With the growing adoption of Machine-Learning-As-A-Service (MLaaS), Private Inference (PI) has emerged as a promising solution to address its security concerns through cryptographic techniques. However, nonlinear operations in neural networks account for most of the computational and communication overhead in PI. Existing studies mainly focus on optimizing and reducing the number of ReLU activations in neural networks, but traditional pruning methods may mistakenly remove ReLUs that are critical to maintaining model accuracy. To accurately evaluate the importance of ReLUs in the network, we propose ReLUPruner, a method that uses Taylor expansion to quantify the impact on loss before and after ReLU replacement. Furthermore, we establish a hierarchical importance metric to guide layer-wise ReLU budget allocation and adopt a progressive pruning strategy that dynamically adjust the pruning rate of each layer according to training progress. Extensive experiments on various models and datasets show that ReLUPruner achieves a good balance between ReLU budget and model accuracy, yielding improvements of 1.89% (12.9k ReLUs, CIFAR-10), 3.62% (50k ReLUs, CIFAR-100) and 2.66% (30k ReLUs, Tiny-ImageNet) over the previous state-of-the-art.

The identification and filtration of malicious texts in social media environments represent a significant technical challenge aimed at protecting users from online violence and disinformation. This complexity stems from the diversity and innovativeness of social media texts, which include unique expressions and special sentence structures. Particularly, malicious texts in interrogative forms pose alignment challenges with traditional corpora due to existing methods’ failure to exploit the text’s deep global semantic representations. This issue is compounded by the scant research on Chinese texts, leading to inefficiencies in recognition accuracy. To mitigate these challenges, we introduce an innovative framework based on a Global Contrastive Semantic Network (GCSN), designed to enhance malicious text recognition efficiency and accuracy by deeply learning global semantic knowledge. It comprises an encoder for global semantic information modelling and a graph-matching network for semantic similarity evaluation between question pairs, enabling the accurate identification and filtering of malicious texts with complex structures. Furthermore, we introduce a semantic consistency-based data augmentation method (COMBINE), using real-world data to generate balanced positive and negative samples, enriching the dataset and enhancing the model’s ability to distinguish semantic consistency through contrastive learning. Experimental validation on two Chinese datasets demonstrates our model’s exceptional performance, affirming its applicationa value in social media malicious text recognition. Our code is available at https://github.com/Wxy13131313131/GCSN-COMBINE

Temporal reasoning is crucial for large language models (LLMs) to understand event concurrency and complex temporal interactions in natural language.Recent approaches rely on the LLM to infer temporal relations between events and largely overlook the inherent structural nature of temporal relationships.In this work, we propose ODL-TempLLM (Ontology-Guided and Description Logic-Constrained Temporal Reasoning with LLMs), a novel paradigm for temporal reasoning with LLMs that shifts focus from internal inference to the explicit modeling of temporal structure.ODL-TempLLM leverages ontology learning to explicitly construct structured temporal knowledge, employs a symbolic reasoner to deductively reason about temporal relations and uses logic-constrained retrieval augmentation to obtain relevant facts.Experiments results evaluated across three datasets via various LLM backbones show that our method outperforms state-of-the-art methods by 2.07-31.83F1 points and 1.00-30.73EM points, exhibiting strong generalization and highlighting the potential of explicit temporal reasoning.

The identification and filtration of malicious texts in social media environments represent a significant technical challenge aimed at protecting users from online violence and disinformation. This complexity stems from the diversity and innovativeness of social media texts, which include unique expressions and special sentence structures. Particularly, malicious texts in interrogative forms pose alignment challenges with traditional corpora due to existing methods' failure to exploit the text's deep global semantic representations. This issue is compounded by the scant research on Chinese texts, leading to inefficiencies in recognition accuracy. To mitigate these challenges, we introduce an innovative framework based on a Global Contrastive Semantic Network (GCSN), designed to enhance malicious text recognition efficiency and accuracy by deeply learning global semantic knowledge. It comprises an encoder for global semantic information modelling and a graph-matching network for semantic similarity evaluation between question pairs, enabling the accurate identification and filtering of malicious texts with complex structures. Furthermore, we introduce a semantic consistency-based data augmentation method (COMBINE), using real-world data to generate balanced positive and negative samples, enriching the dataset and enhancing the model's ability to distinguish semantic consistency through contrastive learning. Experimental validation on two Chinese datasets demonstrates our model's exceptional performance, affirming its applicationa value in social media malicious text recognition. Our code is available at https://github.com/Wxy13131313131/GCSN-COMBINE

Logical reasoning of text requires neural models to possess strong contextual comprehension and logical reasoning ability to draw conclusions from limited information. To improve the logical reasoning capabilities of pretrained language models (PLMs), existing approaches can be broadly categorized into neural architecture-based methods and large language model (LLM)-driven strategies. While neural methods struggle with fine-grained logic that fails to capture detailed semantic roles and constraints, LLM-driven approaches, despite generating multistep reasoning sequences, lack explicit inference control and suffer from error accumulation due to their implicit and stochastic nature. Some works have tried using logical expressions, like first-order logic, but these approaches often fail to handle quantifiers systematically or support clear reasoning processes. Inspired by first-order logic and generalized quantifier (GQ) theory, we propose AQE-RF, a model based on an adaptive quantifier extension and rule-filtering graph network to address this challenge. The first component constructs a fine-grained text logical graph (FTLG) and then performs GQ instantiation based on option attention. The second component performs rule-filtered deductive reasoning, using conflict scores and dynamic programming (DP) to select coherent, interpretable inference paths. Extensive experiments on the LogiQA, ReClor, and AR-LSAT datasets demonstrate the effectiveness and robustness of AQE-RF.

To improve the performance of social recommendation, it can be beneficial to include social relationships, model user interests, and item attraction. Traditional methods use binary social relationships between users and items, known as ’user-user’ and ’user-item’, which rely on historical interactions like purchasing history and clickiing history equally. However, the interests and attractions aggregated by such approaches are incomplete. In addition to binary interactions, there are also a large number of ’user-item-user’ triple social behaviors, such as retweeting or recommending. Considering these triple social behaviors introduces inhomogeneity to the modeling of interests and attractions. Furthermore, beyond the primary explicit in-terest derived from interaction history, users are also influenced by secondary, latent interests in items shared by friends, which makes the modeling more complex. To address this issue, we propose IIMHGCN (Inhomogeneous Interest Modeling via Hy-pergraph Convolutional Networks). In addition to binary social relationships, we incorporate a hypergraph model that utilizes triple social recommendation behaviors and higher-order social connections to model inhomogeneous interests and attraction. We substantiate its effectiveness through extensive experiments on two real-world datasets.

Abstract In today's digital landscape, cybercriminals are constantly evolving their tactics, making it challenging for traditional cybersecurity methods to keep up. To address this issue, this study explores the potential of knowledge graph reasoning as a more adaptable and sophisticated approach to identify and counter network attacks. By leveraging graph structures imbued with human‐like thinking, this method enhances the resilience of cybersecurity systems. The study focuses on three critical aspects: data preparation, semantic foundations, and knowledge graph inference techniques. Through an in‐depth analysis of these components, the research aims to reveal how knowledge graph reasoning can improve cyberattack detection and enhance the overall efficacy of cybersecurity measures, including intrusion detection systems. The proposed approach has undergone extensive experimentation to validate its effectiveness compared to existing methods. The results of the experiment have shown a remarkable advancement in accuracy, speed, and recall for recognition, surpassing current methods. This achievement is a notable contribution in the realm of managing big data in cybersecurity. The study establishes a foundation for the automation of network attack detection, ultimately enhancing overall network security.

Cross-scene hyperspectral image (HSI) classification remains a challenging task due to the distribution discrepancies that arise from variations in imaging sensors, geographic regions, atmospheric conditions, and other factors between the source and target domains. Recent research indicates that convolutional neural networks (CNNs) exhibit a significant tendency to prioritize image styles, which are highly sensitive to domain variations, over the actual content of the images. However, few existing domain adaptation (DA) methods for cross-scene HSI classification take into consideration the style variations both within the samples of an HSI and between the cross-scene source and target domains. Accordingly, we propose a novel content-biased and style-assisted transfer network (CSTnet) for unsupervised DA (UDA) in cross-scene HSI classification. The CSTnet introduces a content and style reorganization (CSR) module that disentangles content features from style features via instance normalization (IN), while refining useful style information as a complementary component to enhance discriminability. A contentwise reorganization loss is designed to reduce the disparity between the separated content/style representations and the output features, thereby enhancing content-level alignment across different domains. Furthermore, we incorporate batch nuclear-norm maximization (BNM) as an effective class-balancing technique that directly exploits unlabeled target data to enhance minority class representations without requiring prior knowledge or pseudolabels, achieving better distribution alignment. Comprehensive experiments on three cross-scene HSI datasets demonstrate that the proposed CSTnet achieves state-of-the-art performance, effectively leveraging content bias and style assistance for robust DA in cross-scene HSI classification tasks. The code is available at:https://github.com/nbdszw/CSTnet.

Misinformation can seriously impact society, affecting anything from public opinion to institutional confidence and the political horizon of a state. Fake News (FN) proliferation on online websites and Online Social Networks (OSNs) has increased profusely. Various fact-checking websites include news in English and barely provide information about FN in regional languages. Thus the Urdu FN purveyors cannot be discerned using fact-checking portals. State-of-the-art (SOTA) approaches for Fake News Detection (FND) count upon appropriately labelled and large datasets. FND in regional and resource-constrained languages lags due to the lack of limited-sized datasets and legitimate lexical resources. The previous datasets for Urdu FND are limited-sized, domain-restricted, publicly unavailable and not manually verified where the news is translated from English into Urdu. In this paper, we curate and contribute the first largest publicly available dataset for Urdu FND, "Ax-to-Grind Urdu", to bridge the identified gaps and limitations of existing Urdu datasets in the literature. It constitutes 10,083 fake and real news on fifteen domains collected from leading and authentic Urdu newspapers and news channel websites in Pakistan and India. FN for the Ax-to-Grind dataset is collected from websites and crowdsourcing. The dataset contains news items in Urdu from the year 2017 to the year 2023. Expert journalists annotated the dataset. We benchmark the dataset with an ensemble model of mBERT, XLNet, and XLM-RoBERTa. The selected models are originally trained on multilingual large corpora. The results of the proposed model are based on performance metrics, F1-score, accuracy, precision, recall and MCC value. F1-score of 0.924, accuracy of 0.956, precision of 0.942, recall of 0.940 and an MCC value of 0.902 demonstrate the effectiveness of the proposed approach for Urdu FND. Comparison analysis with SOTA ML and DL models and existing Urdu benchmark datasets exhibit that the ensemble model outperforms them for Urdu FND. The dataset used for our experiments is publicly available at https://github.com/HjH-Whu-CRC/Ax-to-Grind-Urdu for further analysis and validation.

The aim of steganography detection is to identify whether the multimedia data contain hidden information. Although many detection algorithms have been presented to solve tasks with inconsistent distributions between the source and target domains, effectively exploiting transferable correlation information across domains remains challenging. As a solution, we present a novel multiperspective progressive structure adaptation (MPSA) scheme based on active progressive learning (APL) for JPEG steganography detection across domains. First, the source and target data originating from unprocessed steganalysis features are clustered together to explore the structures in different domains, where the intradomain and interdomain structures can be captured to provide adequate information for cross-domain steganography detection. Second, the structure vectors containing the global and local modalities are exploited to reduce nonlinear distribution discrepancy based on APL in the latent representation space. In this way, the signal-to-noise ratio (SNR) of a weak stego signal can be improved by selecting suitable objects and adjusting the learning sequence. Third, the structure adaptation across multiple domains is achieved by the constraints for iterative optimization to promote the discrimination and transferability of structure knowledge. In addition, a unified framework for single-source domain adaptation (SSDA) and multiple-source domain adaptation (MSDA) in mismatched steganalysis can enhance the model's capability to avoid a potential negative transfer. Extensive experiments on various benchmark cross-domain steganography detection tasks show the superiority of the proposed approach over the state-of-the-art methods.

The objective of this paper is to develop a new effective method for hierarchical segmentation of multitemporal ultrafine-beam synthetic aperture radar (SAR) data in urban areas. Multitemporal RADARSAT-2 ultrafine-beam highresolution horizontal transmit and horizontal receive-Synthetic Aperture Radar (HH-SAR) images acquired in the rural-urban fringe of the Greater Toronto Area during the summer of 2008 are selected for this research. Stationary wavelet transform (SWT) and algebraic multigrid (AMG) method are proposed for segmentation of SAR data. SWT is applied for decomposition of multitemporal SAR images in image preprocessing. The hierarchical and matrix-based AMG method is applied for segmentation. A pyramid of fine-to-coarse grids is constructed by iteration of selecting representative pixels and calculating the interpolation matrix between a fine-level grid and a coarse-level grid. When the pyramid is completed, segments are determined by a top-down scanning based on the interpolation matrices. The AMG techniques provide a complete hierarchical segmentation of SAR data. The experimental results show that our method produces higher accuracy than eCognition.