arXiv subjects

Kamlesh N Pathak

Publications and source records attributed to Kamlesh N Pathak.

3 recordsLinked to original sources

arXiv2021

Deep Learning for estimating parameters of Gravitational Waves

In recent years, improvements in Deep Learning (DL) techniques towards Gravitational Wave (GW) astronomy have led to a significant rise in the development of various classification algorithms that have been successfully employed to extract GWs of binary blackhole merger events from noisy time-series data. However, the success of these models is constrained by the length of time-sample and the class of GW source: binary blackhole and neutron star binaries to some extent. In this work, we intended to advance the boundaries of DL techniques using Convolutional Neural Networks, to go beyond binary classification and predict the physical parameters of the events. We aim to propose an alternative method that can be employed for realtime detection and parameter prediction. The DL model we present has been trained on 12s of data to predict the GW source parameters if detected. During training, the maximum accuracy attained was 90.93%, with a validation accuracy of 89.97%.

astro-ph.HE↗

arXiv2021

A machine learning-based approach towards the improvement of SNR of pulsar signals

Many pulsar folding algorithms are currently deployed to generate strong SNRs for the total intensity profiles. But they require large observation times to improve the SNR effectively. New approaches to de-noise the pulsar total intensity data have sprung up over the years, powered by Machine learning and Deep learning algorithms. In the current work, efforts are made to implement the currently proposed supervised machine learning models, such as ensembling techniques like Decision Tree Regressor, Random Forest Regressor, Adaboost Regressor, Gradient Boosting Regressor (GBR), K-Nearest Neighbours(KNN), and Support Vector Regressor (SVR) to find out the best possible algorithm which can work over a variety of pulsars from the EPN database of pulsars. All the data used in this work is extracted from the European Pulsar Network (EPN) database of pulsar profiles. The training dataset is obtained by post-processing the pulsar profile data from the EPN database hand testing is performed on a preselected portion of the original data. The results are obtained by testing the above algorithms for 10 different pulsars, including some historically significant ones, and the predicted profiles are plotted. We find that Gradient boosting regressor works the best in denoising pulsar data, followed closely by KNN regressor. This work also emphasizes that there is a reduction in the number of periods of folding by 35-40\% when a combination of machine learning models with the existing pulsar folding techniques like Fast Folding Algorithm(FFA) is employed, which in turn can further reduce the pulsar observation times for the telescopes hunting for pulsars today.

astro-ph.HE↗

arXiv2020

Predicting future astronomical events using deep learning

In a quest towards an intelligent decision-making machine, the ability to make plausible predictions is the central pillar of its intelligence. A predicting algorithm's central idea is to understand the governing physical rules and make plausible and apt predictions based on the same governing laws. Extending the study towards the astrophysical phenomenon puts the model's ability to test since the model has to understand various parameters that govern the dynamics of the event and understand the spatial and temporal evolution by applying the plausible laws. This work presents a deep learning model to predict plausible future events that maintain spatial and temporal coherence. We have trained over two broad classes, the evolution of Sa, Sb, S0, and Sd galaxy mergers and evolution of gravitational lenses with a higher redshift of the foreground galaxy having $15M_{\odot}$. We extended our work towards developing a direct measure of the performance metric for any prediction algorithm. We thereby introduce a novel metric, Correctness Factor (CF), which directly outputs how accurate a prediction is.

astro-ph.IM↗