Generative adversarial networks (GANs) are a set of powerful generative models, among which CycleGAN, featuring the unique cycle-consistency loss, has gained special popularity. However, this unique structure and the cycle-consistency loss make watermarking CycleGAN particularly challenging, rendering existing deep neural network (DNN) watermarking methods, whether model-agnostic or GAN-specific, inapplicable. Meanwhile, existing DNN watermarking methods are intrusive in nature, requiring direct or indirect modification of model parameters for watermark embedding, which raises fidelity concerns. To solve the above problems, we propose the first nonintrusive and robust watermarking method for CycleGAN. We empirically show that without modifying the CycleGAN model, a user-defined watermark image can still be extracted from model outputs using a dedicated watermark decoder. Extensive experimental results verify that while achieving the so-called absolute fidelity, the proposed method is robust to various attacks, from image post-processing to model stealing.

Network traffic classification is crucial for both network security and management. Despite advances in deep learning-based multi-task traffic classification, existing models often struggle to jointly handle multiple tasks while providing interpretable insights. In multi-task scenarios, different tasks rely on distinct regions of the traffic sequence, motivating the use of dynamic and interpretable attention mechanisms. To this end, we propose Dynamic Quantized Self-Attention (DQSA), a unified framework specifically designed for multi-task network traffic classification. At its core, the Task Gated Attention Router (TGAR) dynamically associates attention heads with different tasks, enabling adaptive focus on task-specific patterns. This mechanism provides interpretable attention scores, which help analyze misclassifications and guide further model refinement. To improve efficiency and handle diverse network traffic features, we introduce the Soft Quantized Self-Attention Head (SQ-SAH) to reduce computational complexity and extend the Rotary Position Embedding (RoPE) to accommodate these features. Extensive experiments on ISCX VPN-NonVPN and DCI-LTE datasets demonstrate that DQSA consistently outperforms state-of-the-art baselines, achieving 92.85% accuracy on the encapsulation-level task of ISCX VPN-NonVPN and 93.17% accuracy on the application-level task of DCI-LTE, surpassing the strongest existing methods by up to 2.65%, while providing interpretable task-specific attention for efficient multi-task network traffic classification.

Judicial efficiency is critical to social stability. However, in many countries worldwide, grassroots courts face substantial case backlogs, and judicial decisions remain heavily dependent on judges’ cognitive efforts, with insufficient intelligent tools to enhance efficiency. To address this issue, we propose a highly efficient law article recommendation approach combining a Knowledge Graph (KG) and a Large Language Model (LLM). First, we construct a Case-Enhanced Law Article Knowledge Graph (CLAKG) to store current law articles, historical case information, and their interconnections, alongside an LLM-based automated construction method. Building on this, we propose a closed-loop law article recommendation framework integrating graph embedding-based retrieval and KG-grounded LLM reasoning. Experiments on judgment documents from China Judgments Online demonstrate that our method boosts law article recommendation accuracy from 0.549 to 0.694, outperforming strong baselines significantly. To support reproducibility and future research, all source code and processed datasets are publicly available on GitHub (see Data Availability Statement). • Proposed a novel Case-Enhanced Law Article Knowledge Graph (CLAKG) that integrates law articles and historical case data, enhancing the accuracy of law article recommendations. • Introduced an automated CLAKG construction method that uses a Large Language Model (LLM) to reduce manual input and improve scalability. Developed a closed-loop recommendation system, leveraging LLMs and CLAKG for more accurate law article recommendations while mitigating common issues like hallucinations in LLM outputs. • Achieved significant improvement in accuracy: The proposed method improved law article recommendation accuracy from 0.549 to 0.694, outperforming baseline models such as BERT, DPCNN, TFIDF-RAG, Graph-RAG and Light-RAG.

The use of learning prompts to adapt pretrained vision-language models (VLMs) for downstream tasks has gained significant attention due to its potential to reduce training costs compared to model fine-tuning through few-shot learning. Most existing methods rely on a universal prompt for all classes, as it generally delivers consistent performance across various datasets. However, a universal prompt cannot capture class-specific discriminative information. To overcome this limitation, we propose class-specific prompt learning (CPL). CPL represents the context of a prompt using two components: a base vector shared among all classes and a class-specific vector designed for individual classes. This method combines the generalization ability of the base context with the adaptability of the class-specific context. Furthermore, we introduce contrastive CPL, which enhances the ability of the prompt to capture discriminative features unique to each class. Also, we adopt the self-consistency loss to regularize the base context, enhancing its generalization ability. As a result, CPL effectively learns tailored prompts for each class. Extensive experiments demonstrate that CPL achieves superior performance over existing methods in both base-class classification and new class generalization.

The widespread use of encrypted communication in mobile networks poses significant challenges in accurately classifying traffic. Detecting out-of-distribution (OOD) samples, which significantly deviate from known classes, adds complexity to the task. This paper proposes a feature analysis-based OOD detection scheme for traffic classification in Long-Term Evolution (LTE) systems. Our method utilizes Long Short-Term Memory (LSTM) networks for feature extraction, capturing the feature vectors of the traffic series. Principal Component Analysis (PCA) is then applied to obtain principal and residual principal components. Leveraging the residual feature vector, we construct an OOD score to quantify deviation from the ID dataset. Extensive experiments on a large-scale encrypted mobile traffic dataset demonstrate the superiority of our approach, achieving high accuracy in OOD detection compared to existing techniques. Our method contributes to enhanced security and reliable traffic classification in LTE systems, addressing challenges posed by OOD samples.

Product quantization (PQ) is a technique that transforms high-dimensional data into compact binary codes to reduce data storage and improve search efficiency. However, existing PQ methods separate the learning of modality-specific features from the learning of quantization codewords, resulting in suboptimal performance in cross-modal retrieval tasks. In this paper, we propose a joint-neighborhood product quantization (JNPQ) method to simultaneously learn modality-specific features and quantization codewords. To achieve this, we first introduce a cross-modal quantization contrastive learning module that preserves the inter-modal neighborhood of the original data and reduces the quantization error. Then, we design a self-neighbor contrastive learning module that enhances the intra-modal neighborhood within individual modalities. Extensive experiments demonstrate that JNPQ achieves state-of-the-art results in crossmodal retrieval when compared with other unsupervised crossmodal quantization methods.

Transformer-based models have delivered impressive results on many tasks, particularly vision and language tasks. In many model training situations, conventional configurations are often adopted. For example, we usually set the base model with hidden size (i.e. model width) to be 768 and the number of transformer layers (i.e. model depth) to be 12. In this paper, we revisit these conventional configurations by studying the the relationship between transformer configuration and training objective. We show that the optimal transformer configuration is closely related to the training objective. Specifically, compared with the simple classification objective, the masked autoencoder is effective in alleviating the over-smoothing issue in deep transformer training. Based on this finding, we propose “Bamboo”, a notion of using deeper and narrower transformer configurations, for masked autoencoder training. On ImageNet, with such a simple change in configuration, the re-designed Base-level transformer achieves 84.2% top-1 accuracy and outperforms SoTA models like MAE by $0.9%$. On language tasks, re-designed model outperforms BERT with the default setting by 1.1 points on average, on GLUE benchmark with 8 datasets.

Classification of traffic service types is a valuable function for wireless communication networks. Even though some progress has been made, the recognition of the type of the traffic services cannot be done in real time. In this paper, we propose a novel method for classifying traffic series in real time based on transfer learning techniques. We pre-train a deep learning model with long traffic series and fine-tune the model with short traffic series. In this way, the developed model achieves the capability of recognising traffic services in real time. In other words, the model can recognize traffic services by using short traffic series. We collect Downlink Control Information (DCI) from commercial LTE networks when using five common types of traffic services. Then we use the dataset to validate our method. Our experimental results show that, by using proposed method, LSTM accuracy rates will increase to 80% and 88.5% when the length of the traffic series is 5 seconds and 10 seconds respectively, which is higher than the baseline. The strategy is also suitable for one dimension convolution neural network (1D-CNN).

Cross-modal retrieval methods are developed to retrieve relevant data across different modalities. Usually, super-vised cross-modal retrieval methods can achieve higher accuracy than unsupervised methods because they can utilize the semantic information provided by clean labels. However, training data with noisy labels will lead to the performance degradation of supervised cross-modal retrieval methods. In this work, we present a novel framework called Neighborhood Learning for Cross-Modal Retrieval (NLCMR) that is robust against noisy labels by exploiting the information contained in the neighbor-hood. Our NLCMR contains two main components: Clustering with Neighborhood Alignment and Neighborhood Contrastive Learning. The first component focuses on reducing the impact of noisy labels and improving clustering robustness, and the second component learns from noisy data by exploring pairwise and neighborhood information. Extensive experiments are conducted on three multi-modal datasets to demonstrate the effectiveness of NLCMR.

Understanding the relationship among different modalities is a challenging task. The frequently used canonical correlation analysis (CCA) and its variants have proved effective for building a common space in which the correlation between different modalities is maximized. In this paper, we show that CCA and its variants may cause information dissipation when switching the modals, and thus propose to use the continuum regression (CR) model to handle this problem. In particular, the CR model with a fixed variance coefficient of 1/2 is adopted here. We also apply the multinomial logistic regression model for further classification task. To evaluate the CR model, we perform a series of cross-modal retrieval experiments in terms of two kinds of modals, namely image and text. Compared with previous methods, experimental results show that the CR model has achieved the best retrieval precision, which demonstrates the potential of our method for real internet search applications.

Reducing custom software development effort is an important goal in information retrieval (IR). This study evaluated a generalizable approach involving with no custom software or rules development. The study used documents "consistent with cancer" to evaluate system performance in the domains of colorectal (CRC), prostate (PC), and lung (LC) cancer. Using an end-user-supplied reference set, the automated retrieval console (ARC) iteratively calculated performance of combinations of natural language processing-derived features and supervised classification algorithms. Training and testing involved 10-fold cross-validation for three sets of 500 documents each. Performance metrics included recall, precision, and F-measure. Annotation time for five physicians was also measured. Top performing algorithms had recall, precision, and F-measure values as follows: for CRC, 0.90, 0.92, and 0.89, respectively; for PC, 0.97, 0.95, and 0.94; and for LC, 0.76, 0.80, and 0.75. In all but one case, conditional random fields outperformed maximum entropy-based classifiers. Algorithms had good performance without custom code or rules development, but performance varied by specific application.