The availability of large, public, multi-modal astronomical datasets presents
an opportunity to execute novel research that straddles the line between
science of AI and science of astronomy. Photometric redshift estimation is a
well-established subfield of astronomy. Prior works show that computer vision
models typically outperform catalog-based models, but these models face
additional complexities when incorporating images from more than one instrument
or sensor. In this report, we detail our progress creating Mantis Shrimp, a
multi-survey computer vision model for photometric redshift estimation that
fuses ultra-violet (GALEX), optical (PanSTARRS), and infrared (UnWISE) imagery.
We use deep learning interpretability diagnostics to measure how the model
leverages information from the different inputs. We reason about the behavior
of the CNNs from the interpretability metrics, specifically framing the result
in terms of physically-grounded knowledge of galaxy properties.

我们详细介绍了我们创建的多调查计算机视觉模型 Mantis Shrimp，用于光度红移估计，它融合了紫外（GALEX）、光学（PanSTARRS）和红外（UnWISE）图像，并使用深度学习可解释性诊断来衡量模型如何利用不同输入的信息并推理卷积神经网络的行为，特别是在物理根据的星系属性方面。

对螳螂虾的初步报告：多次调查的计算机视觉光度红移模型

Preliminary Report on Mantis Shrimp: a Multi-Survey Computer Vision  Photometric Redshift Model

Sky surveys are the largest data generators in astronomy, making automated
tools for extracting meaningful scientific information an absolute necessity.
We show that, without the need for labels, self-supervised learning recovers
representations of sky survey images that are semantically useful for a variety
of scientific tasks. These representations can be directly used as features, or
fine-tuned, to outperform supervised methods trained only on labeled data. We
apply a contrastive learning framework on multi-band galaxy photometry from the
Sloan Digital Sky Survey (SDSS) to learn image representations. We then use
them for galaxy morphology classification, and fine-tune them for photometric
redshift estimation, using labels from the Galaxy Zoo 2 dataset and SDSS
spectroscopy. In both downstream tasks, using the same learned representations,
we outperform the supervised state-of-the-art results, and we show that our
approach can achieve the accuracy of supervised models while using 2-4 times
fewer labels for training.

使用自监督学习的方式学习天文学中的巨大数据源，通过图像表示的学习，可以在不需要过多标记的情况下获得与使用有监督学习方法获得的结果相当甚至更好的效果，可能可以有效减少标记数量。

天文图像的自监督表征学习

Self-Supervised Representation Learning for Astronomical Images

The need to analyze the available large synoptic multi-band surveys drives
the development of new data-analysis methods. Photometric redshift estimation
is one field of application where such new methods improved the results,
substantially. Up to now, the vast majority of applied redshift estimation
methods have utilized photometric features. We aim to develop a method to
derive probabilistic photometric redshift directly from multi-band imaging
data, rendering pre-classification of objects and feature extraction obsolete.
A modified version of a deep convolutional network was combined with a mixture
density network. The estimates are expressed as Gaussian mixture models
representing the probability density functions (PDFs) in the redshift space. In
addition to the traditional scores, the continuous ranked probability score
(CRPS) and the probability integral transform (PIT) were applied as performance
criteria. We have adopted a feature based random forest and a plain mixture
density network to compare performances on experiments with data from SDSS
(DR9). We show that the proposed method is able to predict redshift PDFs
independently from the type of source, for example galaxies, quasars or stars.
Thereby the prediction performance is better than both presented reference
methods and is comparable to results from the literature. The presented method
is extremely general and allows us to solve of any kind of probabilistic
regression problems based on imaging data, for example estimating metallicity
or star formation rate of galaxies. This kind of methodology is tremendously
important for the next generation of surveys.

本文介绍了一种基于卷积网络和混合密度网络的新方法，并使用 SDSS（DR9）中的数据进行实验，证明此方法相比之前的方法具有更好的性能，可以预测任何类型数据（例如星系、类星体、恒星）的物理参数概率密度函数，是解决基于成像数据的概率性回归问题的一种通用方法。