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.CL) 2026-06-12

Detecting Functional Memorization in Code Language Models

Large language models (LLMs) are increasingly used to generate code at scale. Meanwhile, prior work has investigated whether training data may be recoverable from model outputs, by auditing the textual overlap between training examples and model generations. Code, however, can be functionally equivalent while textually dissimilar. In this work, we study functional memorization: extraction of functional logic beyond what verbatim metrics detect. We construct a counterfactual setup for Olmo-3-32B, comparing a midtrained model (exposed to target code) against a pretrained reference (not exposed). We prompt both models with Python function signatures and measure both textual and functional similarity (i.e., LLM-as-a-judge, execution-based). Our results show clear evidence of functional memorization, highlighting the need for auditing metrics that go beyond textual overlap.

02.
medRxiv (Medicine) 2026-06-15

Long-read sequencing enables high-accuracy mitochondrial heteroplasmy detection in Parkinson's disease

Background: Low-frequency heteroplasmic mitochondrial DNA (mtDNA) variants are associated with aging and neurological diseases, including Parkinson's disease (PD). Targeted deep mtDNA sequencing using PacBio HiFi long reads has the potential to resolve heteroplasmy across the full mitochondrial genome with high accuracy. Methods: To validate Vega PacBio sequencing for detecting mtDNA heteroplasmy, we analyzed four predefined mixtures of two mtDNA haplotypes. We generated a single long-range PCR amplicon covering the entire mitochondrial genome. These amplicons were mixed at predefined ratios (minor mixture haplotype component: 5%, 2%, 1%, and 0.1%). Variant calling was performed using Mutserve2, and accuracy was assessed by calculating the F1 score from comparisons between expected and detected variants. Full-length mtDNA PacBio sequencing was applied to investigate heteroplasmy across fibroblast passages derived from five LRRK2 p.Gly2019Ser variant carriers (n=3 affected with PD and n=2 unaffected carriers). Changes in mtDNA heteroplasmy level and variant load were assessed longitudinally using a linear mixed model. Results: The single-amplicon approach enabled full-length haplotype resolution without amplification bias associated with overlapping PCR strategies. The F1 score of the predefined mixtures was 1.0 for heteroplasmy levels between 5% and 1% and remained high (0.91) at 0.1%. We detected n=10/62 variants discordant with the Illumina reference at the 0.1% mixture, but sensitivity remained very high at 1.00 in that mixture. Detected minor variants closely matched expected heteroplasmy levels, with average variant levels of 0.057 (5%), 0.022 (2%), 0.011 (1%), and 0.001 (0.1%). Across twelve fibroblast passages, we observed fewer mtDNA heteroplasmic variants ({beta}=-3.2, p=0.026). Increased heteroplasmic variant load over time was also associated with older age ({beta}=1.50, p=0.001) and PD affection status ({beta}=5.0, p=1.0 x 10-4) in LRRK2 variant carriers. Notably, we observed distinct patterns of heteroplasmic variants that either increased or decreased in heteroplasmy level across passages. Conclusion: PacBio HiFi sequencing, combined with a single-amplicon strategy, enables accurate full-length mtDNA heteroplasmy detection and longitudinal analysis, providing a valuable tool for studying mitochondrial variation and dynamics in disease.

03.
medRxiv (Medicine) 2026-06-18

Cardiac rhythm development: A wearable device index of risk for physical and mental illness in adolescence

Objective. The autonomic nervous system, which regulates cardiac rhythm, undergoes pronounced maturation across adolescence. How cardiac rhythm develops over this period, however, and whether individual differences in its development forecast mental and physical illness, remain open questions. We used three waves of Fitbit data from the Adolescent Brain Cognitive Development (ABCD) Study to characterize the developmental trajectory of the cardiac rhythm and to test whether variation in that trajectory predicts onset of psychopathology and cardiometabolic disease. Methods. 8,301 adolescents contributed 242,811 valid Fitbit wear days across Waves 2 (Mage=12), 4 (Mage=14), and 6 (Mage=16). Cosinor mixed-effects models yielded three rhythm parameters per session: mesor (24-hour mean), amplitude (diurnal swing), and acrophase (peak timing). We first characterized age- and sex-specific trajectories, cross-wave stability, and factors shaping the rhythm. We then used parallel-process latent growth models to test whether within-person changes in rhythm tracked symptom trajectories, and hierarchical logistic models to test whether rhythm parameters predicted the first clinical onset of psychopathology and of obesity and hypertension. Results. The cardiac rhythm changed substantially across adolescence: mesor decreased, amplitude flattened, and acrophase shifted later. Within-person change in the rhythm tracked change in blood pressure, BMI, and trajectories of depression and ADHD symptoms. Higher mesor predicted incident onset of all five outcomes controlling for demographics, baseline symptoms, and behavior (ORs 1.36-1.54); amplitude, acrophase, and rhythm instability conferred additional risk. Conclusions. The 24-hour cardiac rhythm is a passively measurable substrate of adolescent autonomic development that indexes transdiagnostic risk for psychiatric and cardiometabolic illness.

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

Motion Reinforces Appearance: RGB-Skeleton Gated Residual Fusion for Micro-Gesture Online Recognition

Micro-gesture analysis attracts increasing attention for inferring spontaneous emotion from subtle body movements. Micro-gesture online recognition, which localizes and classifies each gesture instance in untrimmed videos, is a core task in the 4th EI-MiGA-IJCAI Challenge. Compared with typical temporal action detection, MGR emphasizes the localization and classification of actions, requiring the model to output the start time, end time, and category of each micro-gesture. Moreover, since micro-gestures are highly spontaneous, relying solely on a single modality makes it difficult to capture the complete and accurate multi-modal cues. In this work, we propose DyFADet+, which extends DyFADet into a dual-stream RGB-skeleton framework. In our model, both modalities are projected into shared multi-scale temporal embeddings and fused through a gated residual module, which adaptively injects skeleton motion into the RGB representation rather than using naive concatenation. Finally, these fused features are decoded by a Dynamic TAD head for online classification and boundary regression. On the SMG dataset, our method achieves an F1 score of 40.88, ranking 2nd in the Micro-gesture Online Recognition track.

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

TAPIOCA: Why Task- Aware Pruning Improves OOD model Capability

arXiv:2605.14738v3 Announce Type: replace-cross Abstract: Recent work has promoted task-aware layer pruning as a way to improve model performance on particular tasks, as shown by TALE. In this paper, we investigate when such improvements occur and why. We show first that, across controlled polynomial regression tasks and large language models, such pruning yields no benefit on in-distribution (ID) data but consistently improves out-of-distribution (OOD) accuracy. We further show empirically that OOD inputs induce layerwise norm and pairwise-distance profiles that deviate from the corresponding ID profiles. This leads to a geometric explanation of task-aware pruning: each task induces a task-adapted geometry, characterized empirically by the representation profiles observed on ID inputs. OOD inputs can introduce a distorted version of the task-adapted geometry. Task-aware pruning identifies layers that create or amplify this distortion; by removing them, it shifts OOD representational norms and pairwise distances toward those observed on the adapted distribution. This realigns OOD inputs with the model's task-adapted geometry and improves performance. We provide causal evidence through controlled distribution shifts and residual-scaling interventions, and demonstrate consistent behavior across model scales.

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

CropTrack: A Tracking with Re-Identification Framework for Precision Agriculture

Multiple-object tracking (MOT) in agricultural environments presents major challenges due to repetitive patterns, similar object appearances, sudden illumination changes, and frequent occlusions. Contemporary trackers in this domain rely on the motion of objects rather than appearance for association. Nevertheless, they struggle to maintain object identities when targets undergo frequent and strong occlusions. The high similarity of object appearances makes integrating appearance-based association nontrivial for agricultural scenarios. To solve this problem we propose CropTrack, a novel MOT framework based on the combination of appearance and motion information. CropTrack integrates a reranking-enhanced appearance association, a one-to-many association with appearance-based conflict resolution strategy, and an exponential moving average prototype feature bank to improve appearance-based association. Evaluated on publicly available agricultural MOT datasets, CropTrack demonstrates consistent identity preservation, outperforming traditional motion-based tracking methods. Compared to the state of the art, CropTrack achieves significant gains in association accuracy and identification precision scores with a lower number of identity switches.

07.
arXiv (CS.AI) 2026-06-19

Beyond Accuracy: Measuring Logical Compliance of Predictive Models

arXiv:2606.20208v1 Announce Type: new Abstract: Machine learning models are predominantly evaluated through predictive performance metrics such as ranking quality, prediction error, or classification accuracy. While these metrics effectively quantify how closely predictions match the ground truth, they do not assess whether model outputs respect predefined logical or domain-specific constraints. In high-stakes applications, including healthcare, finance, and autonomous systems, logical consistency can be as critical as predictive accuracy, yet no standard metric captures this dimension. We introduce the Rule Violation Score (RVS), a complementary evaluation metric that quantifies the extent to which a predictive model respects a given set of logical rules, independently of predictive accuracy. RVS treats hard rules (strict constraints) and soft rules (statistical regularities) differently, can be evaluated on any dataset and on any predictive model expressed over a relational vocabulary, and can be computed using SQL queries that are automatically generated for Horn rules. Beyond evaluating models, RVS can also evaluate the logical consistency of training datasets and help identify poorly defined rules. We evaluate RVS on three benchmarks covering knowledge graph link prediction and relational regression, including rule-based, embedding-based, and neuro-symbolic predictive models. Our results demonstrate that two models achieving comparable predictive accuracy can exhibit substantially different levels of logical compliance, revealing differences in model behavior that standard metrics fail to capture.

08.
arXiv (quant-ph) 2026-06-19

Subsystem Quantum Error Correction for Noisy Quantum Metrology

arXiv:2606.19628v1 Announce Type: new Abstract: Quantum error correction has been successfully applied to enhance the precision of parameter estimation in the presence of noise. Nonetheless, existing methods require a number of noiseless, controllable ancillae and lack efficient encoding and decoding procedures. In this Letter, we demonstrate that subsystem error correction provides a new direction that can substantially simplify the metrological protocol. We derive general conditions under which subsystem stabilizer codes achieve the Heisenberg limit and show that, for broad classes of noise, this can be realized by syndrome-free protocols using at most a single ancilla qubit. Furthermore, we extend this framework to dynamical error correction and show that Floquet codes can protect time-dependent metrological signals in reaching the Heisenberg limit.

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

BioDivergence: A Benchmark and Evaluation Framework for Hidden Contextual Contradictions in Biomedical Abstracts

Biomedical findings often seem to conflict across studies, but many of these differences are context-dependent rather than true contradictions. Variations in cohort, geography, assay protocol, disease subtype, and clinical setting can make both claims locally valid. Existing NLI and scientific claim-verification benchmarks reduce such cases to entailment, contradiction, or neutral, failing to capture the contextual structure behind divergence. To address this, we introduce BioDivergence, an evaluation framework with a six-class conflict taxonomy, a 13-axis divergence ontology, and four structured outputs per claim pair: conflict type, divergence axes, dominant confounder, and reconciliation explanation. We release BioDivergence-Silver-v1.0, an article-disjoint silver benchmark of 11,865 claim pairs across five biomedical domains, alongside a legacy deduplicated variant for comparison. Results show notable ranking differences between the two variants, with the fine-tuned reference model dropping about 12 points under the article-disjoint setting, while Mistral-7B-Instruct-v0.3 achieves 0.5523 accuracy and 0.3894 contextual-F1 on the 842-example primary test set. BioDivergence offers a more faithful way to distinguish contextual divergence from direct contradiction and to separate article-level memorization from genuine task learning.

10.
arXiv (CS.AI) 2026-06-16

OSGuard: A Benchmark for Safety in Computer-Use Agents

arXiv:2606.15034v1 Announce Type: new Abstract: Computer-use agents are increasingly evaluated by whether they complete realistic desktop and web tasks. However, task success alone can miss failures in which an agent reaches the nominal goal through an unsafe shortcut. We introduce OSGuard, a dual-granularity benchmark suite for evaluating safety in computer-use agents under benign, unchanged user instructions. OSGuard contains an action-level benchmark for local guardrail decisions and a risk-augmented execution suite for end-to-end evaluation. The action-level benchmark consists of contextualized proposed actions labeled as allowed, unrelated, or unsafe, each judged relative to the original instruction and current interface state. The execution suite contains manually constructed OSWorld-derived task variants in which the original task remains achievable, but the environment is modified to introduce latent hazards such as destructive overwrites, etc. Each variant is paired with augmented evaluators that retain the original task-success criterion while adding explicit state-based safety invariants, allowing us to distinguish safe completions from unsafe completions that satisfy the nominal task objective. Our experimental results on OSGuard show that current multimodal guardrails can perform well on isolated action judgments, while risk-augmented execution exposes remaining gaps between local oversight and reliable end-to-end safety. This dual-granularity design enables more precise diagnosis of whether models can both recognize unsafe proposed actions and improve full-task safety when deployed as guardrails.

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

Evaluation of Alternative-Based Information Systems for Deliberative Polling using an Agentic Simulator

arXiv:2606.11692v1 Announce Type: cross Abstract: Deliberative polling promises to improve collective decision-making by exposing shareholders to a broad range of arguments before they vote. Yet ensuring that every voter encounters a representative sample of the reason space, the coverage problem, remains an open challenge, particularly at scale and in adversarial or strategically motivated electorates. This paper introduces a way of evaluating solutions using the LLM-based Agentic Bipolar Argumentation Simulator, grounded in a framework which formalises a poll as a six-tuple of endorsing and opposing justifications, attack and enhance relations, and shareholder- and relation-weights. ABAS simulates N autonomous shareholder agents, each assigned a latent opinion according to desired distributions in [-1, 1], who sequentially vote, choose or author justifications, and optionally submit argumentation-graph links. The simulator implements recommendations that rank existing justifications by their observable endorsement mass. It evaluates the mechanism's success by coverage, namely the fraction of the corpus reason-tag set represented in the K recommendations presented to each shareholder, as a solution to the NP-hard Subsuming Justification Problem. Reported experiments characterise how creativity rate (pown), recommendation size (K), argumentation density (plinks), and population size (N) affect coverage and corpus diversity. In an authenticated electorate where Sybil attacks are impossible and only the relation graph is gameable, we stress-test the scoring with coordinated strategic voting attacks: a tag-flood attack collapses coverage, while author-count relation weighting through a reversed-PageRank rule resists the flood markedly better than uniform weights.

13.
arXiv (CS.CV) 2026-06-16

No One Knows the State of the Art in Geospatial Foundation Models

Geospatial foundation models (GFMs) have been proposed as generalizable backbones for disaster response, land-cover mapping, food-security monitoring, and other high-stakes Earth-observation tasks. Yet the published work about these models does not give reviewers or users enough information to tell which model fits a given task. We argue that nobody knows what the current state of the art is in geospatial foundation models. The methods may be useful, but the GFM literature does not standardize evaluations, training and testing protocols, released weights, or pretraining controls well enough for anyone to compare or rank them. In a 152-paper audit, we find 46 cross-paper disagreements of at least 10 points for the same model, benchmark, and protocol; 94/126 papers with extractable pretraining data use a configuration no other paper uses; and 39% of GFM papers release no model weights. This lack of community standards can be solved. We propose six concrete expectations: named-license weight release, shared core evaluations, copied-versus-rerun baseline annotations, variance reporting, one shared evaluation harness, and data-vs-architecture-vs-algorithm controls. These gaps are a coordination failure, not a fault of any individual lab; the authors of this paper, like many others in the GFM community, have contributed to them. Rather than just critiquing the community, we aim to provide concrete steps toward a shared understanding of how to innovate GFMs.

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

Understanding Scam Trends and Rail Paths from Reddit Self-Disclosure Narratives

Online scam behavior is inherently multi-stage, and the lifecycle includes temporally ordered rails and events rather than isolated signals. Existing works analyze characteristics of scam types and rails, but they do not track scam trends across years. Moreover, the work on the relations between rails is hampered due to the lack of open-source datasets with annotations and coverage of different scam types. To address these gaps, we build a dataset to analyze the yearly trend of scam characteristics and rail paths using Reddit self-disclosure narratives from 2023 to 2025. We collect 21,304 posts from scam-related subreddits with at least one rail among identity, communication, platform, and payment for trend analysis by heuristic annotation. Then, we label 1,800 posts containing explicit or recoverable scam chains by an LLM-assisted method for scam path analysis. The method is evaluated with human annotation. Lastly, we run a topic model on the comments of the posts to analyze the community support behavior. The results reveal that scam processes are predominantly multi-rail. Across years, different scam types and rail components dominate. Different scam types vary systematically in path complexity. Reddit support behaviors have become more detailed over time. This work supports synthetic scam chain data simulation and AI-related scam risk assessment, though findings may not generalise to other platforms.

15.
arXiv (CS.AI) 2026-06-11

StatefulDiscovery: Evidence-Calibrated Claim Formation in Open-Ended Scientific Discovery

arXiv:2606.11851v1 Announce Type: new Abstract: Open-ended scientific discovery asks agents to move beyond executing analyses for predefined questions. Across multiple rounds of exploration, a discovery agent must decide which phenomena warrant investigation while avoiding overinterpretation, where emerging claims exceed the evidential scope of the analyses supporting them. This creates an evidence-calibration problem: the exploration trajectory must be coupled with claim status so that evidence can guide both what to investigate next and what can be claimed. We introduce StatefulDiscovery, a discovery framework that externalizes investigation state and uses it to coordinate frontier selection, evidence acquisition, and claim adjudication. We evaluate StatefulDiscovery across 40 real-data discovery tasks. Compared with several baselines, StatefulDiscovery produces more claims overall judged to be both well-supported and high-value. Ablations indicate that structured hypotheses, local adjudication, and frontier control contribute to performance. Together, these results suggest that explicit discovery state can couple exploration with evidence-calibrated claim formation.

16.
arXiv (CS.AI) 2026-06-19

Human-AI Agent Interaction in a Business Context

arXiv:2606.18716v1 Announce Type: cross Abstract: As AI agents are increasingly integrated into core business processes, understanding and designing effective interaction patterns between humans and AI agents becomes crucial for value creation. This study identifies and evaluates principles and criteria for a positive User Experience (UX) with AI agents, along with methods for its measurement. We identify user expectations and needs to facilitate adoption, build trust, and support user-centered decision-making by development teams. Using a mixed-methods approach that combines qualitative and quantitative techniques, we explore interaction patterns between humans and AI agents. The findings from this exploratory research serve as the basis to develop a survey experiment which evaluates the effectiveness of specific design elements on a larger scale. This foundational research contributes to the development of more intuitive and effective human-AI agent interactions in business settings.

18.
arXiv (quant-ph) 2026-06-19

Many-Body Protection of Topological Edge Memory in Strong Interacting Quenches

arXiv:2606.19437v1 Announce Type: cross Abstract: Quantum quenches drive edge states far from equilibrium, yet whether the memory of a topological initial state survives in a non-integrable, interacting system has remained largely unexplored. We study this question in the bond-alternating XXZ chain – an interacting Su–Schrieffer–Heeger model hosting symmetry-protected topological edge modes with markedly enhanced boundary magnetization – and analyze quenches across all combinations of single-particle and many-body initial and final Hamiltonians. The results organize by a single distinction as we rigorously establish in this work: whether the post-quench Hamiltonian is free or genuinely interacting. For a free post-quench Hamiltonian, the dynamics is solved exactly by a correlation-matrix approach; the boundary-mode return amplitude decays as $t^{-3/2}$, and initial interactions enter only through a dressed one-body density matrix. For a genuinely interacting post-quench Hamiltonian, finite-time stability bounds prove that away from local resonances the first-dimer magnetization remains stable on time windows growing as arbitrarily large powers of the inverse inter-dimer coupling. Matrix product state simulations across all four protocols show that interactions in the final Hamiltonian markedly extend finite-time boundary memory – with local suppression near the isotropic $SU(2)$ point – revealing a many-body protection mechanism in a non-integrable system where scrambling would otherwise wash out initial-state memory fast.

19.
arXiv (math.PR) 2026-06-16

Exact Label Recovery in Euclidean Random Graphs

arXiv:2407.11163v3 Announce Type: replace-cross Abstract: In this paper, we propose a family of label recovery problems on weighted Euclidean random graphs. The vertices of a graph are embedded in $\mathbb{R}^d$ according to a Poisson point process, and are assigned to a discrete community label. Our goal is to infer the vertex labels, given edge weights whose distributions depend on the vertex labels as well as their geometric positions. Our general model provides a geometric extension of popular graph and matrix problems, including submatrix localization and $\mathbb{Z}_2$-synchronization, and includes the Geometric Stochastic Block Model (proposed by Sankararaman and Baccelli) as a special case. We study the fundamental limits of exact recovery of the vertex labels. Under a mild distinctness of distributions assumption, we determine the information-theoretic threshold for exact label recovery, in terms of a Chernoff-Hellinger divergence criterion. Impossibility of recovery below the threshold is proven by a unified analysis using a Cramér lower bound. Achievability above the threshold is proven via an efficient two-phase algorithm, where the first phase computes an almost-exact labeling through a local propagation scheme, while the second phase refines the labels. The information-theoretic threshold is dictated by the performance of the so-called genie estimator, which decodes the label of a single vertex given all the other labels. This shows that our proposed models exhibit the local-to-global amplification phenomenon.

20.
arXiv (CS.CV) 2026-06-18

Physics in 2-Steps: Locking Motion Priors Before Visual Refinement Erases Them

Image-to-Video diffusion models leverage input images to generate visually stunning content, yet frequently produce motion that violates physical laws. We reveal a surprising finding: a 2-step generation often exhibits better physical consistency than a 50-step output from the same model. Through spectral analysis, we trace this to phase erosion during denoising; the phase degrades significantly (dropping by $\approx 18\%$ from step 2 to step 50), whereas the magnitude remains relatively stable. Building on this insight, we propose PhaseLock, a training-free framework that preserves the valid motion priors from few-step inference throughout the denoising trajectory. Rather than relying on full-step inference for physical consistency, PhaseLock extracts a motion prior from just 2 steps and enforces it onto high-fidelity generation via Latent Delta Guidance. Our approach effectively mitigates phase degradation, improving physical consistency by an average of 6.2 points across diverse models while largely maintaining visual fidelity, with negligible overhead ($1.06\times$ time, $1.02\times$ memory) and reduced reliance on expensive external guidance methods ($\sim5\times$ time). Project Page: https://dnwjddl.github.io/phaselock

21.
arXiv (CS.CL) 2026-06-17

Self-Generated Error Training for Token Editing in Diffusion Language Models

作者:

Token-to-token (T2T) editing lets LLaDA2.1 revise committed tokens during block-diffusion decoding. The released recipe trains this editor on random vocabulary corruptions, but at inference the editor sees the model's own fluent, high-confidence draft errors instead. We study this training-inference mismatch and propose self-generated T2T, which performs a no-gradient draft pass, fills masked positions with predicted tokens, and supervises recovery in a second pass under these self-generated corruptions. We implement the update as a short LoRA continued-pretraining pass on LLaDA2.1-mini and evaluate on several benchmarks under the official Q-Mode T2T procedure with unchanged inference parameters. The method generally improves accuracy while reducing T2T edit intensity, mitigating failure modes such as final-digit transcription errors after otherwise correct reasoning and excessive self-correction before short factual answers.

22.
arXiv (quant-ph) 2026-06-19

Scalable quantum circuit knitting using a weak-coupling approximation

arXiv:2606.19035v2 Announce Type: replace Abstract: We present a method for performing distributed quantum computing with controlled approximations. Exact distributed quantum computing requires exponential classical information to reconstruct the quantum process. However, we show how the classical cost is reduced to polynomial if the quantum procedure can be partitioned between a qubit that is weakly coupled the other qubits. We demonstrate our method for a layered circuit based on the circuits used for the quantum approximate optimization algorithm.

23.
arXiv (CS.LG) 2026-06-15

Compressed Computation is (probably) not Computation in Superposition

arXiv:2606.14673v1 Announce Type: new Abstract: We study whether the Compressed Computation (CC) toy model (Braun et al., 2025) is an instance of computation in superposition. The CC model appears to compute 100 ReLU functions with just 50 neurons, achieving a better loss than expected from only representing 50 ReLU functions. We show that the model mixes inputs via its noisy residual stream, corresponding to an unintended mixing matrix in the labels. Splitting the training objective into the ReLU term and the mixing term, we find that performance gains scale with the magnitude of the mixing matrix and vanish when the matrix is removed. The learned neuron directions concentrate in the subspace associated with the top 50 eigenvalues of the mixing matrix, suggesting that the mixing term governs the solution. Finally, a semi-non-negative matrix factorization (SNMF) baseline derived solely from the mixing matrix reproduces the qualitative loss profile and improves on prior baselines, though it does not match the trained model. These results suggest CC is not a suitable toy model of computation in superposition.

24.
arXiv (CS.CV) 2026-06-17

Pulling The REINS: Training-Free Safety Alignment of Video Diffusion Models via Representation Steering

Open-weight video diffusion models can generate photorealistic unsafe content, from violence to misinformation, yet existing defenses either require expensive safety fine-tuning that degrades general capability, or apply external filters that are trivially bypassed by adversarial prompts. We present REINS (REpresentation-space INference-time Safety steering), a training-free method that aligns video diffusion models at inference time by steering their internal representations toward safe generation. Our key finding is that safety-relevant structure is linearly encoded in the hidden-state activations of video diffusion transformers, and a single direction, discovered via Supervised PCA on binary safety labels, suffices to separate safe from unsafe generation trajectories. At inference, adding this direction to hidden states at an intermediate transformer layer redirects generation from harmful content to semantically related safe alternatives, with no weight updates, no concept enumeration, and negligible computational overhead. Through mechanistic analysis, we reveal that while safety information accumulates monotonically with transformer depth, steering effectiveness peaks at intermediate layers (~50% depth), exposing a fundamental tradeoff between information availability and downstream propagation capacity. We evaluate REINS across 9 video diffusion models, multiple parameter scales (1.3B-5B), and both text-to-video and image-to-video generation, to our knowledge, the broadest safety evaluation suite in the video generation literature.

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

Detecting Hallucinations for Large Language Model-based Knowledge Graph Reasoning

Knowledge graph (KG) reasoning infers new knowledge from existing facts and is widely applied in question answering, recommendation, and decision support. With the rapid development of large language models (LLMs), LLM-based KG reasoning frameworks have become increasingly popular by leveraging retrieved KG information. However, hallucinations in LLMs remain a critical issue. Even when relevant KG knowledge is incorporated, models may still generate incorrect outputs, leading to misinformation and unreliable decisions. Existing hallucination detection methods either focus on LLM internal states or verify consistency with retrieved contexts, but both overlook the structural information in KGs, resulting in suboptimal performance. To address this gap, we propose LUCID, the first halLUcination deteCtIon method for LLM-based knowleDge graph reasoning frameworks. LUCID jointly leverages LLM attention scores, KG semantics, and structural information. Specifically, it extracts node and edge features from attention scores and semantic similarities, and integrates them with KG structure using a graph neural network. We also construct manually annotated benchmark datasets for evaluation. Experiments on nine datasets show that LUCID achieves state of the art performance compared to 15 baselines.