SPUCOMNIST, SPUCOANIMALS

SPUCOMNIST和SPUCOANIMALS是两个新开发的视觉数据集。SPUCOMNIST基于CMNIST数据集，通过添加彩色背景作为虚假特征，用于研究不同难度和噪声条件下的虚假特征学习。SPUCOANIMALS则从ImageNet中提取，包含四个类别，旨在更真实地反映现实世界中的虚假相关性。这两个数据集均设计用于评估和改进机器学习模型在处理虚假相关性方面的性能，特别是在视觉识别任务中。

SPUCOMNIST and SPUCOANIMALS are two newly developed visual datasets. SPUCOMNIST is based on the CMNIST dataset, with colored backgrounds added as spurious features, and it is used to study spurious feature learning under different difficulty and noise conditions. SPUCOANIMALS is extracted from ImageNet, includes four categories, and aims to more realistically reflect spurious correlations in the real world. Both datasets are designed to evaluate and improve the performance of machine learning models when dealing with spurious correlations, especially in visual recognition tasks.

搜集汇总

背景与挑战

背景概述

SPUCOMNIST和SPUCOANIMALS是两个新开发的视觉数据集，专注于虚假相关性研究。SPUCOMNIST基于CMNIST数据集，通过添加彩色背景作为虚假特征，用于测试不同难度和噪声条件下的模型学习；SPUCOANIMALS从ImageNet提取，包含四个类别，更贴近现实世界中的虚假相关场景。两者均旨在评估和改进机器学习模型在视觉识别任务中处理虚假相关性的性能。