TY - GEN
T1 - An Energy-Efficient Deep Neural Network Model for Photometric Redshift Estimation
AU - Shreevershith, K.
AU - Sen, Snigdha
AU - Roopesh, G. B.
N1 - Publisher Copyright:
© 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.
PY - 2023
Y1 - 2023
N2 - In cosmological applications, redshift is a distance-measuring metric, and machine learning approaches have produced amazing results in this domain. Deep learning is a promising solution in redshift prediction due to a big set of cosmic data. An artificial neural network (ANN) model was implemented in this work with redshift values in the low (0–3) and high (0–7) ranges, where the data was extremely unbalanced and skewed in the high range. The true redshift value is translated into the logarithmic domain to account for the skewness in the redshift range distribution. Because redshift prediction is a regression job, the metrics used to evaluate model performance include mean absolute error (MAE), mean squared error (MSE), and R2. Furthermore, our experiment shows that by using a limited number of hidden layers, training time and carbon emissions can be minimized while still achieving sufficient performance.
AB - In cosmological applications, redshift is a distance-measuring metric, and machine learning approaches have produced amazing results in this domain. Deep learning is a promising solution in redshift prediction due to a big set of cosmic data. An artificial neural network (ANN) model was implemented in this work with redshift values in the low (0–3) and high (0–7) ranges, where the data was extremely unbalanced and skewed in the high range. The true redshift value is translated into the logarithmic domain to account for the skewness in the redshift range distribution. Because redshift prediction is a regression job, the metrics used to evaluate model performance include mean absolute error (MAE), mean squared error (MSE), and R2. Furthermore, our experiment shows that by using a limited number of hidden layers, training time and carbon emissions can be minimized while still achieving sufficient performance.
UR - https://www.scopus.com/pages/publications/85161470380
UR - https://www.scopus.com/inward/citedby.url?scp=85161470380&partnerID=8YFLogxK
U2 - 10.1007/978-981-19-7455-7_24
DO - 10.1007/978-981-19-7455-7_24
M3 - Conference contribution
AN - SCOPUS:85161470380
SN - 9789811974540
T3 - Lecture Notes in Networks and Systems
SP - 319
EP - 330
BT - Innovations in Computer Science and Engineering - Proceedings of the 10th ICICSE, 2022
A2 - Saini, H. S.
A2 - Sayal, Rishi
A2 - Govardhan, A.
A2 - Buyya, Rajkumar
PB - Springer Science and Business Media Deutschland GmbH
T2 - 10th International Conference on Innovations in Computer Science and Engineering, ICICSE 2022
Y2 - 16 September 2022 through 17 September 2022
ER -