Controllable image synthesis, which enables fine-grained control over generated outputs, has emerged as a key focus in visual generative modeling. However, controllable generation remains challenging for Visual Autoregressive (VAR) models due to their hierarchical, next-scale prediction style. Existing VAR-based methods often suffer from inefficient control encoding and disruptive injection mechanisms that compromise both fidelity and efficiency. In this work, we present SCALAR, a controllable generation method based on VAR, incorporating a Scale-wise Conditional Decoding mechanism. SCALAR leverages a pretrained image encoder to extract semantic control signal encodings, which are projected into scale-specific representations and injected into the corresponding layers of the VAR backbone. This design provides persistent and structurally aligned guidance throughout the generation process. Building on SCALAR, we develop SCALAR-Uni, a unified extension that aligns multiple control modalities into a shared latent space, supporting flexible multi-conditional guidance in a single model. Extensive experiments show that SCALAR achieves superior generation quality and control precision across various tasks.

Unmanned aerial vehicles (UAVs) offer a safe and economical alternative to manual patrols of overhead transmission lines, yet reliable localisation of missing cotter pins remains difficult because the targets are pixel–scale and often masked by cluttered towers. This work presents SADL–YOLOv11, a lightweight single-stage detector embedded in an end-to-end UAV workflow. A visibility-aware waypoint optimiser first plans 3-D trajectories that maximise pin exposure while satisfying safety envelopes reconstructed from LiDAR point clouds. The captured images are analysed on board by the proposed network, which extends a YOLOv11 baseline with a slim-neck (GSConv, VoVGSCSP, DySample), a micro-scale P2 head, an Adaptive Fine-Grained Channel Attention module, a Normalized-Wasserstein Distance loss for sub-pixel box regression, and a detail-enhanced weight-shared head. Group-Taylor pruning and feature–logit distillation further compress the model for edge deployment. On the Pin_Defect_Dataset, SADL–YOLOv11 attains 99.5% [email protected] and 87.8% [email protected]:0.95. Under adverse imaging on the Pin_Low_Quality_Dataset (low light, blur/over-exposure, and very small targets), it achieves 97.0% [email protected] and 73.3% [email protected]:0.95. Cross-domain transfer to PKU-Market-PCB confirms generalisation: [email protected] matches YOLOv11 (78.5%) and [email protected]:0.95 improves by +3.2 points. The detector sustains 66.8 FPS on an RTX 3090 and 18–28 FPS on an RK3568 edge SoC, delivering a practical, high-precision, IoT-ready solution for large-scale power-grid inspection. Source code of proposed SADL-YOLOv11 model and related data-sets are available at the following link: https://github.com/haobai816/ultralytics-yolo11-Pin-Pro_SADL-YOLOv11.git.

Model inference in edge is often regarded as an effective method to alleviate high latency and enhance data privacy in edge-cloud collaborative computing environment. In this paper, we demonstrate that optimizing network communication in edge-cloud environment with limited bandwidth can enhance model inference performance. We first analyze network bottlenecks and the characteristics in model inference, then design a serverless inference system - EINS, to support collaborative optimization of network transmission and inference performance in edge-cloud environment. This system identifies concurrent network communication bottlenecks in multi-model deployment, dynamically scales capacity, and optimizes placement strategies and model transfer sequences. Real-world workload evaluation reveal that EINS can achieve a 5.7x throughput improvement and an average reduction of 62% latency in model instance startup.

With the dramatic increase in the number and size of software in the industry, tremendous research has been studied to automatically detect vulnerabilities. However, existing detection methods have limitations in code semantic modeling and detection granularity, which makes them unable to meet the requirements of high accuracy and fine granularity at the same time. In this paper, we propose a general framework, namely VulDL, which can effectively identify whether a given code snippet has a vulnerability and locate the specific code line where the vulnerability resides. VulDL first represents the source code as a novel semantic data structure, namely the adapted code property graph. After that, tree-based and graph-based neural networks are designed, which learn features according to the hierarchies and neighborhoods, and further realize vulnerability identification and localization. Our evaluation shows that VulDL achieves F1-scores of 98.68% and 94.85% in the identification of buffer error and resource management error vulnerabilities and 97.73% on their combined vulnerabilities. On the FFmpeg+QEMU dataset, VulDL achieves an F1-score of 59.62%, which is more effective than existing methods. Besides, VulDL can locate vulnerabilities at the statement granularity with F1-scores of 97.88%, 98.31%, and 99.16% on the evaluated datasets.

With the explosive development of information technology, vulnerabilities have become one of the major threats to computer security. Most vulnerabilities with similar patterns can be detected effectively by static analysis methods. However, some vulnerable and non-vulnerable code is hardly distinguishable, resulting in low detection accuracy. In this paper, we define the accurate identification of vulnerabilities in similar code as a fine-grained vulnerability detection problem. We propose VulSniper which is designed to detect fine-grained vulnerabilities more effectively. In VulSniper, attention mechanism is used to capture the critical features of the vulnerabilities. Especially, we use bottom-up and top-down structures to learn the attention weights of different areas of the program. Moreover, in order to fully extract the semantic features of the program, we generate the code property graph, design a 144-dimensional vector to describe the relation between the nodes, and finally encode the program as a feature tensor. VulSniper achieves F1-scores of 80.6% and 73.3% on the two benchmark datasets, the SARD Buffer Error dataset and the SARD Resource Management Error dataset respectively, which are significantly higher than those of the state-of-the-art methods.

Non-Chinese speakers hold increasing opportunities and need to process Chinese information and communicate in Chinese. This paper, with the purpose of facilitating the handwriting input of Chinese characters for non-Chinese speakers, is directed towards the development of the handwriting rules and vocabulary for Latin-style anti-cursive characters and the ways of their selection and classification. This aims to build a practical platform by utilizing three characteristics of wavelet neural network — automatically ascertaining the number of hidden layer unit, converging rapidly and never running into the partial minimum of networks — for a simple Latin-style online handwriting input and processing, meanwhile, taking the customary handwriting habits of non-Chinese speakers. The paper, based on profound information of cursive characters, deciphered the genetic code of ancient cursive symbols and made clear the rules for characters changing into its cursive style. As a result, it breaks the bottleneck, which enables non-Chinese speakers to easily input information through handwriting Chinese characters.