Abstract:Data-driven approaches play a crucial role in space computing, and our paper focuses on analyzing data to learn more about celestial objects. Photometric redshift, a measure of the shift of light towards the red part of the spectrum, helps determine the distance of celestial objects. This study used a dataset from the Sloan Digital Sky Survey (SDSS) with five magnitudes alongside their corresponding redshift labels. Traditionally, redshift prediction relied on spectral distribution templates (SEDs), which, though effective, are costly and limited, especially for large datasets. This paper explores data-driven methodologies instead of SEDs. By employing a decision tree regressor and a fully connected neural network (FCN), we found that the FCN outperforms the decision tree regressor in RMS. The results show that data-driven estimation is a valuable tool for astronomical surveys. With the adaptability to complement previous methods, FCNs will reshape the field of redshift estimation.

What problem does this paper attempt to address?

### What problems does this paper attempt to solve? This paper aims to explore a deep - learning - based method to estimate photometric redshift in order to improve the measurement accuracy of celestial distances. Specifically, the paper focuses on the following key issues: 1. **Limitations of traditional methods**: - **High cost of spectroscopic redshift measurement**: Traditional spectroscopic redshift measurement relies on detailed spectral energy distribution templates (SEDs). Although these methods are effective, they are resource - intensive and costly, and are particularly inadequate when dealing with large - scale data sets. - **Limitations of spectroscopic observations**: For fainter celestial bodies, spectroscopic observations may not provide enough information, leading to measurement difficulties. 2. **Advantages of data - driven methods**: - **Efficiency and adaptability**: By using actual observational data (such as magnitudes in different bands) instead of theoretical templates, data - driven methods can process large - scale celestial data more efficiently and have better adaptability. - **Improved prediction accuracy**: Compared with the traditional decision - tree regression model, the fully - connected neural network (FCN) can capture complex relationships in the data, thereby improving the accuracy of redshift prediction. 3. **Specific research objectives**: - **Evaluating the reliability of FCN**: The research aims to evaluate the performance of FCN in estimating photometric redshift, especially when using data from the Sloan Digital Sky Survey (SDSS). - **Comparing the performance of different models**: By comparing the decision - tree regressor and FCN, evaluate the advantages and disadvantages of the two methods in redshift prediction, especially their root - mean - square error (RMS) and overall prediction accuracy. ### Formula representation The formulas involved in the paper include: - **Standard deviation calculation**: \[ \sigma = \sqrt{\frac{1}{N} \sum_{i = 1}^{N}(x_i-\mu)^2} \] where \(\mu\) is the average of all redshift values, \(N\) is the total number of redshift values, and \(x_i\) represents each redshift value. - **Mean - squared - error (MSE) loss function**: \[ L=\frac{1}{n} \sum_{i = 1}^{n}(y_i-\hat{y}_i)^2 \] where \(y_i\) is the true redshift value, \(\hat{y}_i\) is the predicted redshift value, and \(n\) is the number of celestial bodies in the training batch. By introducing data - driven methods, especially deep - learning techniques, this research hopes to provide a more efficient and accurate redshift - estimation tool for astronomical surveys, thereby promoting the understanding of galaxy formation and evolution.

Deep Learning Approach to Photometric Redshift Estimation

Photometric redshift estimation via deep learning

Deep learning based photometric redshifts for the Kilo-Degree Survey Bright Galaxy Sample

Photometric Redshift Analysis using Supervised Learning Algorithms and Deep Learning

Photometric Redshifts Probability Density Estimation from Recurrent Neural Networks in the DECam Local Volume Exploration Survey Data Release 2

Photometric Redshift Estimation for CSST Survey with LSTM Neural Networks

Photo-zSNthesis: Converting Type Ia Supernova Lightcurves to Redshift Estimates via Deep Learning

Photometric Analysis for Predicting Star Formation Rates in Large Galaxies Using Machine Learning and Deep Learning Techniques

Benchmarking and scalability of machine-learning methods for photometric redshift estimation

Using gamma regression for photometric redshifts of survey galaxies

Probability density estimation of photometric redshifts based on machine learning

Machine learning applications in astrophysics: Photometric redshift estimation

Photometric Redshift Estimation of Galaxies in the DESI Legacy Imaging Surveys

Estimation of redshift and associated uncertainty of Fermi/LAT extra-galactic sources with Deep Learning

Feature importance for machine learning redshifts applied to SDSS galaxies

Data-driven photometric redshift estimation from type Ia supernovae light curves

Photometric redshifts from SDSS images using a Convolutional Neural Network

Multimodality for improved CNN photometric redshifts

Photometric Redshifts for Cosmology: Improving Accuracy and Uncertainty Estimates Using Bayesian Neural Networks

AI-assisted Reconstruction of Cosmic Velocity Field from Redshift-Space Spatial Distribution of Halos