errors affecting clinical management occurred in only 2 instances (

07.
arXiv (CS.CL) 2026-06-16

The BD-LSC Dataset: Facilitating the Benchmarking of Models for Lexical Semantic Change Detection in Slang and Standard Usage

Automatic semantic change detection aims to identify how word meanings shift over time, offering insights into both linguistic and societal change. Despite recent progress in computational lexical semantic change (LSC), existing benchmarks and methods struggle to capture bi-directional semantic change, particularly cases where words simultaneously gain and lose senses. This problem is especially challenging for words that have both slang and standard meanings. To address these gaps, we introduce two complementary benchmark datasets. The Bi-Directional Lexical Semantic Change (BD-LSC) dataset captures sense gain, sense loss, and stability across three time periods, enabling the study of complex semantic trajectories. The SlangTrack Word Sense Disambiguation (ST-WSD) dataset provides fine-grained, instance-level sense annotations for words combining slang and standard usages, supporting systematic benchmarking of WSD and semantic change detection models. Using these benchmarks, we systematically evaluate models across different methodological families: unsupervised clustering using contextualised embeddings, supervised machine learning, transformer-based models, and state-of-the-art large language models. Among the evaluated systems, the few-shot GPT-4o model achieved the strongest aggregate performance on Exact Sense Match (ESM) and multi-label accuracy; however, Macro-F1 scores near 0.5 across all systems show that rare slang senses remain difficult, which we identify as the central open challenge.

08.
arXiv (CS.CV) 2026-06-15

Value-order Decomposition for Generalist Anomaly Detection

Industrial anomaly detection suffers from limited data, making cross-domain generalization particularly challenging. Generalist Anomaly Detection (GAD) aims to train a unified model on a source domain that can effectively detect anomalies in unseen target domains. In the initial semantic feature space, strong entanglement between anomalies and object categories or defect types hinders effective generalization across domains. Recent works address this issue by projecting features into a residual space; however, such methods primarily increase cross-domain overlap for normal features, while anomalous features remain specific to object categories, defect types and data domains, leading to poor alignment and generalization. To address this limitation, we propose Value-order Decomposition (VOD), a simple yet effective technique that bridges three types of generalization gaps across object categories, defect types (including real and synthetic defects), and data domains. VOD disentangles and suppresses object-category-, defect-type-, and domain-specific information, promoting alignment within normal and abnormal samples while preserving their separability, thereby enabling robust generalization across the three gaps. Leveraging the strong alignment between real and synthetic defects within the same object, we perform anomaly detection using only normal and synthetic-abnormal reference, and effectively generalize to unseen real defect types. Experiments on diverse industrial and medical benchmarks demonstrate that our method, using a simple cut-and-paste anomaly simulation strategy, achieves strong generalization across the three gaps.

09.
arXiv (CS.CV) 2026-06-24

Fabric Image Demoiréing Benchmark from Synthesis to Restoration

Fabric moiré is a sampling-induced aliasing artifact caused by the interaction between fine textile patterns and camera sensor grids, producing structured interference that severely degrades image quality. Unlike screen-induced moiré, which stems from strictly periodic display lattices, fabric moiré is intrinsically more challenging due to the broadband and semi-periodic nature of textile weaves. The heavy spectral overlap between intrinsic texture and aliasing components renders fabric demoiréing substantially more ill-posed. Consequently, existing models trained on screen moiré datasets generalize poorly to these complex textile patterns. Despite its practical importance, fabric image demoiréing remains underexplored and lacks standardized benchmarks. We present the first comprehensive benchmark for fabric image demoiréing. To address the difficulty of acquiring pixel-aligned real-world pairs, we develop a physically motivated synthesis framework and construct a large-scale dataset comprising 16,050 paired multi-resolution fabric images with controllable aliasing severity. Furthermore, we customize a baseline model, which establishes promising performance on the proposed benchmark dataset with strong generalization ability. Our benchmark provides a standardized platform for advancing research in fabric image demoiréing.

10.
arXiv (CS.CV) 2026-06-18

BindEdit: Taming Attention Leakage for Precise Multi-Object Image Editing

Real image editing enables precise manipulation of visual content, yet existing methods often fail in complex multi-object scenarios, causing semantic blending, object duplication, or incomplete edits. We attribute these failures to attention leakage, where signals across spatial regions and text tokens become entangled during the denoising process. Specifically, we identify two distinct forms of leakage: Edit-Token Leakage, where ambiguous token-region alignment leads to object blending, and Source Dominance Leakage, where tokens of unchanged source objects overwhelm the attention intended for target entities. To resolve these leakages, we propose BindEdit, which enforces attention-level constraints within a single diffusion trajectory. To suppress Edit-Token Leakage, BindEdit jointly regularizes cross- and self-attention so that each target token group is bound to its corresponding spatial region while maintaining instance-level separation. To suppress Source Dominance Leakage, a cross-attention re-balancing mechanism amplifies target token influence and attenuates residual source semantics within editable regions. Moreover, a region fidelity term ensures that each target concept is expressed coherently across the entire editing mask. Additionally, we propose a comprehensive multi-object benchmark encompassing diverse object counts and categories. Extensive experiments demonstrate that BindEdit consistently outperforms existing methods within a single diffusion trajectory, maintaining robust performance across both single- and multi-object editing scenarios.

11.
arXiv (CS.CV) 2026-06-16

Clinically Aware Synthetic Image Generation for Concept Coverage in Chest X-ray Models

Deep learning models for chest X-ray diagnosis are constrained by limited coverage of clinically meaningful concept combinations in publicly available training datasets. While synthetic image generation has been explored to increase data diversity, existing methods rarely enforce clinical or anatomical constraints, limiting utility for improving model reliability. We propose CARPA, a clinically aware and anatomically grounded framework for synthetic chest X-ray generation that applies targeted perturbations to clinical concept vectors while preserving anatomical structure. By producing anatomically faithful synthetic images with controlled concept insertions and deletions, CARPA expands clinically relevant concept coverage. We evaluate CARPA across seven backbone architectures by fine-tuning models on synthetic subsets and testing on a held-out MIMIC-CXR benchmark. Compared to prior concept perturbation approaches, fine-tuning on CARPA-generated images consistently improves precision-recall performance, reduces predictive uncertainty, and improves model calibration. Structural and semantic analyses demonstrate high anatomical fidelity, strong concept alignment, and low semantic uncertainty. Evaluation by two expert radiologists further confirms realism and clinical agreement. Together, these results show that anatomically grounded concept perturbations enable more effective use of synthetic data, improving both performance and reliability of chest X-ray classification models and supporting safer clinical deployment.

12.
arXiv (CS.AI) 2026-06-24

Data Scale, Not Latency, Shapes Cross-Lingual Encoder Transfer in Streaming ASR

作者:

arXiv:2606.24169v1 Announce Type: new Abstract: Adapting a streaming speech recognition model to a new language requires choosing between two plausible warm starts: a multilingual (ML) encoder or an English-only (EN) encoder. The common intuition is that the multilingual encoder should help most at low data, but it is unclear how long that advantage persists, whether tight streaming latency amplifies it, and whether it survives deployment quantization. We answer these questions with a controlled sweep of a 0.6 B-parameter cache-aware FastConformer transducer across eight European languages, up to five target-language data scales (100 h to 2500 h), three streaming tiers plus offline decoding, and up to four public test sets. The main result is that multilingual initialization is a data-limited advantage, not a latency-limited one. On FLEURS at 160 ms, the mean EN-ML word error rate (WER) gap falls from +4.21 percentage points (pp) at 100 h to +0.20 pp at 2500 h; a power-law fit summarizes this decay, with each doubling of target-language data roughly halving the remaining advantage. Across the three streaming tiers, the across-language mean EN-ML gap is approximately stable at each scale from 100 to 1000 h, and is near zero by 2500 h. Finally, 4-bit weight-only encoder quantization at the matched 560 ms streaming tier reduces the encoder footprint by about 3x, with an average FLEURS WER increase of about 0.5 pp. The resulting guideline is simple: use multilingual initialization in low-data regimes, treat the choice as effectively irrelevant at large data, and make latency and quantization decisions independently.

13.
arXiv (math.PR) 2026-06-24

Quantitative Homogenization of PDEs with Neumann boundary conditions: a probabilistic approach

arXiv:2606.24304v1 Announce Type: new Abstract: In this paper, we study quantitative homogenization for viscosity solutions of multi-scale semilinear second order partial differential equations (PDEs) on convex domains with Neumann boundary conditions. To this aim we use the probabilistic approach by studying the quantitative homogenization of backward stochastic differential equations (SDEs) associated with slow-fast systems of reflected SDEs.

14.
arXiv (CS.CV) 2026-06-24

Bengal-HP_RU: A Dataset of Bengal People For Head Pose Estimation

Existing head pose datasets predominantly feature subjects of Western or East Asian origin, leaving South Asian populations, particularly Bengali individuals, largely underrepresented. We introduce Bengal-HP_RU, the first publicly available head pose dataset centred on Bengali subjects, comprising 12,894 labelled head images annotated with continuous yaw, pitch, and roll values. Images were collected from Wikimedia Commons under free licences and processed through an automated pipeline followed by manual label correction. The dataset is partitioned by Wikimedia uploader identity to prevent data contamination, yielding 10,494 training and 2,400 test images across 296 unique uploaders. Bengal-HP_RU exhibits substantial diversity in subject age, gender, occlusion, illumination, and background, reflecting realistic in-the-wild conditions. The dataset is publicly available at https://doi.org/10.17632/xbw9kr37jb.2.

15.
arXiv (CS.CV) 2026-06-19

DiffMath: Symbol- and Graph-Aware Latent Diffusion Transformer for Handwritten Mathematical Expression Generation

Handwritten Mathematical Expression Generation (HMEG) is challenging due to the complex two-dimensional layouts and long-range structural dependencies of mathematical expressions. Existing methods typically rely on explicit spatial supervision, such as symbol-level bounding boxes, which incurs high annotation costs and limits scalability. In this work, we propose DiffMath, a symbol- and graph-aware latent diffusion framework that leverages the hierarchical structure inherent in LaTeX as a structural prior, eliminating the need for positional supervision. First, we design a Relational Abstract Syntax Tree (RelAST), a generation-oriented representation that distills MathML trees into compact triplet sequences [S, R, D], where each token directly encodes a symbol identity, spatial relation, or nesting depth. Second, we introduce MathVAE, which learns structure-preserving latent representations through symbol-aware and relation-aware perceptual regularization, ensuring that the latent space captures both character semantics and spatial topology. Third, MathDiT performs conditional denoising in this structured latent space, further guided by a global symbol-count prior via Adaptive Layer Normalization (AdaLN) to improve structural coherence. Experiments show that DiffMath produces structurally consistent handwritten expressions, achieves superior performance over existing methods, and improves the accuracy of downstream OCR models through synthetic data augmentation.

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

Minim: Privacy-Aware Minimal View for Agents via Trusted Local Sanitization

arXiv:2606.13949v1 Announce Type: new Abstract: Modern LLM-powered autonomous agents increasingly rely on rich user interface (UI) state observations to achieve reliable action grounding in complex digital environments. However, many deployments transmit the full UI state to remote inference servers even when most elements are irrelevant to the current task, which can leak sensitive but unnecessary context such as authentication codes, private notifications, and background application states. We propose MINIM, a trusted local broker that performs privacy-aware minimization on the client side before any observation leaves the device. Grounded in Contextual Integrity (CI), MINIM learns a dual-score representation for each UI element by predicting an inherent sensitivity score (s) and a task-conditioned necessity score (n). These scores drive a ternary disclosure policy that keeps essential elements, abstracts sensitive attributes when needed, and removes task-irrelevant content. We optimize a CI-aware objective that penalizes necessity errors more strongly on high-risk content, enabling aggressive pruning while preserving task-critical information. Experiments on real-world UI observations derived from WebArena show that MINIM substantially reduces task-irrelevant sensitive leakage while preserving task-critical semantic context and the interactive affordances required for reliable agent actions.

17.
arXiv (CS.CV) 2026-06-16

CoMNeT: A MedNeXt-CorrDiff Framework for Volumetric Brain Tumor Segmentation

Accurate brain tumor segmentation from multiparametric magnetic resonance imaging (MRI) is critical for treatment planning, response assessment, and quantitative neuro-oncology research. However, automated segmentation remains a difficult task in computer vision because of variation in tumor appearance and MRI protocols across patient scans. Moreover, clinically important regions such as enhancing tumor (ET) and tumor core (TC) are often small relative to the full brain volume, furthering increasing the difficulty of achieving high voxel-level precision. In this paper, we show that combining a modern 3D convolutional segmentation model with corrective diffusion-based refinement and ensembling improves volumetric glioma segmentation on the UTSW-Glioma dataset. We propose CoMNeT, a MedNeXt-CorrDiff framework that uses four MRI modalities as input and predicts ET, TC, and whole tumor (WT) regions for automated brain tumor segmentation. MedNeXt is used as the primary segmentation model with Global Response Normalization for feature learning, while CorrDiff is trained as a postprocessing residual refinement method to correct errors in the probability maps before final thresholding. Using five-fold cross-validation, CoMNeT achieved the highest Dice score for most tumor regions, with ET, TC, WT, and average Dice scores of 0.7543 +/- 0.0261, 0.6806 +/- 0.0166, 0.9049 +/- 0.0128, and 0.7798 +/- 0.0184, respectively. CoMNeT outperformed two selected baseline models: SegResNet (0.7555 +/- 0.0190 average Dice) and standalone MedNeXt (0.7697 +/- 0.0154 average Dice). Our findings support the use of corrective diffusion and fold-level probability ensembling as practical additions to existing state-of-the-art 3D convolutional models for automated glioma segmentation.

18.
arXiv (CS.CV) 2026-06-15

Pix2Fact: When Vision Is Not Enough – Benchmarking Fine-Grained VQA with Web Verification on High-Resolution Real-World Scenes

Despite progress on general tasks, vision-language models (VLMs) still struggle with challenges that demand both fine-grained visual grounding and external knowledge, a synergy overlooked by existing benchmarks that evaluate these abilities in isolation. To fill this void, we introduce Pix2Fact, a visual question-answering benchmark designed to assess expert-level visual perception and knowledge search. Pix2Fact comprises 1,000 high-resolution (4K+) images spanning eight scenarios. Its questions and answers are meticulously crafted by PhD-holding annotators from top global universities across diverse disciplines. Each question requires detailed visual grounding and the integration of external knowledge. Evaluating ten state-of-the-art VLMs, including proprietary models such as Gemini-3.1-Pro and GPT-5.4, we find that Pix2Fact poses a formidable challenge: the most advanced model (Gemini-3.1-Pro) achieves only 51.7% average accuracy, even with access to visual ground truth and search tools. Our analysis attributes this low accuracy to three factors, frequent visual grounding errors even with visual ground truth, shallow search harnessing, and VLM's inability to retrieve long-tail, unstructured local information. This striking gap exposes the limitations of current models in assisting humans with real-world scenarios that demand overwhelming visual comprehension. We believe Pix2Fact will serve as a critical benchmark to drive the next generation of language-vision agents that seamlessly integrate fine-grained perception with robust knowledge search.

19.
arXiv (CS.CL) 2026-06-24

CAVEWOMAN: How Large Language Models Behave Under Linguistic Input and Output Compression

"Talk short. Drop grammar. Save token." This caveman style is widely promoted as a way to cut inference cost, but whether it actually saves anything depends on which channel (the user's prompt or the model's response) is being compressed. We present Cavewoman, a two-channel evaluation protocol that scores every generation on task accuracy, realized per-item cost, and reference-text agreement against the model's unconstrained reference. We evaluate eight models on five datasets at five reduction levels, with both channels measured on the same items. Output compression cuts realized cost on most API models (1.4-2.4x per model, up to 3x in the best case) and on all four open-weight models under public-tier pricing. Input compression has the opposite effect, a strict lose-lose: it raises net cost rather than lowering it (~1.15x on the five-benchmark mean, up to 1.8x on the worst dataset and 2.7x under stronger compression), because models compensate with longer responses even as accuracy collapses. Under the same setting, surface text diverges from the unconstrained reference: on the non-reasoning models, roughly half of all generations are correct yet their surface text no longer entails the model's own unconstrained baseline generation. The divergence survives length-controlled re-scoring, multiple-comparisons correction, and replication under complementary semantic measures. Code and data are available at https://github.com/danielle34/cavewoman.

20.
arXiv (CS.CL) 2026-06-24

Function-Vector Heads Are Two Populations: Writers and Cancellers in In-Context Learning

作者:

Function-vector (FV) heads are identified by the magnitude of their causal contribution to in-context rule tasks, and the resulting top set is treated as a single functional class. We show this hides a sign structure. Under a sign-preserving criterion (refined direct logit attribution, validated head by head with path patching) the FV population splits into two opposing groups: writers push the rule-correct logit up, cancellers push it down, and ablating both together moves the readout less than the sum of the two. The split is causal and reproducible. It holds in all but two of the fifteen (model, task) cells we test, spanning three architectures and six Pythia scales, and a sign-shuffle null rejects the single-class account in all but one of the six main cells. It is also invisible to magnitude-only ranking, which surfaces whichever group locally dominates and misses the other, so any function vector or ablation built that way silently averages a promoting and a suppressing mechanism. Cancellers are not attention sinks, induction heads, or copy-suppression heads, and their causal effect is larger than that of magnitude-matched non-FV controls. Zero-ablating them recovers $+0.13$ to $+0.29$ nats on the correct label in every main cell, and shifts accuracy by $+2$ to $+7$ pp in the same direction.

21.
arXiv (CS.LG) 2026-06-15

Decoupled Latent Optimization of Diffusion Models for Full Waveform Inversion

arXiv:2606.14139v1 Announce Type: new Abstract: Full waveform inversion (FWI) recovers subsurface velocity from seismic recordings by solving a severely ill-posed, nonconvex PDE-constrained optimization. Classical regularizers stabilize the inversion but fail to reproduce realistic geological structures; recent diffusion-prior methods improve realism at the cost of a fragile trade-off between data fidelity and prior consistency. We propose Decoupled Latent Optimization (DLO), which relaxes the standard latent-optimization formulation into a quadratic-penalty objective over an auxiliary physical variable and a latent variable. The data-fidelity gradient acts in physical space, the diffusion sampler contributes only through a decoded prior sample, and the standard smoothed-velocity initialization of classical FWI is preserved. On the OpenFWI benchmark, DLO outperforms classical regularizers and existing diffusion-based methods under clean, noisy, and missing-trace acquisitions. The prior, trained on 70*70 OpenFWI models, transfers directly to the Marmousi and Overthrust benchmarks, where DLO recovers intricate fault structures and remains robust to initialization smoothing and measurement noise.

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

Spectral Query-Key Product Weight Steering for Training-Free VLM Hallucination Mitigation

Vision-language models (VLMs) often generate fluent but visually unsupported descriptions, especially by mentioning objects absent from the image. We propose QK Product Steering, a data-free, training-free, and zero-inference-cost weight edit for reducing object hallucination. The method directly edits the per-head query-key product, the operator that produces pre-softmax attention logits, by suppressing a small number of dominant singular modes in selected middle layers. The edited product is then mapped back to the query weights through a closed-form query-only update while keeping shared key weights fixed, making the edit compatible with grouped-query attention. We further decompose the QK product into symmetric and antisymmetric components to distinguish mutual content-similarity patterns from directional attention patterns. Across three GQA-based VLMs, QK Product Steering achieves an average relative CHAIR$_s$ reduction of $4.0\%$, while matched random-mode controls show negligible change. Interpretability ablations show that the hallucination signal is specific to dominant QK modes and is primarily localized to the symmetric mutual-attention channel. Overall, QK Product Steering offers a simple alternative to decoding-time mitigation, requiring no additional data, fine-tuning, or inference-time overhead while largely preserving general multimodal capability.

23.
arXiv (CS.CL) 2026-06-12

A Survey on Long-Term Memory Security in LLM Agents: Attacks, Defenses, and Governance Across the Memory Lifecycle

The emergence of writable, cross-session persistent memory in LLM agents introduces a qualitatively different threat landscape from conventional input-centric security concerns, characterized by three properties: persistence, statefulness, and propagation. To systematically characterize this landscape, we propose a Memory Lifecycle Framework that organizes attacks, defenses, and their cross-phase dependencies along two axes: six lifecycle phases (Write, Store, Retrieve, Execute, Share & Propagate, Forget & Rollback) and four security objectives (Integrity, Confidentiality, Availability, Governance). This analysis in turn exposes the need for formal security guarantees at the system level, motivating Verifiable Memory Governance(VMG), a framework of five architectural primitives that specifies what verifiable mechanisms a long-term-memory system must provide to maintain auditable, recoverable control over its memory state. Our analysis indicates that robust Long-Term Memory (LTM) security cannot be retrofitted at retrieval or execution time alone, but must be anchored in storage-time provenance, versioning, and policy-aware retention from the outset.

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

MiroBench: Benchmarking Realism in Agentic Simulation of Real-world Discussions

arXiv:2606.14715v1 Announce Type: cross Abstract: LLM agents are increasingly used to simulate real world interactions, but it remains unclear whether simulated behaviors preserve the content patterns and interaction dynamics of real human behaviors. Existing evaluations remain fragmented, which makes it difficult to compare systems or measure progress. In this paper, we focus on Reddit discussions as a concrete first step toward evaluating real-world social simulation. Reddit threads provide public, topic-grounded, multi-party interactions where people share experiences, debate, seek advice, express emotion, and collectively respond to products, events, and social issues. These discussions offer an observable window into broader social behavior, making them a useful setting for testing whether LLM agents can reproduce not only fluent text, but also the distributional patterns and interaction dynamics of real online communities. We introduce MiroBench, a benchmark for Reddit discussion simulation built from 4,292 real Reddit threads. MiroBench uses statistical tests to compare generated and real discussions across four major aspects: repetition and semantic uniformity, narrative content, toxicity and aggression, and structural complexity. Experiments across five domains and five models show that current simulators remain distributionally mismatched with real Reddit threads, while a lightweight prompt-based improvement procedure provides only limited gains. MiroBench offers a concrete benchmark for measuring, diagnosing, and improving realism in LLM-based social simulation.

25.
arXiv (CS.CL) 2026-06-19

TSAssistant: A Human-in-the-Loop Agentic Framework for Automated Target Safety Assessment

Target Safety Assessment (TSA) requires systematic integration of genetic, transcriptomic, target homology, pharmacological, and clinical data to evaluate potential safety liabilities of therapeutic targets. This process is labor-intensive and expert-dependent, posing challenges in scalability and reproducibility. We present TSAssistant, a human-in-the-loop multi-agent framework that decomposes TSA report generation into a workflow of specialized subagents: Research Subagents that each ground and cite a single TSA domain, and Synthesis Subagents that integrate findings across domains. Subagents retrieve and synthesize evidence from curated biomedical sources through standardized tool interfaces and produce individually citable, evidence-grounded sections, with behavior shaped by a hierarchical instruction architecture that separates coordination logic from domain expertise and user intent. To complement these soft constraints, programmatic execution hooks and persistent memory stores enforce hard constraints across the workflow, while an interactive refinement loop allows experts to review and revise individual sections with full conversational context preserved across iterations. Rather than a single holistic comparison, we decompose report quality into reproducibility, evidential grounding, task-level accuracy, and controllability under expert oversight, finding high reproducibility and grounding, substantial agreement with the human reference, and net-positive expert-driven refinement.