Machine Learning for Astronomical Source Classification

Exploring Diverse Machine Learning Techniques for Astronomical Source Classification Using Images and Photometric Features.

For any queries, please contact at srinadhml99@gmail.com Arxiv link: https://arxiv.org/abs/2408.13634

Classical ML Algorithms

We studied several classical ML algorithms for classification between Star-Galaxy and Star-Galaxy-Quasar using photometric features alone. The files can be found at MLAlgos.

Table 1: Classical ML Baselines (DT, RF, GBDT) Performance

Experiment	Classification Task	Model	Accuracy (%)	Precision (%)	Recall (%)
Experiment 1	Star-Galaxy	DT
		RF
		GBDT
Compact Train/Test	Star-Galaxy-Quasar	DT
		RF
		GBDT
Experiment 2	Star-Galaxy	DT
		RF
		GBDT
Faint+Compact Train/Test	Star-Galaxy-Quasar	DT
		RF
		GBDT
Experiment 3	Star-Galaxy	DT
		RF
		GBDT
Compact Train, Faint+Compact Test	Star-Galaxy-Quasar	DT
		RF
		GBDT

MargNet

Unofficial PyTorch implementation of the paper, Photometric identification of compact galaxies, stars and quasars using multiple neural networks. The official implementation in TensorFlow can be found at here.

MargFormer

This part of the work (studied as MM ViT) was published in the Journal of Astrophysics and Space Science. The arxiv pre-print is here.

Bayesian MargNet and Bayesian MargFormer

Bayesian Neural Networks are also studied with MargNet and MargFormer. For reference, you can have a look at the file Ex1_SG_BayesianMargNet.ipynb.

More details will be shared soon.

Contrastive Learning (CL)

Self-Supervised (SimCLR)

Supervised

For initial experiments please check: https://github.com/srinadh99/Contrastive-Learning-for-Astronomy-

More details and codes will be shared soon.

Conformal Predictions

For initial experiments, please check, https://github.com/srinadh99/AstroFormer/tree/main/MargFormerCP