Vehicle Ad Hoc Networks (VANET) are a technology that allows existing transportation systems to deliver safety and critical information. Due to real-time communication, it generates massive amounts of data, making it vulnerable to eavesdropping and interference. Intrusion Detection Systems (IDS) are crucial devices that detect and respond to malicious activity. This paper develops a novel network intrusion detection system for VANET that leverages Spark-based big data optimization and transfer learning (NIDS-VSB). First, a packet parser is used to crawl network traffic and filter required flow events. Next, the Spark-based optimization method is implemented to process large amounts of data efficiently. Furthermore, a transfer learning strategy is designed to learn the comprehensive feature representation using their semantic anchors. A Convolutional Neural Network (CNN) model is used to extract deep features from semantic anchors. Finally, a stacking generalization ensemble model uses deep features to detect various attacks. The proposed approach is tested on three standard and publicly available datasets, namely the CIC-IoT 2022, CIC-IDS2017, and NSL_KDD datasets. Our findings show that the proposed approach delivers the best detection results for a variety of attacks including audio, HTTP, BruteForce, web, and others.

The rapid advancement of machine learning (ML) technologies necessitates robust security frameworks to protect the integrity of ML model training processes. Proof-of-Learning (PoL) is a critical method for verifying the computational effort in trainingMLmodels, while model watermarking is a strategy for asserting model ownership. This research integrates PoL with feature-based model watermarking, embedding thewatermark directly into the model’s features or parameters. This integration mitigates security risks associated with external key management and reduces computational overhead by eliminating the need for complex verification procedures. Our proposed dual-layered verification architecture embeds unique watermarks during the training phase. It records them alongside PoL proofs, enhancing security against sophisticated spoofing attacks where adversaries attempt to mimic a model’s computational trajectory and watermark. This approach addresses critical challenges, including maintaining watermark robustness and balancing security with model performance. Through a comprehensive analysis, we identify vulnerabilities in existing PoL systems and demonstrate howfeature-basedwatermarking can enhance security.We present a secure PoL mechanism, supported by empirical validation, that significantly improves resilience to spoofing attacks. This advancement represents a crucial step towards securing ML models, paving the way for future research to protect diverse ML applications from various threats.

Community detection in bipartite networks has attracted significant attention due to its broad applications. However, the structural complexity of bipartite networks makes it challenging to directly apply traditional one-mode community detection methods. To address this limitation, this paper proposes a Network Reconstruction Framework (NRF), which enables one-mode algorithms to be directly applied to community detection in bipartite networks by transforming bipartite networks into weighted asymmetric networks. The core of this framework lies in calculating the interaction force based on the common neighbors of node pairs, and reconstructing the network with this as the weight, which retains the original structural information while enhancing the quantitative expression of node associations. To verify the effectiveness of NRF, we integrated three classic one-mode algorithms—Spectral Clustering, Label Propagation, and Louvain—into the framework, conducted experiments on synthetic and real-world network datasets, and compared them with four bipartite network community detection algorithms. The results show that the three NRF-based algorithms not only can stably and accurately identify community structures, but also significantly outperform traditional methods, confirming the superiority of NRF in balancing computational efficiency and structural information retention.

Proof-of-Learning (PoL) attests training effort but remains vulnerable to tolerance-based spoofing, while model watermarking (WM) protects ownership without verifying how a model was trained. We present SecurePoL, a dual-layer framework that couples immutable PoL logs with three watermarking strategies: feature-based triggers, sparse parameter perturbations, and a non-intrusive auxiliary head, so that verification succeeds only when both the training trajectory and the watermark are consistent. This design forces an attacker to satisfy trajectory consistency and watermark integrity at the same time, rather than bypassing each mechanism in isolation. On CIFAR-10 with ResNet-20, SecurePoL raises the cost of blindfold Top-Q and infinitesimal-update attacks while preserving task utility: accuracy changes are 0.00, −0.03, and −0.58 percentage points for the three strategies, with runtime overheads between 0.6% and 17.3% and proof logs below 12 MB. These results, together with an analysis of hyperparameter sweeps and attacker models, indicate that integrating PoL with watermarking is a practical path toward secure, auditable, and tamper-evident ML training.

Medical image segmentation plays a critical role in computer-assisted diagnosis, treatment planning, surgical navigation, and quantitative medical image analysis. Among deep learning-based segmentation architectures, U-Net has become one of the most influential frameworks because of its encoder–decoder structure and skip connections for integrating semantic and spatial information. This review focuses on U-Net and representative variants for two-dimensional medical image segmentation. We systematically analyze their architectural evolution, including attention-based designs, enhanced skip connections, deformable convolutions, and Transformer-based extensions and discuss their advantages, limitations, and application boundaries. Quantitative performance results are synthesized from published studies rather than reproduced through new benchmarking experiments; direct numerical comparisons are restricted to models evaluated within the same study or under clearly defined benchmark settings. We further relate architectural modifications to modality-specific challenges in MRI, CT, ultrasound, fundus imaging, dermoscopy, and microscopy. Finally, current challenges in computational efficiency, cross-domain generalization, data scarcity, interpretability, and clinical deployment are discussed, together with promising directions for future research.