A Shared-Backbone Approach for Multi-Task MedMNIST Classification
研究通过共享骨干网络和任务特定线性头解决多任务MedMNIST分类中的模态与类别分布差异问题,最佳模型使用ConvNeXt-Tiny和标签平滑技术。
研究通过共享骨干网络和任务特定线性头解决多任务MedMNIST分类中的模态与类别分布差异问题,最佳模型使用ConvNeXt-Tiny和标签平滑技术。
该研究提出了一种65M参数的生成式恢复器,通过使用Haar变换和改进的MeanFlow目标函数,在保持高效处理速度的同时实现高质量JPEG图像恢复。
This study addresses the computational challenges posed by high-resolution digital elevation models (DEMs) in calculating topographic prominence at a global scale. The authors propose an approximately linear-time algorithm that efficiently computes the prominence of all peaks worldwide. The key insight is that the prominence of the vast majority of peaks is determined solely by local terrain, with only a small fraction influenced by distant higher summits. Leveraging this observation, the method refines classical algorithms and incorporates memoization to drastically reduce redundant computations and memory usage. Validation on real-world 3-arcsecond SRTM data demonstrates that the approach achieves substantial gains in computational efficiency while preserving the correctness of prominence values, enabling scalable prominence analysis across massive DEM datasets.
This study evaluates the reliability of large language models (LLMs) in static security analysis of smart contracts, investigating whether they can replace or merely complement traditional tools. To this end, we introduce the first automated evaluation framework that systematically assesses LLM performance in vulnerability detection, quantitatively revealing— for the first time—high false positive rates stemming from lexical biases (e.g., identifier naming) and insufficient semantic validation. Through extensive experiments with diverse prompting strategies, we observe a pronounced trade-off between precision and recall. Our framework achieves 92% accuracy in classifying model outputs, demonstrating that current LLMs are ill-suited for standalone security auditing but show promise as collaborative aids to conventional static analysis tools, thereby underscoring the necessity of hybrid approaches.
This work proposes a hybrid optimization framework that integrates a multimodal genetic algorithm with graph neural networks (GNNs) to address the challenge of balancing search efficiency and solution quality in workforce scheduling. For the first time, a GNN is embedded within the genetic algorithm as an enhancement operator, leveraging its capacity to model domain-specific knowledge—such as scheduling constraints and employee preferences—in a structured manner to guide the evolutionary search and improve solution quality. Experimental results on real-world employee scheduling tasks demonstrate that the proposed approach significantly outperforms standalone genetic algorithms or GNNs in both solution quality and computational efficiency, with performance gains confirmed to be statistically significant.
研究通过共享骨干网络和任务特定线性头解决多任务MedMNIST分类中的模态与类别分布差异问题,最佳模型使用ConvNeXt-Tiny和标签平滑技术。
该研究提出了一种65M参数的生成式恢复器,通过使用Haar变换和改进的MeanFlow目标函数,在保持高效处理速度的同时实现高质量JPEG图像恢复。
This study addresses the computational challenges posed by high-resolution digital elevation models (DEMs) in calculating topographic prominence at a global scale. The authors propose an approximately linear-time algorithm that efficiently computes the prominence of all peaks worldwide. The key insight is that the prominence of the vast majority of peaks is determined solely by local terrain, with only a small fraction influenced by distant higher summits. Leveraging this observation, the method refines classical algorithms and incorporates memoization to drastically reduce redundant computations and memory usage. Validation on real-world 3-arcsecond SRTM data demonstrates that the approach achieves substantial gains in computational efficiency while preserving the correctness of prominence values, enabling scalable prominence analysis across massive DEM datasets.
This study evaluates the reliability of large language models (LLMs) in static security analysis of smart contracts, investigating whether they can replace or merely complement traditional tools. To this end, we introduce the first automated evaluation framework that systematically assesses LLM performance in vulnerability detection, quantitatively revealing— for the first time—high false positive rates stemming from lexical biases (e.g., identifier naming) and insufficient semantic validation. Through extensive experiments with diverse prompting strategies, we observe a pronounced trade-off between precision and recall. Our framework achieves 92% accuracy in classifying model outputs, demonstrating that current LLMs are ill-suited for standalone security auditing but show promise as collaborative aids to conventional static analysis tools, thereby underscoring the necessity of hybrid approaches.
This work proposes a hybrid optimization framework that integrates a multimodal genetic algorithm with graph neural networks (GNNs) to address the challenge of balancing search efficiency and solution quality in workforce scheduling. For the first time, a GNN is embedded within the genetic algorithm as an enhancement operator, leveraging its capacity to model domain-specific knowledge—such as scheduling constraints and employee preferences—in a structured manner to guide the evolutionary search and improve solution quality. Experimental results on real-world employee scheduling tasks demonstrate that the proposed approach significantly outperforms standalone genetic algorithms or GNNs in both solution quality and computational efficiency, with performance gains confirmed to be statistically significant.