BLOCKIES
收藏资源简介:
BLOCKIES数据集是由比勒费尔德大学认知交互技术中心(CITEC)的David S. Johnson等人提出的参数化方法生成的模拟诊断任务数据集。该数据集能够控制用于训练实际世界模型的数据特征和偏差,旨在为视觉决策任务中的高风险人机协作评估提供大规模、在线的评价框架。数据集通过Blender生成虚拟生物体Blockies的X射线图像,并根据症状 severity 分配不同的特征分布。BLOCKIES能够支持创建用于特性诊断任务的自定义数据集,并包含用于生成诊断建议的真实世界黑盒模型。该数据集用于评估在不同风险条件下的人机协作方法,如XAI,以提高高风险环境中的人机协作效率。
The BLOCKIES dataset is a simulated diagnostic task dataset generated via a parametric approach proposed by David S. Johnson et al. from the Center for Cognitive Interaction Technology (CITEC), Bielefeld University. This dataset enables control over the data features and biases used for training real-world models, and aims to provide a large-scale, online evaluation framework for high-stakes human-computer collaboration assessment in visual decision-making tasks. The dataset generates X-ray images of virtual organisms named Blockies using Blender, and assigns distinct feature distributions based on symptom severity. BLOCKIES supports the creation of custom datasets for targeted diagnostic tasks, and includes real-world black-box models for generating diagnostic recommendations. This dataset is utilized to evaluate human-computer collaboration methods such as XAI under varying risk conditions, so as to enhance the efficiency of human-computer collaboration in high-stakes environments.




