This page organizes papers related to artificial intelligence published around the world. This page is summarized using Google Gemini and is operated on a non-profit basis. The copyright of the paper belongs to the author and the relevant institution. When sharing, simply cite the source.
Interpreting deep learning-based stellar mass estimation via causal analysis and mutual information decomposition
Created by
Haebom
Author
Wei Zhang, Qiufan Lin, Yuan-Sen Ting, Shupei Chen, Hengxin Ruan, Song Li, Yifan Wang
Outline
End-to-end deep learning models trained on multiband galaxy images are powerful data-driven tools for estimating galaxy physical properties without spectroscopy. However, the lack of interpretability and the associative nature of these models make it difficult to understand how information beyond integrated luminosity (e.g., morphology) contributes to the estimation. This study aims to interpret deep learning-based stellar mass estimates using two interpretability techniques: causal analysis and mutual information decomposition. Using data from the Sloan Digital Sky Survey (SDSS) and the Wide Field Infrared Survey (WISE), we obtain meaningful results that provide a physical interpretation of the image-based models.
Takeaways, Limitations
•
We demonstrate the benefits of combining deep learning and interpretability techniques.
•
We provide a physical interpretation of image-based models for stellar mass estimation.
•
It offers the potential to facilitate data-driven astrophysical research, such as estimation of astrophysical parameters and studies of complex multivariate physical processes.
•
It is limited to two interpretability techniques: causal analysis and mutual information decomposition.
