MACHINE LEARNING APPLICATIONS FOR THE JCMT

MACHINE LEARNING APPLICATIONS in radio astronomy

created by RUPASHRI SATHASIVAM

ML APPLCATIONS IN RADIO ASTRONOMY

WHAT IS RADIO ASTRONOMY?

WHAT IS ML?

MY PROJECT

JCMT

contents

neural networks

artificial intelligence

Click on the +INFO Button to learn more about Deep Learning!

+ INFO

NEXT

BACK

Machine learning is a fascinating branch of artificial intelligence that empowers computers to learn from data and make predictions or decisions without explicit programming. It's like giving machines the ability to learn from examples and improve their performance over time. But wait, there's more! Deep learning takes this to the next level by using neural networks with multiple layers to process complex patterns and extract valuable insights from vast amounts of data.

WHAT IS MACHINE LEARNING (ML)?

Click on the +INFO Button to learn more about radio waves, radio astronomy and radio telescopes.

BACK

NEXT

+ INFO

Radio astronomy is the study of celestial objects and phenomena using radio waves instead of visible light, allowing scientists to observe and analyze cosmic objects such as stars, galaxies, and pulsars.Speaking of radio astronomy, did you know that the James Clerk Maxwell Telescope (JCMT) in Hawaii is one of the world's largest single-dish telescopes operating in the submillimeter wavelength range, providing valuable insights into the birth and evolution of stars?

WHAT IS RADIO ASTRONOMY ?

machines

support vector

Generative Adversarial Networks

Convolutional Neural Networks

MACHINE LEARNING APPLICATIONS IN RADIO ASTRONOMY

+ INFO

+ INFO

+ INFO

Identifying and categorizing celestial objects in radio observations is a fundamental task in astronomy. Machine learning techniques, such as support vector machines (SVMs), random forests, and deep learning models, can automate the process of source detection and classification. These algorithms can learn patterns and features from labeled data and accurately identify different types of galaxies.

Radio telescopes capture vast amounts of data, often in the form of radio interferometric measurements. ML algorithms can be employed to reconstruct high-resolution images from these measurements. Convolutional Neural Networks (CNNs) and Generative Adversarial Networks (GANs) have been used to enhance the quality and resolution of images obtained from radio interferometry.

Redshift is a crucial parameter in cosmology that measures the expansion of the universe. Machine learning algorithms can predict redshift values by learning from the spectral features in radio observations. By training on a large set of labeled data, these algorithms can estimate redshifts accurately and at a much faster rate than traditional methods.

GALAXY CLASSIFICATION

IMAGE RECONSTRUCTION

REDSHIFT ESTIMATION

NEXT

BACK

MORE

+ INFO

Dark matter is an invisible form of matter that does not emit or interact with light. Mapping its distribution in the universe is a challenging task. Machine learning techniques can help analyze large-scale structures in the cosmic microwave background radiation or galaxy surveys and infer the distribution of dark matter. These algorithms can learn complex patterns and correlations in the data, providing insights into the nature and properties of dark matter.

DARK MATTER MAPPING

MACHINE LEARNING APPLICATIONS IN RADIO ASTRONOMY

NEXT

BACK

NEXT

Click on the +INFO Button to learn more about the JCMT!

BACK

+ INFO

The JCMT, or James Clerk Maxwell Telescope, is a powerful astronomical observatory located in Hawaii that studies the universe in the submillimeter wavelength range, helping us understand the formation of stars and galaxies.While machine learning hasn't been extensively utilized at the JCMT yet, it has the potential to revolutionize data analysis and improve our understanding of the cosmos.

JAMES CLERK MAXWELL TELESCOPE (JCMT)

DATA COLLECTION

STARLINK

CNN

BACK

In my project, I am developing and training a Convolutional Neural Network (CNN) algorithm using data from the JCMT to distinguish between regions in space where stars are actively forming and regions where star formation is not occurring.This application of machine learning has the potential to enhance our understanding of stellar evolution and the processes shaping the universe around us.

MY PROJECT: "STAR FORMING REGION CLASSIFICATION: a machine learning apporach"

protostars

Canadian Astronomy Data Center

Click on the +INFO Button to access the data from CADC.

NEXT

+ INFO

BACK

The data set used in this study was specifically taken from the JCMT Transient Survey which can found on the Canadian Astronomy Data Center (CADC).The purpose of this survey was to collect measurements for accretion variability of protostars from eight nearby star forming regions (NGC 1333, IC 348, OMC 2/3, NGC 2024, NGC 2071, Ophiucus Core, Serpens Main, Serpens South).For my project I used the radio imaging from this survey.

DATA COLLECTION

East Asian Observatory

Click on the +INFO Button to get to the website where the tutorials are hosted. If you need any help with installation just click on HELP

Click on the DOWNLOAD Button to get to the website where the software is hosted.

HELP

+ INFO

NEXT

DOWNLOAD

BACK

Starlink is the open source software package used to process the raw data into usable data for projects.The software is currently being maintained by the East Asian Observatory (EAO).Do note that the software is only available for MacOs and Linux so there will be a steep learning curve especially on Linux. But don't worry, there are tutorials on how to get started with the software.

STARLINK

segmentation

object recognition

NEXT

Click on the +INFO Button to learn more about CNNs through a simple google colab tutorial.

BACK

+ INFO

The capacity of Convolutional Neural Networks (CNNs), a class of deep learning algorithms, to automatically extract hierarchical patterns and features from images, makes them a popular choice for computer vision tasks. A CNN algorithm was utilised in this project since CNNs have shown remarkable performance in applications including object recognition, image classification, and segmentation.

CNN

NEXT

BACK

In conclusion, this project has been a great success. I have learned so much about machine learning and radio astronomy and would love to learn more.Thank you for taking the time to read through this infographic about my experience dealing with radio astronomy and machine learning. I hope you have learned something through this infographic and put your new found knowledge into creating your own projects .

CONCLUSION

NEXT

This infographic has been created as a component of my final year project, conducted under the academic purview of the School of Physics, Universiti Sains Malaysia (USM), under the supervision of Dr John Soo Yue Han. All rights, including intellectual property rights and ownership, pertaining to this infographic are held exclusively by USM. The information, data, and visuals presented in this infographic are intended for educational and informational purposes only. They do not represent an endorsement, opinion, or official stance of USM.

acknowledgement

START OVER

Thanks!

Random Forests

Neighbours

k-Nearest

The paper linked tries to evaluate techniques for estimating redshifts of galaxies from a radio-selected survey. The techniques are k-Nearest Neighbours (kNN) and Random Forests. They were not be able to predict a continuous value for high-redshift radio sources, but a majority of them can be identified.

Based on the shift in an object's spectral lines brought on by the expansion of the universe, redshift is a measurement of the distance to a celestial object. The evolution and distribution of galaxies in the cosmos can only be studied with accurate redshift measurements. Based on a galaxy's observed features, such as its spectrum or image, machine learning algorithms can be used to forecast the redshift of the galaxy. Spectra or images of galaxies along with the galaxies' known redshifts usually make up the data for this project.

Click here to see a related study on the subject

encoder

COSMOS

auto-

The paper below reproduced the JCMT SCUBA-2 images from Herschel SPIRE 500 μm COSMOS data with high fidelity using an auto-encoder.

Click here to see a related study on the subject

High-resolution images of the radio sky can be created using radio interferometers, which can detect the radio signals coming from various antennas and combine them. However, this procedure can be computationally costly and result in incomplete measurements that leave out important information. The quality of the image can be increased by using machine learning techniques to reconstruct missing data. Machine learning can also be used to recreate images from different radio telescopes.The interferometric measurements used for this project usually take the form of visibilities or UV data, where each measurement reflects the correlation between the signals received by two antennas.

Cosmicflows-3

EAGLE simulation

The paper linked shows that we can reconstruct the cosmic web from the galaxy distribution using the convolutional-neural-network-based deep-learning algorithm. The mapping was confirmed by applying it to the EAGLE simulation. Finally, using the local galaxy sample from Cosmicflows-3, the dark matter map in the local universe was found

An important portion of the universe's mass is believed to be made up of dark matter, an invisible substance.It is only detectable by its gravitational effects on visible matter because it neither emits nor absorbs light. Based on the gravitational lensing impacts of dark matter on visible galaxies, machine learning algorithms can be used to map the distribution of dark matter in the universe.

Click here to see a related study on the subject

Riley

Fanaroff–

The paper linked focused on the morphological-based classification of radio galaxies known as Fanaroff–Riley (FR) type I and type II via supervised machine-learning approaches.

Click here to see a related study on the subject

Galaxies come in a variety of forms and sizes, and knowledge about their morphology and formation and evolution can be very useful. Astronomers can better comprehend the characteristics and behaviour of galaxies by classifying them based on their morphology using machine learning algorithms. Images of galaxies along with descriptions of each galaxy usually make up the data for this project.