A survey detection channel overrides the pixels in an astronomical foundation model, and biases tomographic mean redshifts

Not provided

multimodalbiasastronomymodel auditing

2608.23626

Builder Relevance

70%

3h ago

Abstract

This paper audits the AION-1 model, revealing that biases in survey segmentation maps significantly affect the accuracy of astronomical measurements.

Reality Card

Core Claim

The study demonstrates that editing the survey segmentation map of the AION-1 model alters reported quantities by 110-4400 times, indicating a critical bias in astronomical measurements.

Method / Result

The model's performance deteriorates nine times worse when provided with contradicted catalogue photometry compared to no metadata.

Limitations

The reliance on incomplete catalogues and the model's sensitivity to survey segmentation maps raises concerns about reproducibility.

Paper to code

Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.

No verified implementation link has been attached yet. AIBuzzHub will keep this panel separate from unverified search results.

← Back to all papers