UniverseTBD/pu-embeddings
收藏资源简介:
README内容描述了一个名为pu-embeddings的数据集,包含用于天文调查的基础模型图像嵌入。这是UniverseTBD等人(2025年)的论文《柏拉图宇宙:基础模型看到的天空是否相同?》的配套数据发布。数据集包括由各种预训练模型应用于不同调查图像生成的嵌入,旨在测试天文数据上的柏拉图表示假设。数据集按目录结构组织,每个目录对应一个特定的调查,并包含用于跨模态比较的配对嵌入。README还提到了使用互k近邻(MKNN)度量来衡量模型之间的表示对齐。
The README content describes a dataset named pu-embeddings, which contains foundation-model image embeddings for astronomy surveys. It is a companion data release for a paper titled The Platonic Universe: Do Foundation Models See the Same Sky? by UniverseTBD et al. (2025). The dataset includes embeddings produced by various pre-trained models applied to different survey imagery, with the goal of testing the Platonic Representation Hypothesis on astronomical data. The dataset is structured into directories, each corresponding to a specific survey, and contains paired embeddings for cross-modal comparison. The README also mentions the use of the Mutual k-Nearest Neighbour (MKNN) metric for measuring representation alignment between models.




