PathAgentBench
收藏资源简介:
PathAgentBench是由新加坡国立大学、PuzzleLogic及北京协和医院等机构联合构建的病理学基准数据集,旨在评估视觉语言模型在全切片图像上的证据寻求能力。该数据集整合了来自癌症基因组图谱的1,822张全切片图像和一个包含190张乳腺癌切片的私有队列,共计涵盖16种器官类型,并提供了由十位认证病理学家标注的17,135条多尺度诊断路径,每条路径包含嵌套的边界框、特定放大倍率的发现以及最终诊断结论。数据集的构建过程基于诊断树框架,通过系统化的标注工作流程,将病理诊断分解为证据解释、验证、获取与整合四个核心阶段。该数据集主要应用于计算病理学领域,旨在解决现有基准在评估模型主动从千兆像素级图像中获取和整合诊断证据能力方面的不足,推动病理学智能体向端到端诊断探索的发展。
FTPrimitiveBench is a quantum computing benchmark suite jointly developed by the Pacific Northwest National Laboratory and Fordham University, aiming to evaluate the robustness of logical computations under hardware-relevant noise models. This dataset includes simulated circuits for surface code Clifford primitives (such as logical storage, lattice surgery, etc.), supports custom noise parameter configuration and spatio-temporal inhomogeneity modeling, and the data is generated via high-performance computing. Its innovation lies in the collaborative optimization of hardware calibration features (such as Pauli bias and measurement noise) and quantum error correction protocols, providing a standardized analytical tool for hardware-software co-design of fault-tolerant quantum architectures.




