中 EN
登录 注册账户

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。

登录 注册账户
01.
arXiv (CS.AI) 2026-06-16

The Energy Blind Spot: NVIDIA's Flagship Edge AI Hardware Cannot Support Process-Level Energy Attribution

作者:
Deepak Panigrahy↗Aakash Tyagi↗

arXiv:2605.27599v2 Announce Type: replace-cross Abstract: Agentic AI workloads - where a single user goal triggers multi-step orchestration, tool calls, retries, and failure recovery - are being targeted for edge deployment, with NVIDIA, Dell, HP, ASUS, MSI, Acer, and Gigabyte all shipping GB10-based desktop AI systems in 2026. We recently demonstrated that orchestration structure dominates agentic energy cost, with workflows consuming 4.33x more energy per successful goal than linear baselines and OOI reaching 7.63x for multi-step reasoning tasks. Separately, Raj et al. show that CPU-side processing accounts for up to 90.6% of total latency and 44% of total dynamic energy in agentic workloads. We report a systematic energy-observability audit of the ASUS Ascent GX10 (GB10 SoC) and find that the platform exposes no CPU energy counter, no INA power-rail monitor, no IPMI/BMC, and no SCMI powercap protocol through any supported software interface. The only on-device energy telemetry is instantaneous GPU power via NVML. We further discover that the MediaTek firmware already computes per-rail energy internally via an undocumented ACPI interface (SPBM), but NVIDIA states there are "no plans to expose CPU rail information." On-device per-process energy attribution - as performed on x86 via RAPL - is therefore not reproducible on this platform through supported interfaces. We formalize a hardware requirements specification for energy-attributed AI, propose an interim calibration bridge for per-domain energy decomposition - confirmed on the Acer Veriton GN100 where CPU energy accumulators are live - and identify a standards-track path via SCMI powercap. Our findings motivate the low-carbon computing community to demand energy observability as a first-class hardware requirement.

阅读与讨论 → 访问原文 →
02.
medRxiv (Medicine) 2026-06-22

Multisite Real-World Validation of an Electronic Health Record-Integrated Generative Artificial Intelligence Tool for Venous Thromboembolism Risk Stratification

作者:
Baughman↗D. J↗Liu↗Jee↗Young↗Knight↗A. M↗Davis↗Yegnasubramanian↗Najjar↗Whitbread↗J. J↗…

Background: Guiding risk-appropriate inpatient thromboprophylaxis requires venous thromboembolism (VTE) risk stratification; however, reliable risk determination remains inconsistent in routine care. Health systems increasingly pilot artificial intelligence (AI) tools, yet few studies demonstrate rigorous evaluation in the context of a learning health system (LHS). We evaluated the performance of a pilot electronic health record (EHR)-integrated generative AI (GenAI) system, inHealth General Reasoner (iHGR), for VTE risk stratification versus clinician order set classifications and physician-adjudicated chart review. Methods: This multisite retrospective validation study included adult inpatient admissions at Johns Hopkins Medicine between June 21, 2025, and Dec 18, 2025 (checklist-based order set from June 21, 2025 - November 19, 2025, and clinician judgement-based order set from November 29 - December 18, 2025). From 758 eligible admissions, we randomly sampled 500 balanced by site and order set periods. iHGR and clinician-selected order set classifications were compared with the reference standard (RS). Primary outcomes were iHGR sensitivity and specificity. Secondary analyses compared the order sets with the same RS to evaluate workflow comparators and error patterns. Results: iHGR achieved 81.8% sensitivity (95% CI 77.3-85.6) and 70.9% specificity (63.6-77.3). The checklist-based order set had 61.3% sensitivity (53.7-68.5) and 86.2% specificity (77.4-91.9). The clinician judgement-based order set had 78.1% sensitivity (71.3-83.7) and 65.4% specificity (54.3-75.0). False-negative iHGR classifications were associated with missed narrative risk factors. Conclusion: iHGR showed higher sensitivity for VTE risk than checklist-based order sets and clinician judgement without introducing systematic bias. In silico evaluation of pilot AI systems within LHSs can identify clinically important performance trade-offs and implementation targets before operational scale-up. Narrative clinical data abstraction remained a key limitation, supporting the use of GenAI to support rather than supplant clinician judgement.

阅读与讨论 → 访问原文 →
03.
arXiv (CS.LG) 2026-06-19

Weighted Bayesian Conformal Prediction

作者:
Xiayin Lou↗Peng Luo↗

arXiv:2604.06464v2 Announce Type: replace Abstract: Conformal prediction provides distribution-free prediction intervals with finite-sample coverage guarantees, and recent work by Snell \& Griffiths reframes it as Bayesian Quadrature (BQ-CP), yielding powerful data-conditional guarantees via Dirichlet posteriors over thresholds. However, BQ-CP fundamentally requires the i.i.d. assumption. Meanwhile, weighted conformal prediction handles distribution shift via importance weights but remains frequentist, producing only point-estimate thresholds. We propose Weighted Bayesian Conformal Prediction (WBCP), which generalizes BQ-CP to arbitrary importance-weighted settings by replacing the uniform Dirichlet $\Dir(1,\ldots,1)$ with a weighted Dirichlet $\Dir(\neff \cdot \tilde{w}_1, \ldots, \neff \cdot \tilde{w}_n)$, where $\neff$ is Kish's effective sample size. We prove four theoretical results: (1)~$\neff$ is the unique concentration parameter matching frequentist and Bayesian variances; (2)~posterior standard deviation decays as $O(1/\sqrt{\neff})$; (3)~BQ-CP's stochastic dominance guarantee extends to per-weight-profile data-conditional guarantees; (4)~the HPD threshold provides $O(1/\sqrt{\neff})$ improvement in conditional coverage. We instantiate WBCP for spatial prediction as Geographical BQ-CP, where kernel-based spatial weights yield per-location posteriors with interpretable diagnostics. Experiments on synthetic and real-world spatial datasets demonstrate that WBCP maintains coverage guarantees while providing substantially richer uncertainty information.

阅读与讨论 → 访问原文 →
04.
arXiv (CS.CL) 2026-06-19

Analyzing Error Propagation in Korean Spoken QA with ASR-LLM Cascades

作者:
Donghyuk Jung↗Youngwon Choi↗

We analyze how automatic speech recognition (ASR) errors propagate through ASR-LLM cascades in Korean spoken question answering (SQA), focusing on downstream semantic failures that conventional ASR metrics cannot fully capture. Our analysis shows that the relative downstream degradation caused by ASR errors is consistent across LLMs with different absolute performance, suggesting that cascade degradation largely tracks ASR-stage information loss. We further identify single-character Korean ASR errors as a Korean-specific loss channel, where even a minimal transcription difference can change the intended question and degrade downstream QA performance. Finally, an auxiliary comparison shows that a large audio language model outperforms an ASR-LLM cascade with an approximately matched language backbone in noisy Korean SQA, indicating the potential of direct audio input to mitigate transcript-induced information loss.

阅读与讨论 → 访问原文 →
05.
arXiv (CS.CL) 2026-06-19

MixSD: Mixed Contextual Self-Distillation for Knowledge Injection

作者:
Jiarui Liu↗Lechen Zhang↗Yongjin Yang↗Yinghui He↗Yingheng Wang↗Weihao Xuan↗Zhijing Jin↗Mona Diab↗

Supervised fine-tuning (SFT) is widely used to inject new knowledge into language models, but it often degrades pretrained capabilities such as reasoning and general-domain performance. We argue this forgetting arises because fine-tuning targets from humans or external systems diverge from the model's autoregressive distribution, forcing the optimizer to imitate low-probability token sequences. To address this problem, we propose MixSD, a simple external-teacher-free method for distribution-aligned knowledge injection. Instead of training on fixed targets, MixSD constructs supervision dynamically by mixing tokens from two conditionals of the base model itself: an expert conditional that observes the injected fact in context, and a naive conditional that reflects the model's original prior. The resulting supervision sequences preserve the factual learning signal while remaining substantially closer to the base model's distribution. We evaluate MixSD on two synthetic corpora that we construct to study factual recall and arithmetic function acquisition in a controlled setting, together with established benchmarks for open-domain factual question answering and knowledge editing. Across multiple model scales and settings, MixSD consistently achieves a better memorization-retention trade-off compared to SFT and on-policy self distillation baselines, retaining up to 100% of the base model's held-out capability while maintaining near-perfect training accuracy, whereas standard SFT retains as little as 1%. We further show that MixSD produces substantially lower-NLL supervision targets under the base model and reduces harmful movement along Fisher-sensitive parameter directions. These results suggest that aligning supervision with the model's native generation distribution is a simple and effective principle for knowledge injection that mitigates catastrophic forgetting.

阅读与讨论 → 访问原文 →
07.
medRxiv (Medicine) 2026-06-18

Device assessed 24-hour movement behaviour and cardiovascular disease mortality amongst cancer survivors.

作者:
Mitchell↗J. J↗Biswas↗R. K↗Blodgett↗J. M↗Koemel↗N. A↗Ahmadi↗M. N↗Fisher↗Friedenreich↗…

Background: Cancer survivors face elevated risks of mortality from cardiovascular disease (CVD). The potential importance of physical activity (PA) and other behaviours across the 24-hour day (e.g. sedentary behaviour (SB) and sleep) for CVD-mortality risk is not well understood in this at-risk population. Objectives: To assess the importance of 24-hour movement behaviour, using a compositional approach, for mitigating CVD-mortality amongst cancer survivors. Methods: Participants with a prior cancer diagnosis were drawn from the UK Biobank accelerometry sub-study (n=6,158). Accelerometer-derived movement (moderate-to-vigorous PA (MVPA), vigorous PA (VPA), moderate PA (MPA), light PA (LPA), SB, sleep) was examined in relation to CVD-mortality, identified from health record linkage data (using Fine-Gray Cox proportional-hazards models adjusted for demographic, health, lifestyle covariates). Results: Median follow-up was 8.0 years (Q1-Q3: 7.4-8.5), with n=500 (8.2%) deaths (CVD-deaths: n=118). Greater MVPA, in place of any other behaviour, was inversely associated with CVD-mortality with e.g. 10% lower hazard if MVPA theoretically replaced 7 minutes (mins)/day SB (Hazard ratio (HR): 0.91, (95% Confidence Interval: 0.86-0.95)), 9 mins/day LPA (HR: 0.90, 0.83-0.97), or 11 mins/day sleep (HR: 0.90, 0.83-0.97). The VPA component of MVPA proved critical, requiring only ~1-2 additional mins/day for equivalent hazard reduction. Sleep duration, was also inversely associated with CVD-mortality. A 10% lower hazard required replacing 29 mins/day of SB with sleep (HR: 0.90, 0.84-0.96); no other behavioural replacement amongst SB, sleep or LPA could provide an equivalent risk reduction. Conclusions: Among cancer survivors, the most potent reduction in CVD-mortality followed theoretically reallocating time to higher intensity movement.

阅读与讨论 → 访问原文 →
08.
Nature Medicine 2026-06-17

General-purpose chatbots outperform clinical AI tools on physicians’ real-world questions

作者: 未知作者

Specialized clinical AI tools are entering medical practice with little independent testing. In a head-to-head evaluation across two public benchmarks and real questions from physicians, three general-purpose frontier large language models outperformed two leading clinical AI tools, which performed no better than Google search AI overview.

阅读与讨论 → 访问原文 →
09.
arXiv (CS.CL) 2026-06-19

Telenor Nordics Customer Service self-help corpus

作者:
Mike Riess↗

This paper presents a multilingual customer service self-help corpus comprising 1,122 manually validated documents in Finnish, Danish, Norwegian, and Swedish, totaling 274,599 words and 1,884,833 characters. The documents have been sourced from the public self-help pages of four Nordic telecommunications operators and subsequently filtered for person-identifiable information and relevance through a combined LLM and human annotation pipeline. Domain-specific datasets for Nordic languages remain scarce, particularly in customer service: a domain of growing importance for retrieval-augmented generation, cross-lingual transfer learning, and emerging agent-based service architectures. An analysis of the corpus reveals substantial variation in document length and structure across operators, reflecting distinct editorial strategies, as well as broad topical coverage spanning network hardware, mobile services, TV and streaming, billing, and account management. The dataset is publicly available under a CC-BY-NC-SA-4.0 license at https://zenodo.org/records/20732652, intended to support reproducible research in Nordic NLP and information retrieval.

阅读与讨论 → 访问原文 →
10.
arXiv (CS.AI) 2026-06-16

Sensory Restoration via Brain-Computer Interfaces: A Unified 2 x 2 Framework and Convergence Roadmap

作者:
Xuan-The Tran↗

arXiv:2606.15091v1 Announce Type: cross Abstract: Millions of individuals worldwide suffer from sensory and communication deficits caused by neurodegenerative diseases, stroke, or trauma. Brain-computer interfaces (BCIs) offer a promising avenue for sensory and motor restoration. However, the scientific literature remains highly fragmented between invasive neuroprosthetics and non-invasive electrophysiological decoders, with a lack of consistent terminology and comparison metrics. This chapter proposes a unified 2 x 2 framework categorizing BCIs along two axes: degree of invasiveness (invasive vs. non-invasive) and signal direction (afferent sensory-IN vs. efferent sensory-OUT). We define and distinguish the paradigms of restoration, substitution, and augmentation. Furthermore, we outline a structural roadmap for the convergence of these modalities over near-, medium-, and long-term horizons, focusing on physical limits and the integrative role of machine learning foundation models.

阅读与讨论 → 访问原文 →
11.
arXiv (CS.CV) 2026-06-18

Pyramid Self-Contrastive Learning for Single-shot Test-time Ultrasound Image Denoising

作者:
Jiajing Zhang↗Bingze Dai↗Xi Zhang↗Yue Xu↗Wei-Ning Lee↗

The inherent electronic and speckle noise complicates clinical interpretation of ultrasound images. Conventional denoising methods rely on explicit noise assumptions whose validity diminishes under composite noise conditions. Learning-based methods are usually pretrained in a limited image domain using a labeled dataset, which implies inevitable domain shift in complex in vivo environments. This study proposes a Pyramid Self-Contrastive Learning (PSCL) framework for test-time ultrasound image denoising without pretraining. Given multiple noisy samples from only one-shot imaging, PSCL disentangles anatomical similarity and noise randomness into separate pyramid latent spaces. The clean image is then decoded from the anatomy space while discarding the noise space. We first apply PSCL to synthetic aperture ultrasound (SAU), where an Aperture-to-Aperture loop serves as a self-supervised proxy task to ensure denoising fidelity. Simulation experiments, including noise levels from 0 to 30 dB and inclusion geometries from simple to complex, demonstrated improvements of 69.3% in SNR and 34.4% in CNR. The in vivo results showed 84.8% SNR and 25.7% CNR gains using only two aperture data of the heart in six echocardiographic views, liver, and kidney. PSCL delivers clear images across diverse imaging targets and configurations, paving the way for more reliable anatomical visualization without domain shift and pretraining costs.

阅读与讨论 → 访问原文 →
13.
arXiv (CS.CV) 2026-06-15

Giving AI a Headache: Acoustic Adversarial Attacks to Computer Vision Applications

作者:
Nicole Villavicencio-Gardu\~no↗Maksim Ekin Eren↗Milo Prisbrey↗Ben Migliori↗Michael Teti↗

Artificial Intelligence (AI) is increasingly used to automate a variety of real-world computer vision (CV) applications, such as autonomous vehicle control, facial recognition, and security cameras. Recent research has shown that acoustic vibration can induce real physical motion in cameras, interfering with their internal stabilization mechanisms. Because the motion falls outside the conditions the stabilization system was designed to handle, the system introduces artifacts into the frame, causing AI-based CV models to misclassify, miss targets, or hallucinate objects. Previous work used ultrasonic frequencies (>20 kHz) to perform short-range attacks, which limits them to short distances due to the attenuation exhibited by high frequencies. In this work, we investigate acoustic attacks using lower frequencies in the audible range (

阅读与讨论 → 访问原文 →
14.
arXiv (CS.LG) 2026-06-18

Hierarchical Planning with Latent World Models

作者:
Wancong Zhang↗Basile Terver↗Artem Zholus↗Soham Chitnis↗Harsh Sutaria↗Mido Assran↗Randall Balestriero↗Amir Bar↗Adrien Bardes↗Yann LeCun↗Nicolas Ballas↗

arXiv:2604.03208v2 Announce Type: replace Abstract: World models are a promising path to zero-shot embodied control through planning. However, existing world model planners struggle on long-horizon, multi-stage tasks: prediction errors compound and naive search is exponential in the planning horizon. Hierarchy mitigates both by decomposing tasks into shorter, tractable subproblems; yet prior hierarchical approaches either amortize control into task-specific policies (hierarchical RL) or assume low-dimensional states and known dynamics (classical hierarchical MPC). We present Hierarchical Planning with Latent World Models (HWM), an architecture and planning paradigm for hierarchical model predictive control (MPC) directly on visual world models trained solely via next-latent prediction. HWM learns world models at multiple temporal scales within a shared latent space, so predictions from the long-horizon model serve as subgoals for the short-horizon model via latent matching, without task-specific rewards, skill learning, or hierarchical policies. To keep long-horizon search tractable, HWM learns an action encoder that compresses primitive action chunks into latent macro-actions. On real-world Franka manipulation, HWM solves pick-and-place from a single goal image at 70% success vs. 0% for single-level planning. Across simulated push manipulation and maze navigation, HWM consistently improves performance on long-horizon tasks while requiring up to 3x less planning compute.

阅读与讨论 → 访问原文 →
15.
arXiv (CS.CL) 2026-06-16

Distilling Examples into Task Instructions: Enhanced In-Context Learning for Real-World B2B Conversations

作者:
Guy Rotman↗Adi Kopilov↗Danit Berger Zalmanson↗Omri Allouche↗

In-context learning (ICL) is the standard method for low-resource classification, yet its efficacy in specialized domains remains largely unexplored. We address the challenge of classifying semantically complex, multi-party B2B conversations, where traditional ICL encounters significant limitations, especially as context length increases due to the concatenation of multiple few-shot examples. We introduce the \texttt{Call Playbook} dataset, featuring five classification tasks derived from real-world B2B conversations targeting core sales concepts. To bridge the gap between performance and practical utility, we propose novel knowledge extraction methods that distill verbose examples into compact, interpretable representations of structured classification criteria and precise task descriptions. Our approach achieves a 99\% reduction in token usage and improves macro-averaged AUC by up to 7\% over traditional ICL. Notably, it remains robust as context grows, unlike advanced token compression baselines which degrade by over 9 F1 points. Importantly, our framework enables direct refinement of classification logic, addressing critical needs for transparency, efficiency, and user interaction in real-world NLP applications.

阅读与讨论 → 访问原文 →
16.
arXiv (math.PR) 2026-06-11

On the Wasserstein distance between a hyperuniform point process and its mean

作者:
Raphael Butez↗Sandrine Dallaporta↗David Garc\'ia-Zelada↗

arXiv:2404.09549v3 Announce Type: replace Abstract: We study the existence of bounds on the expected $p$-Wasserstein distance between a random measure and its mean under the assumption that the $p$-th centered moments of the counting statistics are controlled uniformly in space. The average Wasserstein transport cost is shown to be bounded from above and from below by some multiples of the number of points. $D$-dimensional versions of those results are also obtained. As a corollary, we prove that for any value of $p\geq 1$ the Ginibre point process can be seen as a perturbed lattice with identically distributed perturbations with a finite $p$-th moment.

阅读与讨论 → 访问原文 →
17.
medRxiv (Medicine) 2026-06-22

Multi-omics data fusion reveals divergent molecular signatures of intra-articular micro-fragmented adipose tissue and hyaluronic acid treatment in inflammatory-phenotype knee osteoarthritis

作者:
Primorac↗Molnar↗Brlek↗Bulic↗Jelec↗Prosenc Zmrzljak↗Klaric↗Lauc↗Malod-Dognin↗Przulj↗

Knee osteoarthritis (KOA) affects an estimated 374 million people worldwide and has no approved disease-modifying treatment. Intra-articular micro-fragmented adipose tissue (MFAT) outperformed hyaluronic acid (HA) on patient-reported outcomes in our recent double-blind randomized trial (ISRCTN88966184), yet the molecular basis of this differential efficacy is unknown, and the two interventions have not previously been compared at the level of their in vivo molecular response in human KOA. Here we apply an interpretable artificial-intelligence data-fusion framework, based on non-negative matrix tri-factorization, to longitudinally collected plasma from this cohort, integrating proteomics, N-glycomics, miRNA transcriptomics and patient genetics with prior protein-protein and miRNA-gene regulatory networks at baseline, one and six months. The framework jointly decomposes all data modalities at each timepoint into shared, interpretable factors, from which we derive data-driven pathways of genes and of miRNAs and recover new patient-gene and patient-miRNA associations. These pathways were biologically coherent, showing significant enrichment in Gene Ontology Biological Process and Reactome Pathway annotations. By six months, the two treatments left clearly distinct molecular signatures: HA remained dominated by canonical OA pathogenic processes, including cartilage-degrading effectors such as MMP13 and LIMK2 and markers of synovial inflammation, whereas MFAT shifted the systemic landscape toward chondroprotection, anti-inflammatory signalling and bone-cartilage homeostasis, with prioritized effectors including SIRT7 and NDUFC1. To our knowledge, these are the first systems-level molecular data directly comparing the in vivo response to the two treatments in human KOA, providing initial evidence that MFAT acts as a disease-modifying intervention and demonstrating the value of interpretable data fusion for uncovering treatment mechanisms in small translational cohorts.

阅读与讨论 → 访问原文 →
18.
arXiv (CS.CL) 2026-06-12

From Benchmarks to Skills: Low-Rank Factors for LLM Evaluation

作者:
Aviya Maimon↗Amir DN Cohen↗Gal Vishne↗Shauli Ravfogel↗Reut Tsarfaty↗

Current evaluations of large language models (LLMs) rely heavily on a growing collection of benchmarks and on aggregate benchmark scores, yet it remains unclear what this comparison actually captures, and what these scores reveal about models' underlying capabilities. Here, we propose a new paradigm for LLM evaluation, by asking whether benchmark performance reflects many independent abilities, or rather relies on a small number of shared dimensions. To answer this, we apply Factor Analysis (FA) to a massive performance matrix of LLMs versus benchmarks \((60\times44)\) revealing an intrinsically low-rank structure of that matrix. That is, a small number of latent factors captures most of the structure in the full task space. This low-rank geometry reveals substantial redundancy across existing tasks and explains why many benchmarks appear to be measuring overlapping abilities. We further show that these latent factors correspond to coherent, skill-like, dimensions of LLM behavior. Leveraging this latent skill-space, we deliver three practical tools for LLM evaluation and downstream users: (i)~identifying redundant tasks, (ii)~profiling new models using a small subset of tasks, and (iii)~selecting models aligned with desired skill profiles. Our method provides a solid alternative to the de-facto standard of a single aggregate score, and establishes an interpretable and practical framework for understanding and benchmarking LLM core capabilities.

阅读与讨论 → 访问原文 →
19.
arXiv (CS.CV) 2026-06-16

Scribby: A Multi-Level LLM Framework for Semantic Video Analysis

作者:
Julian Abelarde↗Hugo Garrido-Lestache Belinchon↗

As video content continues to expand across educational platforms, recorded lectures, and live-streamed entertainment, the need for efficient and structured analysis of long-form footage has increased [1]. Although many existing AI programs provide high-level video summaries based on AI-generated transcripts [2,3,4,5], these approaches are often limited to coarse overviews and lack detailed analysis of a video's structure, thematic progression, and semantic relationships, all of which are required for comprehensive video analysis. This paper proposes an LLM-based video summarization framework that balances macro-level comprehension with micro-level semantic analysis [6,12,13]. The first stage of the process indexes the video at a micro level by (1) analyzing the full transcript, (2) analyzing individual transcript sentences, and (3) grouping these sentences by semantic similarity using an LLM as a judge [6,13]. Contextual continuity is retained during sentence-level processing by incorporating both the global transcript analysis and adjacent sentence information into each evaluation prompt. This framework establishes a foundation for video analysis tools that visualize semantic chunking and semantic matching through relevance-based heatmaps. Limitations and future expansions of the framework are also discussed.

阅读与讨论 → 访问原文 →
20.
arXiv (CS.LG) 2026-06-17

Noise-Driven Escape from Metastable Phases explains Grokking in Deep Neural Networks

作者:
Ibrahim Talha Ersoy↗Karoline Wiesner↗

arXiv:2606.17120v1 Announce Type: new Abstract: Deep neural networks (DNNs) exhibit first order phase transitions under variations of the L2 regularization strength, with each transition marking the onset of a new learnable feature. Below a critical regularization strength, all features are in principle learnable, but coexisting metastable states, separated by energy barriers, can trap the network and impede convergence. A strength of DNNs is their ability to generalize. But many open questions remain, among them the origin of so called grokking: the abrupt, delayed onset of generalization after prolonged apparent overfitting. We show for linear DNNs that grokking is consistent with hysteresis in first-order L2 phase transitions: using L2 regularization to engineer deliberate trapping, we demonstrate that a model in a low-accuracy metastable state escapes only when SGD noise drives it across an energy barrier, with escape times following Arrhenius scaling. We reproduce grokking-like delayed convergence across two orders of magnitude in escape time by deliberately trapping models in metastable phases. Using sparse sub-sampling we also reproduce the canonical grokking curve where test error eventually approaches the final training error. Our work suggests that the number of metastable states equals the number of learnable features – one per singular value of the data covariance – the potential for hysteresis grows naturally with task complexity. We provide evidence that the same mechanism likely operates in general nonlinear DNNs. Our results provide routes toward more efficient learning schemes.

阅读与讨论 → 访问原文 →
21.
arXiv (CS.CL) 2026-06-11

Overcoming State Inertia in Full-Duplex Spoken Language Models via Activation Steering

作者:
Cheng-Kuang Chang↗Kai-Wei Chang↗Alexander H. Liu↗James Glass↗

Full-duplex spoken language models (FD-SLMs) enable seamless speech interaction by allowing models to listen and speak simultaneously, yet the internal mechanism by which they coordinate listening and speaking remains underexplored. We analyze the predictive behavior encoded in FD-SLM hidden representations and find that they exhibit stream-specific predictive patterns: during listening, they preferentially predict the incoming user stream, whereas during speaking, they preferentially predict the model output stream. Building on this observation, we show that FD-SLMs dynamically modulate their internal predictive focus between two states: a generative state aligned with model output generation and a perceptive state aligned with incoming user input. However, this modulation can lag behind abrupt changes in conversational context. During user interruptions, the model remains transiently biased toward the generative state before transitioning into the perceptive state, causing it to miss the beginning of the incoming input. We term this delayed internal transition state inertia. To quantify its downstream impact, we introduce the Zero-Buffer Benchmark (ZBB), a diagnostic benchmark for evaluating immediate interruption comprehension when user speech begins abruptly. We evaluate this setting using response correctness and initial-word occurrence rate (IWOR). Finally, we mitigate state inertia through activation steering with a perception vector, a training-free intervention with little additional computational overhead. Across multiple state-of-the-art FD-SLMs, activation steering substantially improves interruption handling; for example, on PersonaPlex, it improves correctness from 28% to 45% and IWOR from 40% to 72% without any fine-tuning.

阅读与讨论 → 访问原文 →
22.
arXiv (CS.LG) 2026-06-16

Tail-Shape Estimation in LLM Evaluation Is Fragile: A Protocol for Diagnosing False Positives

作者:
Luca Zhou↗

arXiv:2606.16511v1 Announce Type: new Abstract: Recent work motivates moving large language model (LLM) evaluation from mean-based to tail-aware metrics, including conditional value-at-risk and tail-index estimates of reward-model error. We ask whether the canonical extreme-value-theory tail-index parameter, which isolates how heavy a tail is from how large the tail mass is, adds discriminative information beyond the mean and a standard tail-magnitude statistic in LLM evaluation. We pre-register a protocol covering admissibility, goodness-of-fit, threshold-stability, and effect-size requirements for any positive tail-shape claim. The protocol is the contribution of this paper; the empirical study below is a demonstration of what its gates catch. Applied to a standard LLM toxicity-evaluation setup under two structurally different scorer families, the protocol catches three distinct modes of false positives that a naive analysis would have published, and rejects the headline tail-shape claim on both scorers. We conclude that tail-shape estimation in the LLM toxicity-evaluation setups we examined is more fragile than the recent literature suggests, and recommend the protocol as a starting point for tail-index claims in similar setups.

阅读与讨论 → 访问原文 →
23.
arXiv (CS.CL) 2026-06-16

When Cognitive Graphs Meet LLMs: BDEI Cognitive Pathways for Panic Emotional Arousal Prediction

作者:
Mengzhu Liu↗Long Qin↗Chuan Ai↗Zhengqiu Zhu↗Hongru Liang↗Chen Gao↗Yong Li↗Xin Lu↗Quanjun Yin↗

Predicting individual panic emotional arousal timing before manifestation is essential for proactive emergency intervention. Existing methods incorporate cognitive elements but none explicitly model the emotional arousal process, making them ill-suited for emotional arousal timing prediction. We argue that grounding prediction in appraisal emotion theory is necessary because it explicitly models this process, but three problems must be solved. (1) Appraisal theory posits that emotion arises from simultaneous evaluation across multiple threat dimensions, yet no prior work fuses these inputs into risk perception. (2) Existing cognitive models lack an Emotion node, decoupling threat appraisal from emotional arousal and forcing emotions to be inferred indirectly from behaviors. (3) Given their generalizable cognitive reasoning, current approaches adopt LLMs as the primary decision-maker, yet overlook the fragility and hallucination-proneness of their outputs. To address these issues, we introduce PanicCognitivePath (PCP), a framework that addresses all three. A Psychological Safety Distance (PSD) model, grounded in psychological distance theory, maps four-domain signals into a unified risk metric as the entry condition for subsequent cognitive reasoning. An explicit Emotion node grounded in appraisal emotion theory is introduced into BDI, forming a Belief-Desire-Emotion-Intention (BDEI) pathway. Agents whose risk metric exceeds the PSD threshold enter this pathway, coupling threat appraisal directly to emotional arousal. The BDEI pathway governs all state transitions while the LLM is confined to parameter estimation for the Belief-to-Desire transition, confining hallucinations to a single step and preventing error propagation. Experiments on Hurricane Sandy show PCP improves arousal timing accuracy by 10.68% over baselines, reduces peak count error to 7.07%.

阅读与讨论 → 访问原文 →
24.
arXiv (CS.CL) 2026-06-16

Hidden Ghost Hand: Unveiling Backdoor Vulnerabilities in MLLM-Powered Mobile GUI Agents

作者:
Pengzhou Cheng↗Haowen Hu↗Zheng Wu↗Zongru Wu↗Tianjie Ju↗Zhuosheng Zhang↗Gongshen Liu↗

Graphical user interface (GUI) agents powered by multimodal large language models (MLLMs) have shown greater promise for human-interaction. However, due to the high fine-tuning cost, users often rely on open-source GUI agents or APIs offered by AI providers, which introduces a critical but underexplored supply chain threat: backdoor attacks. In this work, we first unveil that MLLM-powered GUI agents naturally expose multiple interaction-level triggers, such as historical steps, environment states, and task progress. Based on this observation, we introduce AgentGhost, an effective and stealthy framework for red-teaming backdoor attacks. Specifically, we first construct composite triggers by combining goal and interaction levels, allowing GUI agents to unintentionally activate backdoors while ensuring task utility. Then, we formulate backdoor injection as a Min-Max optimization problem that uses supervised contrastive learning to maximize the feature difference across sample classes at the representation space, improving flexibility of the backdoor. Meanwhile, it adopts supervised fine-tuning to minimize the discrepancy between backdoor and clean behavior generation, enhancing effectiveness and utility. Extensive evaluations of various agent models in two established mobile benchmarks show that AgentGhost is effective and generic, with attack accuracy that reaches 99.7\% on three attack objectives, and shows stealthiness with only 1\% utility degradation. Furthermore, we tailor a defense method against AgentGhost that reduces the attack accuracy to 22.1\%. Our code is available at \texttt{anonymous}.

阅读与讨论 → 访问原文 →
25.
arXiv (CS.LG) 2026-06-11

IAPO: Input Attribution-Aware Policy Optimization for Tool Use in Small Multimodal Agents

作者:
Yifan Yang↗Zhen Zhang↗Jiayi Tian↗Liyan Tan↗Zheng Zhang↗

arXiv:2606.11652v1 Announce Type: new Abstract: This paper investigates reinforcement learning (RL) methods for improving tool-calling capabilities in multimodal small language model (SLM) agents. While existing works have explored various reward designs to improve agentic tool-calling ability, these approaches face inherent limitations for SLM training, especially under multimodal scenarios. First, many existing methods evaluate tool use correctness through exact matching against certain ground-truth or predefined formats. However, this assumption is often unsuitable for multimodal tasks, where multiple tool use paths may be valid and annotated tool trajectories are typically unavailable. Second, such sparse and brittle binary rewards provide little guidance on how to improve the underlying decision process, making them particularly difficult for multimodal SLM to learn from. To address these issues, we propose Input Attribution-Aware Policy Optimization (IAPO), an RL algorithm for improving tool use in multimodal SLM by aligning the model's attribution across input components with that of a stronger teacher. Experiments on Qwen2.5-VL-3B show that the proposed method improves visual question answering accuracy by an average of 3% across six test sets compared with existing visual tool use work, by helping the model attend to the most relevant input evidence.

阅读与讨论 → 访问原文 →