AR-2019-2020

RESEARCH BY VISITING ASSOCIATES Sheelu Abraham Application of convolutional neural networks for stellar spectral classification Due to the ever-expanding volume of observed spectroscopic data from surveys such as SDSS and LAMOST, it has become important to apply artifi- cial intelligence (AI) techniques for analysing stel- lar spectra to solve spectral classification and re- gression problems like the determination of stellar atmospheric parameters T eff , logg, and [Fe/H]. We propose an automated approach for the classifica- tion of stellar spectra in the optical region using Convolutional Neural Networks. Traditional ma- chine learning (ML) methods with “shallow” archi- tecture (usually up to 2 hidden layers) have been trained for these purposes in the past. However, deep learning methods with a larger number of hidden layers allow the use of finer details in the spectrum which results in improved accuracy and better generalisation. Studying finer spectral sig- natures also enables us to determine accurate dif- ferential stellar parameters and find rare objects. We examine various machine and deep learning al- gorithms like Artificial Neural Networks (ANN), Random Forest (RF), and Convolutional Neural Network (CNN) to classify stellar spectra using the Jacoby Atlas, ELODIE and MILES spectral libraries as training samples. We test the perfor- mance of the trained networks on the Indo-U.S. Library of Coud´e Feed Stellar Spectra (CFLIB). We show that using convolutional neural networks, we are able to lower the error up to 1.23 spectral sub-classes as compared to that of 2 sub-classes achieved in the past studies with ML approach. We further apply the trained model to classify stel- lar spectra retrieved from the SDSS database with SNR > 20. This work has been done in collaboration with Kaushal Sharma, Ajit Kembhavi, Anirudha Kembhavi, T. Sivarani and Kaustubh Vaghmare. Detecting outliers in SDSS using convolutional neural network We propose an automated algorithm based on Con- volutional Neural Network (CNN) for the detection of peculiar objects in large databases using their spectral observations. A convolutional neural net- work is a class of deep-learning algorithms which al- lows the detection of significant features/patterns in sequential data like images, audio, time-series etc. by applying convolutional neurons (kernels) along the sequence. For detecting unusual spec- tra, we use eight-layer deep convolutional network with autoencoder architecture on ∼ 60,000 spec- tra collected from the Sloan Digital Sky Survey. The training of the network is done in an unsuper- vised manner. We show that the trained network is able to retrieve the spectra of rare objects from a large collection of spectra. Such algorithms can easily be rescaled to other surveys and therefore can serve as a potential component of the data re- duction pipelines for automatically detecting spec- tra with unusual features and recovering defective spectra. This work has been done in collaboration with Kaushal Sharma, Ajit Kembhavi, Anirudha Kembhavi, T. Sivarani Dharam Vir Ahluwalia Elko under spatial rotation Under a rotation by an angle ϑ , both the right- and left- handed, Weyl spinors pick up a phase factor exp ( ± iϑ/ 2). The upper sign holds for the posi- tive helicity spinors, while the lower sign for the negative helicity spinors. For ϑ 2 π radians this pro- duces the famous minus sign. However, the four- component spinors are built from a direct sum of the indicated two-component spinors. The effect of the rotation by θ = 2 π radians on the eigen- spinors of the parity - that is, the Dirac spinors – is the same as on Weyl spinors. It is because for these spinors the right- and left- transforming components have the same helicity and the rota- tion induced phases, being same, factor out. But for the eigenspinors of the charge conjugation op- erator, i.e., Elko, the left- and right- transform- ing components have opposite helicities, and there- fore, they pick up opposite phases. As a conse- quence, the behaviour of the eigenspinors of the charge conjugation operator (Elko) is more subtle: for 0 < ϑ < 2 π , a self conjugate spinor becomes a linear combination of the self and antiself conjugate spinors with θ dependent superposition coefficients - and yet the rotation preserves the self/antiself conjugacy of these spinors! This apparently para- doxical situation is fully resolved. This new effect, to the best of our knowledge, has never been re- ported before. The purpose of this communication is to present this result and to correct an interpre-