Research map / Vision-language and multimodal / Cross-modal learning
COSMIO: A Benchmark for Cross-Survey Modality Imputation
Dichang Zhang, Yixuan Shao, Jiali Cui, Haotian Yin, Yuanpeng Liu, Jiaiq Deng, Xiang Gao, Zhiqiang Lao, Heather Yu, Simon Birrer, Dimitris Samaras
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces COSMIO, a benchmark for cross-survey modality imputation over aligned multi-survey astronomical observations, where each object is seen by only a subset of surveys.
Problem
Large-scale astronomical surveys yield heterogeneous, partially observed multi-survey data: each object is observed by only some surveys, and surveys differ in resolution, wavelength coverage, noise, and footprint. Missing survey modalities must be imputed.
Approach
Formalizes the setting as cross-survey modality imputation and provides a benchmark of aligned multi-survey observations to evaluate imputation of missing survey modalities.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use MRI-specific imputation architectures instead of simple diffusion | medical imputation baselines (MMGAN, UMMI, DiffM4RI) · COSMIO common galaxies -> Roman | CRPS: 0.159 (MMGAN) vs 0.069 (Asyn-Diff) | Hurt |
| use asynchronous modality-level diffusion (Asyn-Diff) instead of pattern-specific Palette | diffusion imputation · COSMIO common galaxies -> Roman | CRPS: 0.072 -> 0.069 | Improved |
| train unified single model (RePaint) instead of pattern-specific Palette | diffusion imputation · common galaxies -> Roman | CRPS: 0.072 -> 0.090 | Hurt |
| evaluate on rare strong lenses instead of common galaxies (zero-shot) | Asyn-Diff · Roman target | CRPS: 0.069 -> 0.090 | Hurt |
| evaluate on rare strong lenses instead of common galaxies | Palette · Roman target | CRPS: 0.072 -> 0.095 | Hurt |
| use imputed Roman images from better imputation | downstream lens classifier · Roman lens detection | AUPRC: 0.5027 (RePaint) -> 0.6104 (Asyn-Diff) | Improved |
| lower assumed lens prevalence from 0.1 to 0.001 | Asyn-Diff classifier · Roman lens detection | AUPRC: 0.6104 -> 0.0411 | Hurt |
| physically motivated baselines instead of learned diffusion | PSF-Linear · Euclid -> LSST | RMSE: 0.285 vs 0.332 (Asyn-Diff) | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CodeBrain — CodeBrain: Towards Decoupled Interpretability and Multi-Scale Architecture for EEG Foundation Model (ICLR 2026) · better on 5 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Experimental setup
Benchmarks: COSMIO
Closest papers by meaning
- Learning What's Real: Disentangling Signals and Measurement Artifacts in Multi-Sensor Data, with Applications to Astrophysics (NeurIPS 2026 · similarity 0.84)
- CosmoBench: A Multiscale, Multiview, Multitask Cosmology Benchmark for Geometric Deep Learning (NeurIPS 2025 · similarity 0.84)
- The Multimodal Universe: Enabling Large-Scale Machine Learning with 100 TB of Astronomical Scientific Data (NeurIPS 2024 · similarity 0.83)
- A Data-Driven Prism: Multi-View Source Separation with Diffusion Model Priors (NeurIPS 2025 · similarity 0.82)
- VESTA: Visual Exploration with Statistical Tool Agents (NeurIPS 2026 · similarity 0.82)
- A systematic evaluation of vision-language models for observational astronomical reasoning tasks (NeurIPS 2026 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
