A Comparative Study of Counterfactual Explainers for Graph Neural Networks Enabling Multiple Types of Graph Edit
研究对比六种最新模型在图神经网络中的反事实解释方法,旨在通过添加或删除边来最小化修改以改变预测结果,评估其性能以指导未来研究。
研究对比六种最新模型在图神经网络中的反事实解释方法,旨在通过添加或删除边来最小化修改以改变预测结果,评估其性能以指导未来研究。
本文提出一种方法,使用大型语言模型自动构建符合FAIR数字对象规范的知识图谱,以解决文化遗产记录中元数据值缺乏持久标识符的问题。
This study addresses the significant challenges posed by inconsistent legal definitions and classification standards for child sexual abuse and exploitation material (CSAM/CSEM) across jurisdictions, which impede cross-institutional collaboration and automated processing. To overcome this, the work proposes PreventCSA@EU, a semantic-driven ontology framework that integrates, for the first time, INHOPE UCS labels, Dublin Core-DMCI metadata, and Schema.org to construct a hierarchical ontology model tailored for CSA/CSE investigations. Centered on core entities—such as media objects, content, persons, depictions, and investigative reports—the framework systematically harmonizes existing ontologies and international metadata standards. This integration enables consistent CSAM/CSEM categorization, supports child identification, and facilitates case analysis, thereby substantially enhancing interoperability across systems and laying a critical technical foundation for the European Union’s planned CSAR central database.
This work systematically investigates fundamental cryptographic primitives in quantum cryptography beyond key distribution, with a focus on quantum one-way functions and their associated constructs—such as one-way state generators and pseudorandom quantum states. By integrating quantum information theory, computational complexity, and quantum state preparation techniques, the study clarifies the mechanisms of one-wayness, physical realizability, and noise resilience under both computational and information-theoretic security frameworks across various adversarial models. The paper delineates conceptual relationships among diverse quantum cryptographic primitives, reviews and compares existing constructions, and identifies key open problems, thereby laying a theoretical foundation for practical quantum cryptographic systems that extend beyond quantum key distribution.
This study addresses the challenge of fine-grained food recognition, which is hindered by high intra-class variation and strong visual similarity among dishes, thereby limiting the accuracy of image-based dietary assessment. Building upon the PaliGemma-2-3B architecture, the authors employ parameter-efficient fine-tuning via LoRA and integrate three European dietary datasets into a unified dish vocabulary to train a vision-language model capable of multi-task instruction-based question answering. This approach achieves state-of-the-art performance in a domain-specific setting, surpassing large proprietary models such as Gemini, GPT, and Claude. In 3-fold cross-validation, the model attains 92.96% Top-1 dish recognition accuracy and 90.79% Exact-Set ingredient accuracy, significantly outperforming both CNN baselines and existing state-of-the-art methods.
研究对比六种最新模型在图神经网络中的反事实解释方法,旨在通过添加或删除边来最小化修改以改变预测结果,评估其性能以指导未来研究。
本文提出一种方法,使用大型语言模型自动构建符合FAIR数字对象规范的知识图谱,以解决文化遗产记录中元数据值缺乏持久标识符的问题。
This study addresses the significant challenges posed by inconsistent legal definitions and classification standards for child sexual abuse and exploitation material (CSAM/CSEM) across jurisdictions, which impede cross-institutional collaboration and automated processing. To overcome this, the work proposes PreventCSA@EU, a semantic-driven ontology framework that integrates, for the first time, INHOPE UCS labels, Dublin Core-DMCI metadata, and Schema.org to construct a hierarchical ontology model tailored for CSA/CSE investigations. Centered on core entities—such as media objects, content, persons, depictions, and investigative reports—the framework systematically harmonizes existing ontologies and international metadata standards. This integration enables consistent CSAM/CSEM categorization, supports child identification, and facilitates case analysis, thereby substantially enhancing interoperability across systems and laying a critical technical foundation for the European Union’s planned CSAR central database.
This work systematically investigates fundamental cryptographic primitives in quantum cryptography beyond key distribution, with a focus on quantum one-way functions and their associated constructs—such as one-way state generators and pseudorandom quantum states. By integrating quantum information theory, computational complexity, and quantum state preparation techniques, the study clarifies the mechanisms of one-wayness, physical realizability, and noise resilience under both computational and information-theoretic security frameworks across various adversarial models. The paper delineates conceptual relationships among diverse quantum cryptographic primitives, reviews and compares existing constructions, and identifies key open problems, thereby laying a theoretical foundation for practical quantum cryptographic systems that extend beyond quantum key distribution.
This study addresses the challenge of fine-grained food recognition, which is hindered by high intra-class variation and strong visual similarity among dishes, thereby limiting the accuracy of image-based dietary assessment. Building upon the PaliGemma-2-3B architecture, the authors employ parameter-efficient fine-tuning via LoRA and integrate three European dietary datasets into a unified dish vocabulary to train a vision-language model capable of multi-task instruction-based question answering. This approach achieves state-of-the-art performance in a domain-specific setting, surpassing large proprietary models such as Gemini, GPT, and Claude. In 3-fold cross-validation, the model attains 92.96% Top-1 dish recognition accuracy and 90.79% Exact-Set ingredient accuracy, significantly outperforming both CNN baselines and existing state-of-the-art methods.