Improving Photometric Redshifts By Varying Activation Functions In Artificial Neural Networks
In recent years, the astronomical community has faced a data deluge of up to exabytes due to the advancement of telescope technology. In fact, the application of machine learning techniques has simplified the task of analysing data for astronomers. Photometric redshift estimation or photo-z is on...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | en |
| Published: |
2024
|
| Subjects: | |
| Online Access: | http://eprints.usm.my/63701/1/24%20Pages%20from%20IMDAD%20BINTI%20MAHMUD%20PATHI.pdf http://eprints.usm.my/63701/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In recent years, the astronomical community has faced a data deluge of up to exabytes
due to the advancement of telescope technology. In fact, the application of machine
learning techniques has simplified the task of analysing data for astronomers. Photometric
redshift estimation or photo-z is one of the relevant applications of machine learning
and statistical methods in cosmology. Due to its significant contribution to cosmology,
there is an increasing demand for precise photo-z for forthcoming astronomical surveys,
including the Euclid mission and LSST. The accuracy and performance of the photo-z
algorithm have been improved by adopting and modifying machine learning hyperparameters.
The Artificial Neural Network Redshift (annz) algorithm is a fast and simple
machine learning photometric redshift estimator. One of the hyperparameters in the
artificial neural network (ANN) is the activation function. It acts as a decision-making
unit, which introduces non-linearity into the model, leading to better differentiation
capabilities and potentially boosting the ANN’s performance if optimally tuned. We
test the performance of annz by varying the activation functions, replacing the original
logistic sigmoid with tanh, Softplus, SiLU, ReLU, Leaky ReLU and Mish. The training
is demonstrated on the Luminous Red Galaxy (LRG) sample of the Sloan Digital Sky
Survey (SDSS), Stripe-82 survey, and Physics of the Accelerating Universe Survey
(PAUS). We also tested the performances of these activation functions by varying the
depth and width of the ANN architectures. |
|---|