Pu Embeddings: Foundation-Model Image Embeddings for Astronomy Surveys

The companion data release for The Platonic Universe: Do Foundation Models See the Same Sky? (UniverseTBD et al. 2025, arXiv:2509.19453). This dataset contains image embeddings from foundation models applied to astronomical survey data, testing the Platonic Representation Hypothesis. It was authored by UniverseTBD and last updated on 2026-04-20.

Use Cases

• Test the Platonic Representation Hypothesis based on embeddings from different foundation models mentioned in the description.
• Compare latent representations of astronomical images based on embeddings generated by models with different training objectives.
• Analyze the convergence of learned features across different model modalities based on the astronomical data described.
• Benchmark foundation model performance on astronomical tasks based on the provided image embeddings.

Strengths

• Companion to a specific 2025 research paper (arXiv:2509.19453) with a clear research question.
• Last updated on 2026-04-20, indicating recent maintenance.
• Focuses on a novel research area testing representation convergence across foundation models.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

UniverseTBD

Collection Method

Likely generated by processing astronomical survey images through various foundation models.

Freshness

Last updated 2026-04-20 23:34:21.