FIX Benchmark
收藏资源简介:
FIX Benchmark是由宾夕法尼亚大学开发的一个用于评估特征解释性的基准数据集,包含6个不同领域的数据集,涵盖宇宙学、心理学和医学等多个应用领域。数据集包括图像、文本和时间序列信号等多种数据类型,总共有超过20万条数据。数据集的创建过程涉及与领域专家的合作,旨在通过自动提取与专家知识对齐的特征来提高模型的解释性。FIX Benchmark的应用领域广泛,旨在解决高维数据中特征解释性的问题,特别是在医疗、法律和教育等需要高度透明性的领域。
FIX Benchmark is a benchmark dataset developed by the University of Pennsylvania for evaluating feature interpretability. It consists of six datasets across diverse application domains including cosmology, psychology, and medicine. The dataset supports multiple data modalities such as images, text, and time-series signals, with a total of over 200,000 data instances. The development of FIX Benchmark involved collaboration with domain experts, with the objective of improving model interpretability by automatically extracting features aligned with expert knowledge. FIX Benchmark has broad application potential, aiming to resolve feature interpretability issues in high-dimensional data, particularly in fields requiring high transparency such as healthcare, law, and education.

- 1The FIX Benchmark: Extracting Features Interpretable to eXperts宾夕法尼亚大学 · 2024年



