In recent years, evolutionary multitasking (EMT) methods have demonstrated significant advantages in addressing feature selection (FS) problems. However, the feature redundancy and negative transfer happened when auxiliary task construct and knowledge transfer in most of EMT-based FS methods. To reduce feature redundancy and negative transfer, this paper proposes an evolutionary multitasking with elbow principal component analysis and negative transfer optimization for high-dimensional feature selection (PN-EMTFS). PN-EMTFS integrates an elbow principal component analysis (PCA) and a negative transfer optimization strategy, thereby enhancing the performance of high-dimensional FS algorithms. The proposed algorithms are evaluated on 26 high-dimensional datasets ranging from 2308 to 22,883 dimensions. Notably, it excelled comparative algorithms in 20 datasets concerning classification accuracy, 8 datasets concerning the number of selected features, and 14 datasets concerning CPU runtime out of the 26 datasets considered. The results showed that PN-EMTFS performed at performance levels exceeding 90% for F1-score,sensitivity and sensitivity metrics, respectively.
With the advancement of digitalization, electronic documents in various formats such as Word and PDF have become the primary carriers of information. While facilitating information exchange, they also pose security risks due to their potential to conceal sensitive information. To address this, this paper proposes a cleaning, transcription, and restoration algorithm applicable to both layout-based and flow-based documents. The algorithm first converts the document into an image to eliminate information hidden within the formatting; then employs advanced OCR technology for character recognition and layout analysis. Heuristic rules are applied to precisely reconstruct the document's structure and formatting, ensuring consistency with the source document's characteristics. This paper details the algorithm's principles and workflow, verifying its feasibility through a Java-based implementation. Experiments demonstrate that this method effectively achieves cross-format document cleansing, transcription, and restoration, providing reliable technical support for promoting document interoperability and preventing information leakage.
Due to the dynamic nature of network environments, the long-term evaluation performance of website fingerprinting (WF) analysis models face significant challenges from concept drift. Traditional static approaches struggle to effectively adapt to network changes, while purely online updating methods may lead to catastrophic forgetting. To establish a continuous evaluation benchmark, this paper proposes a novel WF analysis method. We design an incremental learning framework that integrates generative replay, leveraging a generative adversarial network to synthesize pseudo-features for training, thereby eliminating the need to store original data. To tackle the issue of norm distribution discrepancies in generated traffic features, we introduce L2 normalization at the feature level. Furthermore, we adopt a frozen-generator strategy combined with knowledge distillation to mitigate memory bias in the generator. Experimental results demonstrate the effectiveness and advantages of the proposed method under concept drift scenarios, and also offer new insights for the design of defense strategies.
The molecular identification of maize varieties is important for protecting the legitimate rights and interests of farmers and maintaining national food security. However, scientific, simple, rapid, and systematic methods for germplasm identification remain limited, despite the critical need for accurate maize variety identification to prevent seed-market infringement and support agricultural security. In this study, we developed a high-throughput DNA barcoding system based on fluorescent simple sequence repeat (SSR) markers combined with capillary electrophoresis to distinguish 104 maize germplasms. After rigorous screening of 70 SSR primer pairs, 18 polymorphic loci distributed across eight chromosomes (Chr. 1–3, 5, and 7–10) were selected, demonstrating high discriminatory power, with an average polymorphism information content (PIC) value of 0.5791. The system showed strong sensitivity, with a detection limit of 0.125 ng DNA; assay specificity, with no cross-reactivity with wheat, rice, or human DNA; and reproducibility across two platforms, GA118-16B and SeqStudio™ 8 Flex. Genetic diversity analysis identified 103 allelic variants, with observed heterozygosity ranging from 0.1635 to 0.8462. Unweighted pair-group method with arithmetic means (UPGMA) clustering and principal component analysis (PCA) grouped the materials into three categories that corresponded broadly to maturity period and ecological zone. A digital DNA molecular identity card was constructed by encoding SSR banding patterns into unique quick response (QR) codes, providing a practical tool for maize varietal authentication. In addition, an optimized multiplex polymerase chain reaction (PCR) protocol incorporating allele ladder standards was developed to ensure consistent genotyping under different experimental conditions. This standardized SSR-based protocol provides a rapid, reproducible molecular identification approach for maize germplasm authentication and seed quality control, with potential application in variety protection and regulatory practice.
Open-source vulnerability evidence is distributed across vulnerability databases, project advisories, version records, code repositories, issue/PR threads, proofs of concept, and tests. These artifacts may share upstream sources, apply to different scopes, or contain inconsistent conclusions, while security fixes may evolve through backports, follow-up fixes, bypass fixes, and cross-branch variants. This paper investigates a multi-source evidence-driven framework for representing and recovering vulnerability conditions in open-source components. The method normalizes heterogeneous artifacts into provenance-, time-, and scope-anchored atomic claims, constructs patch families to recover scope-aware fix invariants, and applies temporal–scope constraints to model evolutionary supersession, alternative triggers, and genuine conflicts while preserving Unknown or Conflict states. We conduct a mechanism-oriented evaluation on 75 real vulnerabilities with traceable project-side evidence from the Maven, PyPI, and Go ecosystems. The complete framework does not outperform simpler baselines on overall condition recovery or demonstrate a consistent advantage in fix-invariant recovery. On cases containing Unknown conditions, abstention reduces Unknown-to-Known errors at the cost of determinate coverage; patch-relation and genuine-Conflict recovery remain unsuccessful in the present evaluation. The results characterize the limits of the current framework and identify specific trade-offs and unresolved requirements for evidence-grounded repair and uncertainty reasoning.