A functional commitment (FC) commits to a value x, and can later generate a proof for a function value y = f (x) with respect to some function f∈F . In contrast, the dual functional commitment (dual FC) allows a committer to commit to a function f∈F , and later produces an opening proof π for the function value y = f (x) given an input x. We propose a new construction of dual FC scheme from lattices. Our dual FC scheme can support arbitrary circuits of bounded sizes. Moreover, our dual FC scheme enjoys computational binding, and we prove the computational binding property of our dual FC based on the l-succinct H - SIS assumption, a falsifiable generalization of the l-succinct SIS assumption, in the random oracle model. In addition, our dual FC scheme is quasi-succinct with a succinct commitment and a quasi-succinct opening proof.
Quantum simulation is a rapidly advancing field poised to revolutionize our understanding of complex quantum systems by harnessing the unique capabilities of quantum computers. In this review, we present a concise overview of key developments in quantum simulation algorithms, with a focus on time-independent digital Hamiltonian simulation of quantum dynamics. We review seminal methods—from traditional Trotter formulas to techniques such as truncated Taylor series combined with the linear combination of unitaries, and quantum signal processing. Additionally, we examine error analyses and recent innovations that enhance the efficiency and accuracy of quantum simulations.
Negative focuses are the most prominent negated texts for a negative cue or verbal negation in a negative statement. Although previous work adopted sequence labelling framework using LSTM and CRF networks, Negative Focus Detection (NFD) is still faced with several disadvantages involving with data limitation, coarse-grained encoding, and insufficient dependencies of the sequence of words. To solve these problems, we firstly apply data augmentation driven by Large Language Models (LLMs) to produce more samples. Then, we propose a novel HyperGraph attention LSTM network (LSTM-HyG) to capture high-level semantics for sentences, and negated verbs, negative cues. Finally, we predict negative focuses by a fine-grained label scheme that can learn adequate sequential dependency relationship of words. Experimental results on PB-FOC and CNeSp datasets can prove that our proposed model is superior to state-of-the-arts.
Recent advancements in native 3D generation have demonstrated remarkable capabilities in producing high-quality 3D assets from image or text prompts. However, these methods face a critical challenge: insufficient alignment between generated meshes and input conditions. In this paper, we propose Multimodal Triplane Diffusion Transformer to address the issue, featuring two core components: A triplane-based 3D variational autoencoder that compresses point clouds, sampled uniformly from mesh surfaces and concentrated near sharp edges, into a triplane latent space, and a diffusion model trained on this latent space, empowered by multimodal diffusion transformer blocks to establish cross-modality interactions between latent representations and conditions. Extensive experiments demonstrate that our method achieves not only superior generalization capability but also significantly enhanced geometric alignment with input images compared to state-of-the-art approaches in the image-to-3D task.
Generative model (GM)-assisted product design has become increasingly popular. The key is to find generated samples (GSs) satisfying the design goal from the GM’s latent space (GLS). Most existing works rely on defining objective functions or writing prompts to search for GSs in the GLS. Unlike them, we propose a progressive approach that relies on multiple rounds of neighborhood exploration to choose desirable GSs from the GLS. Notably, the approach allows users to concretize and refine their goals during the exploration, thus applying to abstract or unspecific goals, which is unavailable for all existing techniques. The approach integrates two techniques to solve the challenges of achieving and applying it. First, many GSs are highly similar or irrelevant to the neighborhood center. Those GSs do not allow users to make comprehensive comparisons for rational choices and cannot be excluded by classic methods. Thus, we propose a method to avoid collecting them, which makes collected GSs have representative feature variations from the neighborhood center. Second, we need a system for applying the approach. The system should fulfill many visualization requirements to efficiently drive exploration and keep it always in the right direction. Thus, we followed the mountain-climbing metaphor to design the system and developed a series of visual and quantitative techniques to achieve these requirements. Cases on multiple real-world datasets and GMs, results of quantitative experiments, and performance and feedback of participants in user studies prove the approach’s effectiveness and usability.
Multimodal Large Language Models (MLLMs) have made remarkable progress in video understanding and consistently perform well on vision-centric benchmarks. However, existing benchmarks primarily evaluate factual or event-based comprehension, while neglecting audience insights. It is a critical yet underexplored dimension of video understanding, reflecting a deep comprehension of cognitive processes from the audience’s perspective. As a result, MLLMs, shaped by such benchmarks, often produce responses that are factually correct but misaligned with audience’s interests. To bridge this gap, we leverage audience insights derived from video comments as a direct proxy to guide the annotation process and introduce A3Bench, an audience-aligned benchmark for evaluating video audience insights with large-scale videos and high-quality multilingual comments. Furthermore, inspired by neuro-imaging studies, we propose Cognition Interaction of Thought (CIoT), a structured reasoning framework that emulates key aspects of cognitive processes. Extensive experiments on A3Bench reveal that current MLLMs struggle to understand audience insights, particularly compared to human-level understanding. In contrast, CIoT can improve the performance of these models, highlighting its potential to enhance the MLLMs’ capability of understanding audience insights in future research.
Temporal Knowledge Graph (TKG) reasoning plays a pivotal role in predicting emerging facts based on historical data. However, existing TKG reasoning methods typically aggregate all historical facts within a given time window indiscriminately, which often introduces outdated or irrelevant information. This information redundancy can significantly hinder the reasoning performance, especially as TKGs continue to grow in scale and complexity. Effectively filtering out irrelevant facts is thus essential for improving inference accuracy and efficiency. To address this critical challenge, we focus on how to refine the TKGs and propose a Temporal knowledge graph reasoning model via Multi-granularity Knowledge Refinement (T-MKR). Specifically, we propose a multi-granularity knowledge refinement approach to prune historical TKGs, which selectively removes irrelevant edges and unnecessary nodes at both the edge and node levels. The resulting refined subgraphs are used for representation learning. To effectively combine information from both refinements, we introduce a subgraph gating integration module. Additionally, we leverage contrastive learning for subgraph alignment to emphasize the relationships between the two refined subgraphs. Extensive experiments on six commonly used datasets demonstrate the superiority of T-MKR compared with many state-of-the-art baselines.
Existing methods for Large Language Models (LLMs) personalization typically rely on extensive pre-collected user data. However, practical personalization for LLMs-based chatbots often starts from a “cold-start” scenario with limited interaction history, where forming an accurate initial impression is crucial for user acquisition and retention. This critical challenge of “cold-start” personalization is further compounded by the absence of dedicated benchmarks for its evaluation. To address this gap, we introduce ColdChat, the first benchmark designed to assess LLM personalization using brief interaction histories. The collection process for ColdChat involved tasking human annotators with (1) engage in multi-session open-domain dialogues with LLM, (2) annotate personalized user profiles based on their dialogue history, and (3) label a user-specific test set for the evaluation of personalization. Our experiment on ColdChat reveals that state-of-the-art LLMs struggle to be personalized in this “cold-start” setup. To enhance LLM personalization in this scenario, we propose EPIC, a novel framework for Extracting user Profile from Interaction Context. Experimental results on ColdChat reveal that the profiles extracted by EPIC yield substantial improvements for personalization, yielding a +27.2
Network motifs are fundamental tools for analyzing complex systems, which provide deep insights into the functional abilities of networks. Temporal networks have attracted growing attention for modeling the dynamics of real-world systems, and there is a growing need to properly reinterpret network motifs for various practical applications. In this article, we provide a comprehensive review of the studies on temporal network motifs. First, we systematically introduce and analyze the various concepts of temporal network motifs and their corresponding discovery algorithms. Second, we review existing applications of temporal network motifs. Finally, we list the challenges and opportunities in temporal network motif research. We hope this article provides valuable insights for researchers interested in temporal network analysis.
Radiance fields, such as NeRFs, 3D Gaussians, and their variants, have emerged as the leading representations for 3D scene reconstruction due to their exceptional performance in novel view synthesis. However, their effectiveness depends on input images captured in well-lit, static environments, making dark scenes a significantly challenging case. Prior works employ low-light enhancement for low-light scenes (e.g., candlelight), but completely dark scenes remain an unsolved problem. However, this is a very common case when exploring unknown scenes, such as caves or nighttime forests, or derelict buildings. To solve the problem, we propose capturing images with a camera-mounted flashlight for exploring such scenes, which is an easily accessible setting for robots. The flashlight’s parameters are modeled and optimized in the reconstruction pipeline, including the flashlight’s angular and distance attenuation, position, rotation, and intensities. Under this setting, the captured images are under dynamic lighting conditions, i.e., lighting is changing for each image. We formulate a photometric stereo (PS) problem of input images by a grouping-and-merging strategy, leveraging its results as supervision priors. As a result, the method enables reconstruction and relighting of dark scenes. Experiments show that the method outperforms state-of-the-art approaches in decomposition, geometry, and relighting.
In class-imbalanced semi-supervised learning, the goal is to leverage abundant unlabeled data when labeled examples are scarce in a class-unbalanced setting. Classifiers of pseudo-label-based algorithms tend to become biased and suffer from degraded representation quality due to the utilization of skewed pseudo-labels for training. Previous pseudo-label-based algorithms employ the classifier itself to generate pseudo-labels for unlabeled data, leading to suboptimal performance on unbalanced tasks. The classifier is optimized to achieve uniform accuracy across all classes, mitigating the bias toward majority classes, while pseudo-labeling strives to accurately annotate the training unlabeled data in a class-imbalanced distribution. This misalignment causes confirmation bias, reinforcing bias in the pseudo-labeling process. To address this issue, we propose a novel semi-supervised framework that disentangles pseudo-label generation from the classification task via designing a dedicated pseudo-label generator to align the class distributions between labeled and unlabeled data. Specifically, we alternately train the pseudo-label generator and the predictive model, where the pseudo-label generator is trained on a debiased label enhancement objective, and the predictive model then leverages these pseudo-labels along with class-level debiasing. Experiments on the imbalanced benchmark datasets validate the effectiveness of the proposed framework.
Imbalanced kernel clustering, distinguished by differing sample counts among diverse clusters, has gained significant prominence in a multitude of real-world nonlinear data mining scenarios. Nevertheless, the computational requirements of such approaches are often associated with the kernel matrix and display a quadratic increase in relation to the data volume, making it unfeasible for scenarios involving large-scale imbalanced datasets. Moreover, despite the importance of theoretical analysis in machine learning, fast imbalanced kernel clustering methods still lack solid statistical guarantees. Understanding the statistical properties of fast imbalanced kernel clustering therefore remains an important and underexplored problem. To solve these problems, we propose a framework of fast Imbalanced Kernel k-Means (IKKM), exploring both computational demands and statistical analysis. According to the theoretical analysis, the proposed fast IKKM can take less time to attain a similar accuracy of exact IKKM, when operating with a sketching dimension of approximately Ω(√(n)) with n denoting the sample count. In particular, we establish the first optimal excess clustering risk bound for the fast IKKM under mild conditions. Comprehensive experiments validate the theoretical analysis of the fast IKKM in addressing the computational challenges of large-scale imbalanced clustering.
Introduction Artificial intelligence (AI) holds promising potential for generating educational resources such as personalized assessments. Despite this appeal, the impact of AI-assisted multiple-choice question (MCQ) authoring on item quality and on teachers' editing behavior remains insufficiently studied.Methods This study (N = 152 items from 19 teachers) compares MCQ quality across three conditions-teacher-only (T), AI-only (AI), and teacher-AI collaboration (T-AI)-using a 19-criterion item-writing-flaw (IWF) rubric and logged authoring-interaction density.Results Neither human raters nor AI models reliably differentiated item provenance, indicating that LLM-generated MCQs have reached a level of surface quality largely indistinguishable from human-authored material. Yet both AI and T-AI items carried significantly more flaws than T items, with the T-AI condition showing a marked increase (d = 1.06) consistent with an automation-bias pattern: teachers accepted AI drafts with minimal critical engagement, as evidenced by significantly lower interaction density in the AI-supported condition. Criterion-level analysis revealed a complementary pattern: AI items exhibited specific cueing biases-notably lexical overlap between stem and correct answer-but simultaneously avoided structural format flaws (e.g., true/false questions) that teachers commonly produced.Discussion These results underscore that the human-AI interaction workflow, rather than AI capability per se, is the critical quality determinant, and that AI-assisted assessment creation requires structured quality review processes that leverage AI's structural consistency while correcting its cueing biases.
This article proposes Disruptive Neuroepistemological Pedagogy (DNEP) as a conceptual and design-oriented framework for STEM education. The central claim is that some STEM concepts should not be taught only as stable curricular content, but as products of scientific disruption: they emerged because previous explanations became insufficient, because anomalies required new models, or because their use transformed society. DNEP integrates three dimensions: epistemological disruption, neuroeducational meaningfulness, and ethical-social responsibility. Its novelty does not lie in adding neuroscience, epistemology, or critical pedagogy as independent domains, but in coordinating them around disruptive STEM concepts that require conceptual revision, model-based reasoning, and responsible use. The article follows a conceptual and integrative review methodology. It defines the research gap, presents the core principles of DNEP, specifies design criteria, proposes a partial instructional model, and identifies boundary conditions for implementation. The framework is positioned against behaviorism, cognitivism, constructivism, holistic-humanistic education, and STE(A)M through an analytical comparison focused on learning, disruption, emotion, curriculum, teacher mediation, inquiry, and assessment. The article concludes that DNEP can contribute to STEM education by helping students understand why scientific ideas change, how such change can be learned meaningfully, and why scientific knowledge requires ethical and social responsibility.
IntroductionThis paper proposes an integrated model of student digital profiling based on multisource educational data. The aim of the model is to improve the accuracy of predicting academic performance and learning risks. Unlike traditional approaches that rely on a limited set of academic indicators, the proposed model integrates academic, research, social, and behavioral characteristics to form a holistic representation of learning activity.MethodsThe methodological framework includes semantic clustering of disciplines using NLP methods, calculation of an integrated grade point average (IGPA), and the application of multimodal machine learning models. SHAP analysis was used to interpret the results. Experimental validation was conducted on real data from a university information system using the AUC-PR, LogLoss, KS, and PSI metrics.ResultsThe results demonstrate that the use of multimodal data increases the accuracy of learning risk prediction by 8–12% compared with baseline models. Dynamic behavioral features enable the early identification of academic underperformance risks at initial stages of study. The model demonstrated robustness when applied across different faculties and scalability under conditions of increasing data volume.DiscussionPractical implementation in the form of interactive dashboards confirmed the applicability of the proposed approach for supporting pedagogical and managerial decision-making. The model provides a foundation for personalized academic support and the development of digital educational analytics.
IntroductionThis study examines a Pepper-based coding activity in primary education through three complementary dimensions: changes in coding performance, perceived cognitive workload, and pupils' post-activity engagement responses. Rather than testing the causal effectiveness of Pepper in isolation, the study investigates how a robot-mediated coding activity can be implemented and evaluated in an authentic classroom context.MethodsTwenty-nine third-grade pupils (8–9 years old) from an Italian primary school participated in a teacher-orchestrated coding activity in which they programmed the Pepper robot to navigate a floor grid. Coding performance was assessed using curriculum-aligned pre- and post-tests scored on a 0–10 scale. The perceived workload was measured with a child-adapted NASA Task Load Index (NASA-TLX) questionnaire. Post-activity engagement responses were collected using an adapted version of the User Engagement Scale-Short Form (UES-SF), with particular caution in interpretation due to the limited internal consistency observed for several subscales.ResultsThe coding scores increased significantly from the pre-test (M = 5.83, SD = 3.09) to the post-test (M = 7.97, SD = 2.54), t(28) = 3.75, p < 0.001, with a size of the effect size of medium-to-large paired-samples (dz = 0.70). NASA-TLX responses indicated low mental and temporal demand, moderate physical demand, high perceived performance, and low frustration, suggesting that the activity was perceived as manageable by the pupils. The adapted engagement questionnaire provided limited descriptive evidence of positive post-activity responses, especially on the Reward subscale, while the other UES-SF dimensions were interpreted cautiously due to weak reliability.DiscussionThe findings provide preliminary classroom-based evidence that a Pepper-mediated coding activity was associated with increased coding scores, manageable perceived workload, and positive descriptive indications of perceived reward. However, because of the single-group pre–post design, small sample size, unequal pre- and post-test formats, short-term scope, and limitations of the adapted engagement measure, the results should not be interpreted as causal or generalizable evidence of Pepper's effectiveness. Instead, the study offers methodological and design insights for future comparative research on robot-mediated coding activities in primary education.
The increasing number of cyber threats demands a robust, real-time detection system that can accurately classify attacks while maintaining computational efficiency in real-time and within reasonable resource limits. Most real-time applications in cybersecurity still rely on traditional machine learning methods with arbitrary configurations due to the difficulty in resolving the trade-off between accuracy and speed within the system. This work proposes a modification to the standard Gaussian Naive Bayes (GNB) classifier, utilizing the Weighted Classification Strategy (WCS-GNB), to enhance the real-time detection of cyberattacks evaluated under simulated streaming conditions on commodity CPU hardware. It aims to address the limitations of traditional probabilistic classifiers as applied in cybersecurity. The WCS-GNB model seeks to preserve the detection accuracy of the model while incorporating class-dependent scaling and traditional Bayesian approaches through a formally derived log-posterior extension of the standard GNB framework. The methodology is evaluated on NSL-KDD and CICIDS2017, which consists of class-specific variance scaling and symmetric Bayesian inference on streaming network data with adaptive feature weighting systems. The proposed WCS-GNB model achieved a detection accuracy of 94.3% with a processing time of 2.9 ms, significantly outperforming the traditional GNB (85.2% accuracy) and competing with complex methods like Random Forest (91.7%), while maintaining a superior processing speed. The WCS-GNB model demonstrated robust performance across various attack types, including DDoS (96.2%), DoS (97.3%), Port Scanning (92.8%), Web Attacks (94.1%), and Botnet activities (89.5%). Throughput reaches ≈8.5 k records/s on commodity hardware. All performance improvements are confirmed statistically significant via paired t-tests (p < 0.05, Bonferroni-corrected) with large effect sizes (Cohen’s d ≥ 0.52). These results indicate WCS-GNB offers a practical, interpretable, and deployment-ready IDS core for high-throughput environments. The WCS-GNB approach successfully connects the gap between accuracy and efficiency in real-time cybersecurity apps. The integration of weighted features and class-specific scaling provides a practical solution for high-throughput network monitoring, while also maintaining the interpretability advantages of Bayesian methods.
Deep learning is increasingly applied in detecting DDoS attacks while potentially bringing new security risks. In this paper, we propose a novel adversarial approach against deep learning based DDoS detection systems, and also explore its defense methods. The purpose of this adversarial approach is to reduce the detection accuracy of deep learning based detection systems by deceiving their built-in deep learning based detection models with adversarial theories. It first stealthily collects relevant network data from its directly-connected network device of a target detection system, and thus it could obtain DDoS flow samples and normal flow samples by deliberately launching DDoS or not, respectively. Critical features to detect DDoS could be selected after observing the reactions from the target detection system. Then, a local estimation model is established to approximate the real built-in detection model. Owing to the established estimation model, adversarial samples against the real detection model could be generated by an adversarial sample generation method based on local saliency function. At last, according to the generated adversarial samples, each adversarial DDoS attack flow is forge by a traffic generator and directed to the target system. To prevent this attack approach, we also explore its defense method. Further, we conduct and evaluate the proposed adversarial approach and its defense method based on a real-world network topology and dataset. The experimental results indicate that this approach is capable of degrading the detection accuracy significantly and the defense method is effective by using detection accuracy.
Energy depletion in battery powered Wireless Sensor Networks (WSNs) causes coverage gaps and can eventually lead to the complete failure of the network. Mobile charging robots (MCRs) can be used to replenish energy and maintain the continuity of the network. Most existing work in this area assumes that MCRs are reliable in their operations; however, in real-world scenarios, robots may malfunction or deliver incomplete charges. We propose a trust-based MCR selection mechanism that allows sensors to evaluate and select charging robots on the basis of their performance in past interactions. Trust computation is done locally at each sensor with minimal overhead and does not require centralized coordination. Simulation results show that trust-based selection can outperform random selection in terms of network coverage across different network sizes, battery capacities, and robot populations without increasing disconnection time by a large margin.
IntroductionAlthough AI virtual news anchors have gained attention for their accurate and uninterrupted broadcasting, existing research has mainly focused on technical efficiency, leaving insufficient understanding of normative deviations in their speech and the impact on audience experience.MethodsThis study conducted in depth interviews with 11 Chinese news consumers and 2 technology practitioners from the Chinese state-media sector to examine audience perceptions of Chinese AI virtual anchor performance within the state-media context.ResultsResults showed that participants widely perceived deficiencies in sentence stress, intonation, and rhythm. These perceived normative deviations were reported by participants as reducing clarity, emotional resonance, and audience engagement. Beyond technical dissatisfaction, 9 out of 11 participants expressed value rational concerns, focusing on lack of human connection, aesthetic quality, and weakened social ritual functions of news broadcasting.DiscussionBased on these findings, the study proposes a working model in which perceived linguistic deviations are linked to experienced communication failure and an inferred sense of value imbalance. The findings suggest that sustainable development of virtual anchor technology may benefit from recalibrating the relationship between instrumental and value rationality, treating technical efficiency as a means rather than an end. These findings offer empirically grounded directions for human machine communication research and for exploring how humanistic values might be integrated into AI mediated news practice.
