Cover page

The simultaneous emergence of Orthetrum chrysis (reddish) and Zyxomma petiolatum (greenish-yellow) was

recorded at airis4D biosciences lab. The bioscience director Geetha Paul Jacob has documented the emergence

of over two dozen species in the lab. This photograph from her collections illustrates the miracle of life on

our planet. A full-ﬂedged dragonﬂy with a completely diﬀerent morphology and much bigger size than its tiny

nymph unfold out of its exuvial shell about an hour past midnight.

Managing Editor Advisory Board Chief Editor Correspondence

Ninan Sajeeth Philip K Babu Joseph Abraham Mulamootil The Chief Editor

Geetha Paul airis4D

Arun Kumar Aniyan Thelliyoor - 689544

India

Jorunal Publisher Details

Publisher : airis4D, Thelliyoor 689544, India

Website : www.airis4d.com

Email : nsp@airis4d.com

Phone : +919497552476

Contents

I Artiﬁcial Intelligence and Machine Learning 1

1 Connecting the Data Dots with Graphs 2

1.1 Prologue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.2 Graphs for Tech . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.3 Graphs for Science . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.4 Graphs for Finance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.5 Graphs for Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.6 Epilogue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 History of learning: From how to what 8

2.1 Data drive - learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.2 From numbers to image and language . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.3 Deep Learning Architectures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.4 Machine Learning in day-to-day life . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.5 The Future . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3 Machines that can understand Human Languages 13

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3.2 History of Natural Language Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

3.3 Challenges of NLP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

II Astronomy and Astrophysics 17

1 A sneak peak into the science behind telescopes 18

1.1 Importance of Aperture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

1.2 Focal length, Field of View and Image magniﬁcation . . . . . . . . . . . . . . . . . . . . . . 20

2 Evolution of Satellite Imaging 23

2.1 Landsat mission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

2.2 Sentinel Mission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3 Eclipsing Binaries 27

3.1 Light Curve of Eclipsing Binary Stars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3.2 Types of eclipsing binary systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

III Biosciences 32

1 Jewels of air- A Dragonﬂy story 33

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

1.2 Story of a Dragonﬂy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

CONTENTS

1.3 History & Science behind the story of a dragonﬂy. . . . . . . . . . . . . . . . . . . . . . . . . 34

1.4 Dragonﬂy Metamorphism - Hemimetabolous insects . . . . . . . . . . . . . . . . . . . . . . 35

1.5 Common Dragonﬂy species . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

1.6 Dragonﬂies as Ecological indicators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

1.7 Conservation goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

2 Healthcare developments during the Covid era 39

IV Computer Programming 41

1 Mastering R: A Challenge for Beginners 42

V Fiction 47

1 Curves and Curved Trajectories 48

1.1 Episode 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

1.2 Episode 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

iii

Part I

Artiﬁcial Intelligence and Machine Learning

Connecting the Data Dots with Graphs

by Arun Aniyan

airis4D, Vol.1, No.1, 2023

www.airis4d.com

1.1 Prologue

Imagine you buy a house in a part of the town where houses are on either side of a central street. You arrive

there and see blue ﬂags in front of each house on the left side of the street and green ﬂags on the houses on the

right side. You happen to have no ﬂag and the house is on the left side. By identifying the pattern of the ﬂag

colours on either side of the street you quickly deduce that your house must have a blue ﬂag. You have simply

mimicked one of the oldest data science algorithms called the “nearest neighbor” algorithm. The fundamental

idea is ﬁnding patterns through relations. In this case, you looked at a pattern which is the relation between the

side of the street and colour of the ﬂag. Such relations give context or more background to the observations one

makes.

Data is generally deﬁned as individual forms of facts or items of information. For example, the “age” of a

person as 46 is a data point. Collection of such data points form “information”. The following table shows a

collection of data points forming a piece of information.

Name Age Gender

Tim 46 M

Martha 40 F

Jim 18 M

Janet 16 F

Dorothy 78 F

Margaret 57 F

Table 1.1: Information on the Name, Age, and Gender of six people

Table 1.1 gives useful information on the Name, Ages, and Gender of six people. One can infer very

simple statistics and information about the group of people mentioned here. Now let us add another piece of

data which is the “relation” between the diﬀerent people in the table. This requires us to move away from the

tabular representation of the data because there may be multiple connected relations between each individual.

We will use more of a graphical representation of data points to represent two types of relations in this case.

Figure 1.1 shows the new “relational” representation of Table 1.1. This new representation adds a completely

new dimension and story to the tabular data in Table 1. Most importantly this form of relational representation

1.2 Graphs for Tech

Figure 1.1: Graph representation of data points

adds “context” to data. This context enables us to bring in a completely new narrative to the observations made.

Apart from the explicitly mentioned relations, we can uncover other relations. For example, we can add a new

relation called “sibling” between Janet and Jim, which gives us another set of new information. We also ﬁnd that

Margaret is a stranger at this point with respect to others. The representation shown in Figure 1.1. is generally

called “Graph Representation” and such data on a graph is called “Graph Data”. In graph data, there are mainly

two components, namely “node” and “relation”. An entity also known as a “vertex”, are individual entities that

are connected by one or more “relations”. Relations also referred to as “edges”, can be either unidirectional or

bi-directional between two nodes. Both nodes and edges can have one or more properties. The idea of using

graphs to uncover hidden relations is not new. You may have seen movies like Sherlock Holmes, where he put

pictures of individuals, places, and objects and connects them with strings to solve hard investigations. This

kind of representation help to make serendipitous discoveries that are not obvious in the ﬁrst place.

1.2 Graphs for Tech

Today we all interact with the application of machine learning and data science on a daily basis. In many of

those applications, graphs do play a major role. One such application is “recommendation systems”. All those

product recommendations that we get on Amazon, the new post recommendations on Facebook, are generated

using recommendation systems. Recommendation systems mostly work on the principle of the nearest neighbor

algorithm using algorithms such as collaborative ﬁltering. With the advent of graph-based algorithms, they have

become much more eﬃcient and accurate in terms of giving us recommendations. Let us look at an example

of movie recommendations used in applications such as Netﬂix. First, a graph database with all the movies to

date is created. The nodes will be entities such as movie, actor, director, etc, and relations will be “acted in”,

“directed” etc. A sample model of such a graph is shown in Figure 1.2. This is also termed graph schema which

is a general representation of the database.

With the graph model shown in Figure 1.2., it is possible to answer questions such as the following:

1. Which person acted or directed a movie?

2. Which movie(s) was directed by person X?

1.2 Graphs for Tech

Figure 1.2: Graph model for movie recommendations

3. What are the genres of movies in which person X acted?

Moreover, the basic nodes and relations, each of the nodes are enriched with additional properties such as “year

of release”, “country of origin” etc. to add more context to the graph. Such metadata enriches the graph to

get additional insights. Now let us take an example where a user who is a fan of Tom Hanks, wants to get

recommendations for his movies which has an IMDB rating greater than 8. A query made on the graph asking

“movies acted in by Tom Hanks with a rating greater than 8” yields the graph shown in Figure 1.3.

Figure 1.3: Movies acted in by Tom Hanks which has IMDB rating greater than 8

The nodes in Figure 1.3 are the best recommendations generated for the user with speciﬁc preferences.

They can now be sorted based on the individual rating and presented to the user. Like recommendation systems,

graphs are widely used in social network analysis. Graphs in social networks help to identify inﬂuential

individuals that advertising agents can target and recommend new connections. The suggestions one gets on

1.3 Graphs for Science

Facebook and LinkedIn are generated through graph algorithms. In those cases, a graph is connected with

the list of our friends and their friends. A new possible friend recommendation is generated by looking at the

similarities between us and the friend of a friend. For example similarity in terms of location, education, or

even personal likes. This is a common application all of us come across in our daily life.

1.3 Graphs for Science

There are even complex applications of using graph representations of data, for example in medicine. One

of the major breakthroughs that graph methods have provided to science and speciﬁcally to chemistry is the

ability to represent chemical elements and molecules as graphs. Molecules are basically set of elements or

compounds which are connected by bonds. Therefore it is very natural to represent them in graph form. This

has facilitated new computational methods that enabled the discovery of novel medicines and treatments.

One of the most recent applications of graphs in medicine is discovering new drugs for diseases. Some

of the new cancer medications were discovered using graph methods. Graphs have also enabled the rapid

discovery of Covid-19 vaccines. Designing vaccines usually takes a period of months. But with the application

of advanced graphical methods as opposed to conventional methods, we were able to discover vaccines for

Covid-19 in a matter of months. This is a great achievement that has a powerful impact on society.

1.4 Graphs for Finance

Another interesting and practical application of graph data is in ﬁnance and ﬁnancial crime investigations.

In the ﬁnancial sector, one of the major applications is detecting money laundering and banking fraud. Money

laundering when represented on a graph has a speciﬁc circular pattern. So, what banks and other ﬁnancial

institutions generally do is take the transactional data for speciﬁc periods and represent them on a graph, where

each node is an account and the relations are the direction of money ﬂow. Figure 1.4. shows a circular pattern in

which the cash ﬂow starts from one node and ends on the same node after hopping through many other accounts.

Figure 1.4: Example pattern of money laundering represented on a graph (Image Courtesy: sctr7.com)

1.5 Graphs for Machine Learning

If you closely observe all the diﬀerent nodes are interconnected and all the relations ﬁnally end on node

13747. If such a pattern is detected from a series of transactions then it is an indicator of money laundering.

1.5 Graphs for Machine Learning

Machine learning applications can be replicated and also be augmented with graphs. For example, if

we have a dataset similar to Table 1. for classiﬁcation or regression purposes, the model performance can be

augmented with features extracted from a graphical representation of the same data. Graph Neural Networks

are a byproduct of such a methodology. They are extremely useful applications in which the relation between

distinct data points is important. Similar models are used in the applications mentioned for drug discovery and

chemical applications. It is also possible to replicate classiﬁcation and even clustering algorithms with graphs.

Figure 1.5 shows a large graph with many diﬀerent nodes which have diﬀerent types of connections.

Figure 1.5: Large graph from the movie recommendation data

Simply looking at the colour of the nodes, it is possible to observe patterns in which diﬀerent nodes group

or cluster together. We can perform classiﬁcation or clustering on the graph itself using both node and relation

properties. It is also possible to predict if there is a relation between two nodes that are not connected to each

other in the ﬁrst place. This is a powerful method that is also used to make discoveries that are serendipitous in

nature. For example, discovering a new property of chemical compounds or a relation between two people who

are at the two ends of a large graph.

1.6 Epilogue

All the above examples show both the advantages and diﬀerent applications of a graph representation of

data. Graphs and Graphical methods are extremely powerful such that they are able to enhance the process of

1.6 Epilogue

doing science and make interesting discoveries. There is a tremendous amount of research and development

taking place both in the academic and commercial space for graph-based technologies. One of the most exciting

topics in this area is called ”Knowledge Graphs” which is enhancing artiﬁcial intelligence development and is

gaining great traction. There are still many areas of industry and even applied research where graphical methods

are not used but may provide a powerful utility.

References

Introduction to Graph Data Science, P. Srinivasan, Analytics Vidhya, August 3, 2022

How to build a knowledge graph, Stardog Blog, Feb 23, 2022

The Basics of Data Modeling, B. M. Sasaki, July 24, 2018

About the Author

Dr.Arun Aniyan is leading the R&D for Artiﬁcial intelligence at DeepAlert Ltd,UK. He comes from

an academic background and has experience in designing machine learning products for diﬀerent domains. His

major interest is knowledge representation and computer vision.

History of learning: From how to what

by Blesson George

airis4D, Vol.1, No.1, 2023

www.airis4d.com

Cognitive development is one of the distinctive characteristics of humans. Although the ability to learn

exists in all creatures, humans have been perfectly blessed with it. From the moment of birth, until our mind is

conscious, we keep learning. What does a young infant learn? Babies learn how to adapt to speciﬁc individuals

or objects that are connected to certain stimuli, which can be both negative or positive. Humans have the

predisposition to learn. We are curious about the environment we live in and our enormous human brains are

primed to learn.

The period known as the ”industrial revolution” can be described as the time when individuals began

to work with machines to perform their physical chores more eﬀectively. Men’s labour was aided by wheels,

manufacturing equipment, and military weapons. They created machines, and the machines slavishly reproduced

what was supplied to them. Later, people started employing computers and other intelligent systems to assist

them in doing computations and other intellectual work as the information revolution developed. In all of these

endeavours, machines merely followed instructions like slaves. They were provided instructions on how to

complete the duties in great detail. Later, the man began to reﬂect once more about this. Why are machines

unable to learn like people do instead of simply repeating duties like a slave? When exposed to their environment

and experiences, humans learned from them. They are picking up knowledge from their environment. Artiﬁcial

intelligence began to take shape when the idea of imitating the human brain emerged. Or we may refer to it as

learning driven by data.

2.1 Data drive - learning

According to Merriam-Webster’s dictionary, Machine Learning is deﬁned as “the process by which a

computer is able to improve its own performance (as in analyzing image ﬁles) by continuously incorporating

new data into an existing statistical model”. Before we proceed into further details, let us look at the diﬀerence

between conventional computer programming and machine learning. In conventional programming, a human

manually constructs the code and establishes the rules. No reasoning is being applied by the programme on

its own. On the basis of the rules speciﬁed in the programme, it generates the desired output from any input

data. In contrast, we just provide the input and output data for machine learning. There are no ﬁxed rules for

producing the result. The rules are automatically generated by the computer using the data. The invention of

data-driven models may be traced back to 1813 when Sir Thomas Bayes created the ground-breaking Bayes

Theorem. Warren McCulloch and Walter Pitts, two American physicists, created the mathematical model that

mimics the biological neuron later in 1943. The introduction of Perceptrons by Frank Rosenblatt, Nearest

2.2 From numbers to image and language

Figure 2.1: Illustration of traditional computer programming and machine learning. Image Courtesy:

https://futurice.com/blog/diﬀerences-between-machine-learning-and-software-engineering

neighbour algorithms, recurrent neural networks, Bayesian classiﬁers, and backpropagation algorithms laid the

groundwork for modern machine learning approaches. The development of Convolutional Neural Networks

(CNN), Boosting Algorithm, Gradient Descent Method, ImageNet, and Deep Belief Networks were signiﬁcant

steps in the growth of machine learning. Amazingly, all of these discoveries and formulations were developed

in less than four decades. IBM Watson, Google Brain (2011), DeepFace(2014), Platforms and APIs for machine

learning, AlphaGO (2016, Waymo (2017), and AlphaFold(2018) are just a few of the applications that have

revolutionised the lives of humans in the twenty-ﬁrst century (2018),

2.2 From numbers to image and language

Machine learning is not limited to processing numerical data. Its realm of application grew to include

two other important human domains: language and vision. These technologies are called language processing

and image processing. The advancement of machine learning in language processing began in the 1950s.

IBM researchers utilised a rudimentary version of machine learning to enhance the IBM 701’s capacity to

comprehend natural language commands. In the 1990s, Natural Language Processing (NLP) saw tremendous

advancements due to the widespread availability of massive volumes of digitised text data and the development

of new machine learning methods. Currently, machine learning is an integral component of NLP. Many of

the most modern NLP systems, such as those used for machine translation, signiﬁcantly rely on machine

learning to increase their accuracy and performance. Computer vision is a branch of artiﬁcial intelligence that

focuses on training computers to understand and interpret visual data. In 1974, optical character recognition

(OCR) was created to assist the decipherment of text written in a variety of typefaces. In the 1980s, Japanese

neurologist Dr. Kunihiko Fukushima developed Neocognitron, a hierarchical, multilayered neural network

capable of recognising complicated visual patterns such as corners, curves, edges, and basic forms. Later,

LeCun introduced Convolutional Neurel Networks (CNN) which is widely used for image/object recognition

and classiﬁcation. A growth in object recognition research in 2000–2001 paved the foundation for the ﬁrst real-

time facial recognition algorithm. It turned out to be a groundbreaking concept that altered the entire landscape

of computer vision. Several advanced computer vision technologies, such as facial recognition, medical picture

analysis, semantic segmentation, etc., are anticipated to be utilised extensively in numerous industries, ranging

2.3 Deep Learning Architectures

from banking to retail.

2.3 Deep Learning Architectures

Deep Learning is a cutting-edge technique based on neural networks that attempt to mimic the functioning

of the human cortex. The architecture of deep learning comprises of neural networks with diverse topologies.

In general, neural networks are composed of numerous data-processing layers, including an input layer (raw

data), hidden layers (which analyse and aggregate input data), and an output layer (it produces the outcome:

result, estimation, forecast, etc.). In the next part, we will explore some of the most common and eﬃcient deep

learning architectures.

RNN: Recurrent Neural Networks:

RNN is one of the foundational network designs that other deep learning architectures are derived from.

RNNs consist of a variety of architectures for deep learning. They can utilise their internal state (memory)

to handle input sequences of varied length. RNNs possess a memory. Every piece of processed data

is collected, saved, and utilised to compute the ultimate result. RNNs are extremely eﬀective in sectors

where the order of delivered information is crucial. Common applications include natural language

processing (chatbots), speech synthesis, and machine translations.

LSTM: Long Short-Term Memory:

LSTM is another sort of RNN. It has feedback connections. This means that not only single data points

(such as photos) but also complete data sequences may be processed. The LSTM algorithm is derived

from neural network topologies and is based on the notion of a memory cell. The memory cell may keep

its value for a short or long period of time as a consequence of its inputs, allowing it to remember more

than just its most recently computed value.

A cell, an input gate, an output gate, and a forget gate make up a standard LSTM design. These three

gates govern the ﬂow of information into and out of the cell, which may remember values across arbitrary

time intervals.

CNN: Convolutional Neural Networks

CNN architecture is frequently employed for image processing, image recognition, video analysis, and

natural language processing. CNN is able to take in an image, give priority to various aspects/objects

inside the image, and discriminate between them. The term ’convolutional’ is derived from a mathematical

technique that involves the convolution of many functions. CNNs include an input layer, an output layer,

and a number of hidden layers. Typically, CNN’s hidden layers consist of a sequence of convolutional

layers.

SOM: Self Organizing Maps

In 1982, Dr. Teuvo Kohonen devised the self-organized map (SOM), also known as the Kohonen map.

SOM is an unsupervised neural network that reduces the dimensionality of the input data set in order to

form clusters. SOMs diﬀer from conventional artiﬁcial neural networks in several ways.

DBN: Deep Belief Network

DBN is a multilayer network (usually deep, with several hidden layers) where each pair of linked layers is

a Restricted Boltzmann Machine (RBM). Consequently, we may assert that DBN is a collection of RBMs.

DBN is constructed of many layers of latent variables (also known as ”hidden units”), with linkages

between levels but not between units within each layer.

Autoencoder

Autoencoder is a sort of neural network in which the output layer has the same number of dimensions as

2.4 Machine Learning in day-to-day life

the input layer. Autoencoders are also known as replicating neural networks since they reproduce data

from input to output without supervision. Three components make up an autoencoder:

1. Encoder: An encoder is a feedforward, fully connected neural network that encodes the input

image in a reduced dimension by compressing it into a latent space representation. The distorted

compressed version of the original image.

2. Bottleneck or Code: This component of the network represents the input that is supplied to the

decoder.

3. The Decoder is also a feedforward network identical to the encoder and has the same architecture.

This network is responsible for reconstructing the input’s original dimensions from the code.

GAN: Generative Adversarial Network

Generative adversarial networks (GANs) are an innovative development in machine learning. GANs

are generative models; they produce new instances of data that resemble the training data. GANs may

generate pictures that resemble photos of human faces, despite the fact that the faces do not belong to

actual people. GANs accomplish this degree of realism by partnering a generator that learns to create the

desired output with a discriminator that learns to separate actual data from the output of the generator. The

generator attempts to deceive the discriminator, while the discriminator attempts to avoid being deceived.

2.4 Machine Learning in day-to-day life

Machine learning is increasingly being used in many aspects of daily life. Some examples of how machine

learning is used in day-to-day life include the following:

Personal assistants like Siri, Alexa, and Google Assistant use machine learning to understand and respond

to natural language voice commands.

Recommender systems on websites and apps, such as those used by Netﬂix and Amazon, use machine

learning to make personalized product or content suggestions.

Email spam ﬁlters use machine learning to automatically identify and block unwanted messages.

Fraud detection systems in banking and ﬁnance use machine learning to identify and prevent fraudulent

transactions.

Medical imaging systems use machine learning to automatically analyze medical images and identify

potential abnormalities.

Social media platforms use machine learning to automatically identify and remove inappropriate or

oﬀensive content.

2.5 The Future

The future of machine learning will likely be marked by continuing innovation and expansion. Quantum

computing has the potential to transform and advance machine learning because it permits simultaneous multi-

state operations, hence speeding data processing. AutoML or Automated Machine Learning is the process

of applying machine learning algorithms to real-world applications in an automated manner. AutoML makes

machine learning accessible to a larger audience, indicating its potential to transform the technology landscape.

The pharmaceutical and healthcare sectors believe that the future of healthcare technology will be built on

machine learning, since disease identiﬁcation and diagnosis become more simple and less prone to error with this

technology. Tesla, Honda, and Waymo are among the automobile manufacturers investigating the incorporation

of self-driving technology into their vehicles. Automakers have already developed partially automated vehicles,

2.5 The Future

but fully autonomous vehicles are still in development. Modern manufacturing technology incorporates machine

learning into production processes, such as predictive algorithms used to arrange equipment maintenance on an

adaptive rather than a ﬁxed timeline. Industry giants like Microsoft, GE, Bosch, Siemens, NVIDIA Fanuc, and

Kuka are already investing extensively in industrial AI with machine learning technologies to improve every

aspect of manufacturing.

With the development of new technologies, machine learning algorithms may be utilized more eﬃciently.

The future of machine learning entails a huge expansion in its use in several industries.

References

1. Machine Learning History: The Complete Timeline-Andrea Jacinto

2. Machine Learning: The Next Step in Advanced Analytics

3. How AI And Machine Learning Will Impact The Future Of Healthcare-Bernard Marr

4. Artiﬁcial Intelligence for All-Andrew NG

5. Sarker, I.H. Machine Learning: Algorithms, Real-World Applications and Research Directions. SN

COMPUT. SCI. 2, 160 (2021).

6. Deep Learning Architecture-Edwin Lisowski

7. Deep learning architectures - Samaya Madhavan, M. Tim Jones

About the Author

Blesson George is currently working as Assistant Professor of Physics at CMS College Kottayam,

Kerala. His research interests include developing machine learning algorithms and application of machine

learning techniques in protein studies.

Machines that can understand Human

Languages

by Jinsu Ann Mathew

airis4D, Vol.1, No.1, 2023

www.airis4d.com

(Image courtesy: https://analyticsindiamag.com/how-i-used-bidirectional-encoder-representations-from-transformers-bert-to-analyze-twitter-data/)

3.1 Introduction

How do humans interact with one another? We communicate in two ways: verbally and nonverbally.

Verbal communication refers to the use of words to exchange information with other people, either orally or in

writing. Nonverbal communication employs nonverbal techniques such as body language, which includes facial

expressions, gestures, and so on. Due to our common experiences in the world we live in, we can understand

what other people are saying. Speciﬁcally, we can associate symbols, traits, and images to words depending on

our prior experiences in the real world. Can a machine carry this out? If yes, how does a device accomplish this?

Computer code is the language of machines. In other words, rather than using words and dialogues, machines

speak a language made up of zeros and ones that produce logical actions. This is one of the main issues with

human-computer communication. Due to its complexity, computer language is diﬃcult for many individuals to

understand on their own and it’s not the cup of tea that most people would want to go through. This issue is

addressed by Natural Language Processing(NLP). The ﬁeld of Natural Language Processing is used to bridge

the communication gap between machines and people.

NLP is a branch of Artiﬁcial Intelligence that helps computers understand, interpret and manipulate human

language. In other words, NLP enables computers to understand natural language as humans do. As people, we

use text and voice to communicate. We refer to this data as unstructured since it is messier or not organized in

3.2 History of Natural Language Processing

a predetermined way. Computers only understand structured data. Structured data is information that is kept in

databases and spreadsheets and has elements that are addressable for eﬃcient analysis. Alternatively, we can

say that Structured data contains information with patterns, such as telephone numbers or email addresses and

Unstructured data contains information without patterns, such as text messages and images. Figure 3.1 depicts a

diagrammatic representation of structured and unstructured data. What NLP does is turn our unstructured data

into structured data so that computers can use it.

(image courtesy: https://blog.stealthbits.com/securing-structured-data/)

Figure 3.1: Structured and Unstructured data

3.2 History of Natural Language Processing

You could now be wondering, whether this technology is new or old. Actually, NLP has been with us

for more than 70 years. It started in the 1940s, following World War II. At this time, people recognised the

importance of translation from one language to another and they hoped to create a machine that could do this

type of translation automatically. However, the task was obviously not as simple as people had ﬁrst imagined.

Figure 3.2 depicts the timeline of NLP.

Figure 3.2: Timeline of Natural Language Processing

1940 : The concept of language translation was created, which inspired the concept of human to machine

language translation.

3.3 Challenges of NLP

1950 : The ﬁrst theory about NLP came up in the article “Computing Machinery and Intelligence ” by

Alan Turing that showcases the criterion of intelligence around the Turing test.

1960 : Implemented some of the ﬁrst successful applications of NLP. One such example is SHRDLU,

which was developed at MIT by Terry Winograd. Also, the ﬁrst bot ELIZA was created.

1970 : During this time many chatterbots were written for eg. Parry.

1980 : Symbolic approaches (also called expert systems) were used in NLP, with hard-coded rules and

ontologies.

1990 : Statistical models took over the symbolic approach. Statistical models were able to learn by

themselves, through machine learning, the multitude of hard-coded rules of the expert systems. This was made

possible above all by the increase in computational resources.

2000 : Begin to represent words with dense vectors of numbers called word embedding, so that words with

similar meanings are associated with similar vectors.

2010 : The deep learning revolution, made possible by the increase in data availability and processing

power, allows RNNs to outperform statistical methods.

2020 : Large language models like GPT3(transformer neural network consisting of 175 billions of param-

eters and requiring approximately 800GB of storage)were introduced.

3.3 Challenges of NLP

Teaching computers to make sense of human language has long been a goal of computer scientists. However,

this is a diﬃcult task due to the complexity, ﬂuidity, and unpredictability of human language. Below are some

of the diﬃculties that nlp has to deal with.

Contextual Words and Phrases: Same words and phrases can have diﬀerent meanings according to the

context of a sentence.While humans can easily understand the diﬀerence between words spoken at work, at

home, at store or at school, none of these diﬀerences are apparent to a computer algorithm.

Eg: I left my phone on the left side of the room.

Irony and Sarcasm: Irony and sarcasm present problems for machine learning models because they

generally use words and phrases that may be positive or negative, but actually imply the opposite. That is,

the reality of the situation is usually completely in contradiction with the reader’s or listener’s most basic

assumptions about it. For example, consider the illustration in Figure 3.3, although sarcasm in the comment

would be obvious to humans, it would be challenging to teach a machine how to interpret this phrase.

(image Courtesy:https://www.standoutbooks.com/irony-and-sarcasm/ )

Figure 3.3: Example of sarcasm

Synonyms: Because humans utilize numerous words to describe the same notion, synonyms can present

problems in NLP. Additionally, some of these terms may have the exact same meaning while others (such

3.3 Challenges of NLP

as, little, tiny, and minute) may have levels of complexity, and various people employ synonyms to indicate

somewhat diﬀerent meanings within their own personal vocabularies. As a result, it’s crucial to include all

conceivable synonyms and deﬁnitions of words while developing NLP systems.

Errors in text: Text analysis might be complicated by improperly spelled or used terms. A machine may

ﬁnd it challenging to understand spoken language because of mispronunciations, accent problems, stutters, etc.

Domain-speciﬁc language: A domain-speciﬁc language (DSL) is a language designed to be used ex-

clusively inside a speciﬁed domain. Language used by various ﬁrms and industries might vary greatly. For

instance, an NLP processing model required for the processing of medical records might diﬀer greatly from one

required for processing legal documents.

Even while NLP has its drawbacks, it nonetheless provides enormous, all-encompassing advantages for

every organization. Many of these limitations will be overcome in the upcoming years as new methods and

technologies emerge on a daily basis. As more research is being carried out in this ﬁeld, we expect more

breakthroughs that will make machines smarter at recognizing and understanding the human language.

References

Khurana, D., Koli, A., Khatter, K, et al. Natural language processing: state of the art, current trends and

challenges. Multimed Tools

Guzman, A. L., Lewis, S. C. (2020). Artiﬁcial intelligence and communication: A Human–Machine

Communication research agenda. New Media & Society, 22(1), 70–86.

Natural Language Processing:Bridging communication between Humans and Machines,Bereket Kas-

saye,Youthtime Magazine,2022

Natural Language Processing and Machine Learning, Ximena Bola

nos, encora,2021

What Writers Like You Need To Know About Irony And Sarcasm, Robert Wood, Standout Books

Unstructured Data vs Structured Data Explained with Real-life Examples, Yunee Ham, medium, 2020

What is Structured Data vs. Unstructured Data?, M-Files

Complete Natural Language Processing for beginners,Buggyprogrammer, 2021

A Beginner’s Guide to Natural Language Processing, Appventurez, 2022

A Brief Timeline of NLP, Fabio Chiusano, Medium, 2022

About the Author

Jinsu Ann Mathew is a research scholar in Natural Language Processing and Chemical Informatics.

Her interests include applying basic scientiﬁc research on computational linguistics, practical applications of

human language technology, and interdisciplinary work in computational physics.

Part II

Astronomy and Astrophysics

A sneak peak into the science behind

telescopes

by Linn Abraham

airis4D, Vol.1, No.1, 2023

www.airis4d.com

To the untrained human eye, most objects in the night sky resemble stars. However, to the astronomer they

become planets, galaxies, nebulae, Active Galactic Nuclei (AGNs), supernovae or even the International Space

Station (ISS). When Galileo Galilei used the telescope to look into the heavens for the ﬁrst time, a whole new

world slid into view. That historical moment was part of a sequence of events that would usher in the scientiﬁc

revolution and subsequently change mankind’s history. Today’s astronomical surveys employ the latest and

greatest in terms of telescopes and imaging systems. These large telescope systems do not have individual

astronomers sitting behind them to view the sky. Rather images of the sky are continuously taken and stored

in large data centers for further processing and analysis. Take the homepage of any astronomical survey and

you are bound to ﬁnd terms like the aperture size, ﬁeld of view and the number of megapixels among the

speciﬁcations. Let us try to understand the relevance of these parameters in observational astronomy and the

science behind them with the help of the humble pinhole camera.

1.1 Importance of Aperture

Contrary to popular perception, the utility of a telescope is not in its ability to make distant objects visible.

The human eye by itself is capable of viewing upto a distance of 2.5 million light years (the Andromeda Galaxy).

Telescopes are beneﬁcial to the astronomer because of two things, they make faint objects appear brighter (light

gathering power) and they allow objects that are seen as a single entity to be separated into its parts (resolving

power). Both these depend to a large extent on the aperture size of the telescope objective.

1.1 Importance of Aperture

Figure 1.1: On the left, Two images showing the eﬀect of lesser (top) and greater (bottom) light gathering

power. On the right, Images showing the eﬀect of worse (top) to better (bottom) resolving power. (Image

Courtesy: Chaisson/McMillan, Astronomy-4th edition))

What is the importance of aperture in a telescope? Let’s start that discussion by ﬁrst deﬁning what an

image is. In the most elementary terms, an image is the representation of a physical object. So by deﬁnition,

an image is not a real object; it’s unreal. But images themselves are classiﬁed as “real” or “virtual.” A “real

image” is where real light rays join together; hence, they can be caught on a screen. A virtual image like the

one you see in a plane mirror appears to come from some point behind the mirror - yet if you go behind the

mirror, you cannot ﬁnd it. Let’s limit our discussion to real images for now. All you need to get a “real” image

of something, is a pinhole camera. Let’s revisit the pinhole camera experiment that we were introduced to, in

elementary school. The basic setup needed to make a pinhole camera is pretty simple. Take an empty cardboard

box with one side open. Cover the open side using a translucent paper; this becomes the screen. On the opposite

side of the screen, make a hole using a pin. Keep this setup in a darkened room with only a candle ﬂame as the

light source. After adjusting the distance of the candle from the pinhole, voila!, an inverted image appears on

the screen.

1.2 Focal length, Field of View and Image magniﬁcation

Figure 1.2: Camera Obscura. The earliest published drawing of a camera obscura can be found in Gemma

Frisius’ 1545 book De Radio Astronomica et Geometrica. In the book, Frisius, a Dutch physician, mathemati-

cian, and instrument maker, describes and shows how he used the camera obscura to observe the solar eclipse

of 24 January 1544. (Image Courtesy: Wikipedia.org)

Why did we need an aperture in the ﬁrst place? What prevents an image from forming, in absence of a

pinhole ? The pinhole acts like a ﬁlter allowing only light rays that come from a particular point on the object

and in a particular direction to reach the screen. If you increase the size of the aperture, each point on the screen

starts to receive light rays from diﬀerent parts of the scene. And if the aperture becomes suﬃciently large the

image gets completely washed out. This is illustrated in Figure 1.1.

Figure 1.3: Without a barrier, there will be no image as all points on the screen receive light rays from multiple

points on the scene. (Image Courtesy: Udacity)

But, if a pinhole is suﬃcient to create an image, why do you need a lens? We saw that with pinholes,

in order to get a sharp image, you need to have a suﬃciently small aperture. This has the undesired eﬀect of

making the image dim. A pinhole cuts out a lot of parallel rays coming from adjacent points on the object. A

convex lens solves this issue as it can bend parallel rays onto a single point. Even though lenses replace the

aperture in a pinhole camera, modern camera lenses do have an aperture added on top of the lens similar to a

pinhole. In such a case, the aperture size of the lens refers to the actual diameter of the lens itself. Thus lenses

help us to collect light coming from a very large area much larger than the size of our pupil and focuses them

onto a point making even faint objects visible.

1.2 Focal length, Field of View and Image magniﬁcation

Returning to the pinhole camera, let’s see what happens when you increase the distance between the hole

and the screen. In the image that is formed, subjects close to the camera grow in size, whereas parts of the

background are lost. When this happens, we say that the ﬁeld of view decreases. In the case of a pinhole camera,

the distance between the hole and the screen is called the focal length. Thus the ﬁeld of view decreases when

1.2 Focal length, Field of View and Image magniﬁcation

the focal length increases and conversely, when the focal length decreases, the ﬁeld of view increases. This is

seen in the animation shown in Figure 1.4 Notice that the image that appears magniﬁed also appears dull and

fuzzy. This is because the same information is now spread over a larger surface area.

Figure 1.4: The eﬀect of moving your screen away from the pinhole is a decreased ﬁeld of view (Image

Courtesy: khanacademy.org/computing/pixar).

Figure 1.5: The larger the focal length of the lens used, the narrower the ﬁeld of view (Image Courtesy:

Udacity).

This relation between the focal length and the ﬁeld of view holds true even for lenses. This can be seen in

ﬁgure 1.5. The decreased ﬁeld of view leads to higher magniﬁcation for lenses with a longer focal length. (Note

that this is not the case with microscopes where a shorter focal length leads to better magniﬁcation). Telescopes

makes use of a compound lens system where the long focal length lens (objective) focuses an image of a distant

object much closer to our eyes. This image can be much smaller than the actual object however it has a greater

angular magniﬁcation as it is now much closer to our eyes. This image is now viewed through the eyepiece

which further magniﬁes it. The total magniﬁcation of the telescope is the product of these two magniﬁcations.

Thus the magniﬁcation is not an essential feature of a telescope as it can be increased arbitrarily by increasing

the magniﬁcation of the eyepiece.

However the resolving power of the telescope is a characteristic property which has a theoretical limit set

1.2 Focal length, Field of View and Image magniﬁcation

by the equation

sin θ ≈ θ = 1.22λ/D

Where D is the aperture of the telescope objective. Magniﬁcation beyond this limit is useless as no more

resolution is obtained and the image ends up being mushy. The resolving power is further limited in modern

digital cameras by the pixel size of the CCD sensors.

Finally, lets try to understand the importance of the number of megapixels that is often quoted with cameras.

It can be seen from Figure 1.4 that because a shorter focal length results in a wider ﬁeld of view, the size of

the screen needs to be suﬃciently big in order to capture the entire scene. With cameras, what this means is

that when you talk about more megapixels (higher resolution) for a sensor, what you get is an increased ﬁeld of

view given the same optics, scene, and pixel size. If you keep the sensor size the same and decrease the size

of individual pixels, what you get is more digital zoom capability. Digital zoom lets you uncover details that

otherwise would be washed out with a lower-resolution sensor. However this is often at the expense of reduced

sharpness and increased noise in the image.

References

1. First Principles of Computer Vision, Shree Nayar, Columbia University

2. Introduction to Computer Vision, Dr. Aaron Bobick, Georgia Tech University

3. Crash Course on Astronomy, Phil Plait, Youtube.com, 2012

4. Resolution versus Field of View, Adimec.com , September 28, 2017

5. Pixar in a box, Eben Otsby, Khan Academy

About the Author

Linn Abraham is a researcher in Physics, specializing in A.I. applications to astronomy. He is currently involved

in the development of CNN based Computer Vision tools for classiﬁcations of astronomical ZTF sources and

from PanSTARRS optical images. He has used data from a several large astronomical surveys including SDSS,

CRTS, ZTF and PanSTARRS for his research.

Evolution of Satellite Imaging

by Robin Jacob Roy

airis4D, Vol.1, No.1, 2023

www.airis4d.com

Satellite imaging has come a long way since the ﬁrst artiﬁcial satellite, Sputnik, was launched into orbit

in 1957. Initially, satellites were primarily used for military and intelligence gathering purposes, but over time,

they have become an important tool for a wide range of scientiﬁc and commercial applications.

One of the key developments in satellite imaging was the launch of the ﬁrst Earth observation satellite,

the NASA-USGS Landsat 1, in 1972. This satellite was equipped with a multi-spectral scanner that allowed it

to capture images of the Earth’s surface in diﬀerent wavelengths of light, providing valuable information about

the composition and condition of the land. The Landsat program has continued to evolve, with each successive

satellite improving upon the capabilities of its predecessor.

Another major advance in satellite imaging came with the development of radar imaging technology. This

allows satellites to capture images of the Earth’s surface using microwaves, which can penetrate clouds and

other obstacles that would block visible light. This has enabled the creation of high-resolution, detailed images

of the Earth’s surface, even in conditions that would be diﬃcult or impossible to observe using optical sensors

alone.

In recent years, satellite imaging has also become more accessible to the general public. The proliferation

of small, inexpensive satellites, known as CubeSats, has made it possible for individuals and organizations to

launch their own imaging satellites. This has opened up new possibilities for citizen science and crowd-sourced

data collection, as well as providing new sources of imagery for a variety of applications including monitoring

natural disasters, tracking global agricultural production, and studying the eﬀects of climate change.

Satellite imagery has a wide range of uses, including environmental monitoring, urban planning, weather

forecasting, and military surveillance. Environmental organizations often use satellite imagery to monitor

deforestation, oil spills, and other environmental changes. Urban planners can use satellite images to study

population density, land use, and other factors in order to make informed decisions about the development

of cities. Weather forecasters use satellite imagery to track storms and other weather patterns, and military

organizations use satellite imagery for surveillance and intelligence gathering. Overall, satellite imagery is a

powerful tool for understanding and monitoring the Earth and its environment.

Let’s look into two major satellite missions in existence today: the Landsat Mission and the Sentinel

Mission:

2.1 Landsat mission

Figure 2.1: Landsat mission timeline. Image Courtesy: USGS website

The Landsat program, a joint venture between NASA(National Aeronautics and Space Administration) and

the USGS(U.S. Geological Survey), is the longest-running satellite imagery acquisition program in operation.

The ﬁrst satellite in the program, the Earth Resources Technology Satellite, was launched in 1972 and was later

renamed Landsat 1 in 1975. The most recent satellite, Landsat 9, was launched in September 2021. The entire

timeline of the various Landsat missions are depicted in Figure 2.1. The instruments on the Landsat satellites

have collected millions of images, which are available for global change research and applications in a variety

of ﬁelds, including agriculture, cartography, geology, forestry, regional planning, surveillance, and education.

These images can be accessed through the USGS EarthExplorer website. Each Landsat scene is approximately

185 km by 185 km in size and has a temporal resolution of 16 days.

Lets look at the various spectral bands of the latest Landsat-9 mission. Landsat 9, like Landsat 8, is

designed to capture four visible spectral bands, one near-infrared (NIR) band, three shortwave-infrared(SWIR)

bands at 30 m spatial resolution, one panchromatic band at 15 m spatial resolution, and two thermal bands with

a spatial resolution of 100 m. Additionally, Landsat 8 and Landsat 9 have two spectral bands, which helps

scientists to measure high thin clouds and water quality. First one being the Aerosol/Coastal band(Band 1) used

to measure chlorophyll concentrations in oceans coming from phytoplankton. This band is also used for aerosol

detection. The second one being the Cirrus Band(Band 9) used to spot high, thin Cirrus clouds which can alter

the values of the data.

2.2 Sentinel Mission

Figure 2.2: The diﬀerent Sentinal missions. Image Courtesy: ESA website

The Copernicus Sentinel program is a collaboration between the European Space Agency (ESA) and the

European Commission to provide a continuous, accurate, and up-to-date satellite observation of Earth. The

program consists of a series of dedicated satellites and instruments, known as Sentinels, that are speciﬁcally

designed to monitor various aspects of Earth’s environment, including land, oceans, atmosphere, and climate.

The information collected by the Sentinels is used for a wide range of applications, including climate and

environmental monitoring, disaster management, and support for various scientiﬁc and commercial activities.

The program is named after the Renaissance astronomer Nicolaus Copernicus, who is credited with developing

the ﬁrst comprehensive heliocentric model of the solar system.

The Sentinel program comprises of seven subsequent missions starting from Sentinel-1, which was launched

in April 2014. Figure 2.2 shows the entire missions coming under the Sentinel program which we will discuss

in detail.

Sentinel-1 is a polar orbiting day and night Land and ocean radar imaging mission which can acquire

data in all weather conditions.

Sentinel-2 comprises of two polar orbiting satellites with the objective of land monitoring. The mission

captures multi-spectral high resolution data of vegetation, inland waterways, soil and water cover. It was

launched in June 2015.

Sentinel-3 is a multi instrument mission, whose primary objective is to measure the ocean surface

topology, sea and land surface temperature and sea and land color with high accuracy and reliability.

Sentinel-4 is dedicated to atmospheric monitoring. The mission focuses on the continuous air quality

monitoring at high spatial and temporal spatial resolution.

Sentinel-5 is a payload which monitor the air quality from the polar orbit. This mission has a global

coverage and continuously monitors the composition of the Earth’s atmosphere.

Sentinel-5P, also known as the Sentinel-5 Precursor aims to perform the measurement of the Earth’s

2.2 Sentinel Mission

atmosphere, with high spatio-temporal resolution, relating to trace gases and aerosols aﬀecting air quality

and climate.

Sentinel-6 measures the global sea surface height using radar altimeter, primarily for climate studies and

operational oceanography.

The future of satellite imaging looks bright. Advances in technology are expected to continue to improve

the resolution and accuracy of satellite images, making them even more useful for a wide range of applications.

At the same time, the growing availability of satellite imagery is likely to continue to democratize access to

this valuable resource, making it possible for more people and organizations to use it to better understand and

protect our planet.

Overall, the evolution of satellite imaging has had a profound impact on our understanding of the Earth

and its environment. By providing detailed, high-resolution images of the planet, satellites have enabled

researchers to gain new insights into a wide range of scientiﬁc, commercial, and environmental issues. As

satellite technology continues to improve, the potential for new discoveries and applications will only continue

to grow.

References:

Fifty Years of Earth Observation Satellites, Andrew J. Tatem, Scott J. Goetz, Simon I. Hay, American

Scientist, Volume 96, Number 5, 2008, 390

The Sentinel Mission

The Copernicus Sentinel program

The Landsat Missions

Landsat-9 Spectral Bands

About the Author

Robin is a researcher in Physics specializing in the applications of machine learning for remote

sensing. He is particularly interested in using computer vision to address challenges in the ﬁelds of biodiversity,

protein studies, and astronomy. He is currently working on classifying satellite images with Landsat and Sentinel

data.

Eclipsing Binaries

by Sindhu G

airis4D, Vol.1, No.1, 2023

www.airis4d.com

Do you know what eclipsing binaries are? They are one type of variable star. Okay. So before we are

talking about eclipsing binaries we have to look at what variable stars are. Simply put, a variable star is a

star that changes its brightness over time. These changes can be very drastic. For example, in the case of a

supernova, the magnitude change can be several thousands! So it might appear that a new star has come up

from nowhere in the middle of the night! Some of these stars are so bright that they become visible even during

the day and remain visible for several days before vanishing into the dark sky. There are also many other types

of stars in the variable universe, so it is called, that changes their brightness from fractions of a second to many

years. The take home message is that all these changes depend on the type of the star.

Figure 3.1: Variable stars (Source : AAVSO)

Variable stars are mainly classiﬁed into intrinsic variable stars and extrinsic variable stars. In the case of

intrinsic variables, the brightness changes are due to the changes in the physical properties of the star. Extrinsic

variables are stars in which the variation in brightness is due to the inﬂuence of external factors such as eclipse

or rotation. Eclipsing binaries belong to the category of extrinsic variable stars. In eclipsing binary stars the

brightness variations are due to the eclipse of one star by the companion object. The best known example of an

eclipsing binary is Algol.

Eclipsing binaries are also one type of binary star systems. Binary star systems contain two stars that orbit

around their common center of mass. Center of mass of a binary star system is a point at which the whole of

the mass of this system appears to be concentrated. Binary stars are gravitationally bound to each other. In a

binary star system the brighter star is the primary star, while the dimmer star is the secondary star. What is the

case when these companion stars are of equal brightness? In this case the designation given by the discoverer

3.1 Light Curve of Eclipsing Binary Stars

is respected. Many stars in this universe are in binary star systems. Figure 3.2 shows a binary star system.

Eclipsing binaries are binary stars orbiting in a plane containing the line of sight of the observer so that they

will eclipse each other.

Figure 3.2: Centre of mass of binary system (Source: Cosmos )

3.1 Light Curve of Eclipsing Binary Stars

Figure 3.3: Example light curve of an eclipsing binary star system. (Source: NASA)

Light curves are important tools in astronomy. Light curve is a plot that shows the brightness of an

astronomical object over a period of time. Figure 3.3 shows a light curve of an eclipsing binary system. Light

curves can be periodic or aperiodic. We can identify the objects from their light curves. Light curves can be

used to determine the size, temperature, mass and distance to stars. Light curves are also used to understand the

processes that occur within the astronomical object. If the data is periodic over a longer period of time, we use

phase folded light curves. Here we plot magnitude as a function of phase rather than time in the light curve. In

phase folded light curves the data is folded to ﬁt within a mathematically deﬁned or determined period. Figure

3.4 shows a phase-folded light curve of an eclipsing binary.

3.2 Types of eclipsing binary systems

Figure 3.4: Phase-folded light curve of an eclipsing binary (Source : ATNF)

Figure 3.5 shows the brightness variation of eclipsing binaries. Eclipsing binaries have periodic light

curves. That means they show periodic change in their brightness. Light curves of eclipsing binaries show

periodic drops in brightness that occur when one of the stars is eclipsed by another. They show periods of

constant light when one component does not come in front of the other component. The brightness of this

binary system may drop twice during the orbit.The eclipse with greater drop in brightness is called the primary

eclipse. The eclipse with a shallow dip in brightness is called a secondary eclipse. Do you know when the

primary and secondary eclipses happen? When the brighter companion is eclipsed by the fainter one then the

primary eclipse occurs. The secondary eclipse happens when the fainter one is eclipsed by the brighter one.

What happens when the two stars have the same brightness? In this case both eclipses will be equal. For

visual observers, the secondary eclipse may not be observed if one companion is much fainter than the other.

Secondary eclipse may occur midway between primary eclipses if the binary stars have circular orbits. If the

orbits of the binary stars are elliptical then the secondary eclipse will occur earlier or later than midway. Light

curve’s shape depends mainly on the size and relative brightness of the stars. Shape of the light curve also

depends on orbital inclination of the binary stars as seen from Earth.

Figure 3.5: Change in the light intensity due to transit (Source: Wikipedia)

We can ﬁnd the orbital period of the eclipsing binaries from the light curve of the system. Orbital period

of the eclipsing binary system is the time between the midpoints of the sequential primary minima. Orbital

periods of these systems may not be constant. Mass transfer occurs if the two stars are close together. In this

case the period of this system can change. Amateur astronomers made a valuable contribution to detect the

changes in the period over months, years and decades.

3.2 Types of eclipsing binary systems

Based on the shape of the light curve, eclipsing binaries are classiﬁed into three main types. They are

3.2.1 EA - Algol type

They are eclipsing binary systems with spherical or slightly ellipsoidal components. It is possible to

specify the moments of the beginning and the end of the eclipses from the light curves. The light remains

almost constant or diﬀers insigniﬁcantly between the eclipses. Secondary minima may be absent. The period of

this system ranges from 0.2 to 10000 days. Light amplitudes of the EA system may reach several magnitudes.

Stars in this system may be detached or semi-detached. Figure 3.6 shows the light curve of Algol type eclipsing

binary.

Figure 3.6: Light curve of EA type eclipsing binaries. (Source : VS-COMPAS Project)

3.2.2 EB - Beta Lyrae type

They are eclipsing binary systems with ellipsoidal components. It is impossible to specify the exact

moments of the beginning and the end of the eclipses. Secondary minima is observed in every case. Depth

of secondary minimum considerably smaller than the primary minimum. Between the primary and secondary

eclipses there is no time of constant light. Periods of the EB system are mainly longer than one day. Light

amplitudes of this eclipsing binary systems are usually less than 2 magnitudes in V. The components are

invariably semi-detached. Figure 3.7 shows the light curve of Beta Lyrae type eclipsing binary.

Figure 3.7: Light curve of EB type eclipsing binaries. (Source : VS-COMPAS Project)

3.2.3 EW - W Ursae Majoris type

They are eclipsing binary systems with ellipsoidal components. In this type the component stars are almost

in contact. They may share a common envelope of material. It is impossible to specify the exact moment of

onset and end of the eclipses. The primary and secondary minima have almost equal depth or its depth diﬀers

3.2 Types of eclipsing binary systems

insigniﬁcantly. They have periods shorter than one day. Light amplitudes of EW type eclipsing binaries are

usually less than 0.8 magnitudes in V. Figure 3.8 shows the light curve of W Ursae Majoris type eclipsing binary.

Figure 3.8: Light curve of EW type eclipsing binaries. (Source : VS-COMPAS Project )

3.2.4 Importance of the study of eclipsing binaries

Period changes may be evidence of the presence of a third component in the eclipsing binary system. It

can be the evidence of the evolution of the stellar system. Eclipsing binaries are used to determine the stellar

quantities such as radius, mass, etc. From the study of eclipsing binaries we can understand the mass transfer

between the component stars with time. If an eclipsing binary is also a spectroscopic binary then we can

measure the period, separation, velocities and inclination angle.

References

Variables: What Are They and Why Observe Them?,AAVSO

Variable star classiﬁcation and light curves, AAVSO, 2012

Variable Stars,Australia Telescope National Facility, CSIRO

Types of Binary Stars,Australia Telescope National Facility, CSIRO

Eclipsing Binary Observing Guide,British Astronomical Association

About Light Curves,AAVSO

Variable Stars and Phase Diagrams,AAVSO

Light Curves Phase Diagrams,Boyce Research Initiatives and Education Foundation

Project VS-COMPAS,2015

Why are eclipsing binary stars important?,AAVSO

GCVS variability types, Samus N.N., Kazarovets E.V., Durlevich O.V., Kireeva N.N., Pastukhova

E.N.,General Catalogue of Variable Stars: Version GCVS 5.1,Astronomy Reports, 2017, vol. 61, No. 1,

pp. 80-88 2017ARep...61...80S

About the Author

Sindhu G is a research scholar in Physics doing research in Astronomy & Astrophysics. Her research

mainly focuses on classiﬁcation of variable stars using diﬀerent machine learning algorithms. She is also doing

the period prediction of diﬀerent types of variable stars, especially eclipsing binaries and on the study of optical

counterparts of x-ray binaries.

Part III

Biosciences

Jewels of air- A Dragonﬂy story

by Geetha Paul Jacob

airis4D, Vol.1, No.1, 2023

www.airis4d.com

1.1 Introduction

Figure 1.1: The dragonﬂies in your backyard have a history older than the dinosaurs

Delicate, dainty ﬂuttering wings, fascinating hovering motions; who has not been attracted by dragonﬂies

and damselﬂies that ﬂutter around our backyard since childhood? It is not just that they are beautiful, but they

are also one of nature’s lifeline signals on the overall health of our ecosystems. In reality, odonates are very

sensitive to climate, water quality and landscape changes. Being natural predators, they also control mosquitoes

and other pathogens on the planet. Dragonﬂies existed from the Carboniferous period that dates back to over 330

million years. Fossil records and carbon dating have shown that dragonﬂies have a much longer evolutionary

history than most species.

However, the diversity of dragonﬂies has undergone a drastic decline in the past few decades. The fact that

they ﬁnd it diﬃcult to survive in the current environment is regarded as the tip of the iceberg that demonstrates

how human greed has been exploiting the wealth of this planet.

Dragonﬂies are highly eﬃcient ﬂyers capable of making intercontinental ﬂights. They can hover, ﬂy

backwards, forwards and sideways. They have evolved and adapted to the changes on this planet, and in fact,

some of them can breed and survive even in the most contaminated environments (Edgehouse, et al.,2014).

While odonates are known predators of harmful insects, mosquito larvae, and other pathogens, a world without

odonates will be a world of pathogens and epidemics. This is a very alarming situation and is a signiﬁcant

concern.

1.2 Story of a Dragonﬂy

Odonates may be visually classiﬁed by closely observing the colour and patterns on their forehead,

prothorax, thorax, abdomen and wings. Depending upon the species’ kind, the details may vary. The colours

and patterns may vary between male and female species as well.

1.2 Story of a Dragonﬂy

Once in a little lagoon, down below on the surface of a muddy bottom, there lived a little water beetle with

a community of water beetles. They lived a comfortable and busy foraging life in the lagoon. One ﬁne day their

community came to face a sad event; one fellow beetle among them climbed up to the stem of the leaf of a water

lily and never be seen again. They believed that their friend was dead and gone forever. Then again, one ﬁne

day, another beetle felt an irresistible urge to climb up the stem above water. He decided to climb and check for

his gone friend and determined that he would not leave his friends. When he climbed up to the top and reached

out of the water, he got tired and ﬁnally went to sleep, tightly clasping the stem of the water lily. As he slept, he

could feel some bodily changes taking place within him, his outer skin got dried and shrunk, and when he woke

up, he felt suﬀocated within the dry skin. The dry outer integument broke and came out as a beautiful red-tailed

slender dragonﬂy with transparent and broad wings designed for ﬂight. So, he decided to ﬂy. He soared up in

the sky and saw the beauty of the universe in a much superior way that he had not seen before. He remembered

his old beetle friends and wanted to go back and tell them that he was now more alive than he had ever been

before. But his new body would not go down into the water. Then he understood that their time would come

when they, too, would know what he knew now. So he raised his wings and ﬂew oﬀ into his joyous new life.

1.3 History & Science behind the story of a dragonﬂy.

The Order Odonata is among the most ancient of Earth’s fauna. Fossils of the order Protodonata, the

ﬁrst recognisable progenitors of present-day dragonﬂies, were known from the Upper Carboniferous period

320 million years ago. Odonata comprises three groups - Anisoptera (Dragonﬂies), Zygoptera(Damselﬂies)

and Anisozygoptera. Odonates lay their eggs in freshwater, and the more signiﬁcant part of their lives as lar-

vae/nymphs is spent in aquatic habitats such as rivers, lakes, ponds or water-ﬁlled tree holes. The metamorphosis

of Odonata has only three stages, egg, larva and adult. The larval life span varies from a few weeks to several

years, during which period they grow in size by shedding their exoskeletons. The fully grown larvae emerge

from the water, and the aerial stage of life begins. Life as a ﬂying insect lasts only a few months. The life of

the adult odonate is spent foraging, establishing territory and ﬁnding a mate to ensure progeny. Odonates are

carnivorous; they are cannibalistic too. Most species spend their lives near water bodies. Species like Pantala

ﬂavescens migrate long distances from India to Africa. Their next generation migrates back, and the cycle

continues forever. Odonate behaviour is a fascinating subject to study. They are very aggressive, agile ﬂiers

(can ﬂy forward and backwards, upward and downward, and hover). They hunt and feed primarily in mid-air on

the wing. Their compound eyes each have up to 30,000 ommatidia, and the visual ﬁeld is almost 3600. They

can detect colour, UV light and movement, which makes them perfect hunting machines. It has been reported

that they have a successful hunting rate of 95% compared to 50%of Great WhiteShark or 25% of African Lions.

The habitats of these beautiful insects are threatened as humans relentlessly destroy the environment, and water

bodies dry up or get land-ﬁlled or contaminated beyond redemption. The number of identiﬁed odonate species

is over 6300 worldwide, 493 in India,196 in the Western Ghats and 175 in Kerala. Many more remain to be

discovered. Unless the natural environment is protected and conserved, the world will lose them forever.

Dragonﬂies belong to the Phylum Arthropoda, Class Insecta, Order Odonata and Suborder Anisoptera.

1.4 Dragonﬂy Metamorphism - Hemimetabolous insects

Dragonﬂies were some of the ﬁrst winged insects to evolve 300 million years ago. Modern dragonﬂies have

wingspans of only two to ﬁve inches, but fossil dragonﬂies have been found with wingspans of up to two feet.

Dragonﬂies are agile ﬂiers, can hover in mid-ﬂight for almost one minute, rotate 360 degrees, ﬂy backwards,

and intercept live prey from mid-air.

1.4 Dragonﬂy Metamorphism - Hemimetabolous insects

Figure 1.2: Life cycle- Hemimetabolism stages- Dragonﬂy & Damselﬂy

Unlike other insects, dragonﬂies don’t have the pupa stage in their lifecycle and therefore have incomplete

metamorphosis. In complete metamorphism, there are all the stages- egg, larva, pupa and adult. Dragonﬂies

are called hemimetabolous (See Figure 1.2 as they do not have a pupa stage in their life cycle. Odonates lay their

eggs in freshwater, and the larger part of their lives as larvae is spent in aquatic habitats such as rivers, lakes,

ponds and even water-ﬁlled tree holes. The metamorphosis of odonate has only three stages, egg, nymph/larva

and adult. The nymphal stage varies from a few weeks to several years, during which period they grow in size by

shedding their exoskeletons. The fully grown larva emerges from the water, and the aerial stage of life begins.

Life as a ﬂying insect lasts only a few months.

Stage 1

Larva climbs above water up to the stem; legs take a ﬁrm grip on the stem or leaf, and breathing becomes

adapted to air rather than underwater.

A split appears on the back of the thorax of the larval case, and the thorax of the emerging adult pushes

backwards through the slit. The head, legs and upper part of the abdomen emerge.

Stage 2

The suspended insect swings forward and backwards to gather energy before ﬂinging itself upwards to grab

with its legs on the head of the larval case and then pulls out its abdomen and continues to cling to the larval

case, called the ‘exuvia’.The new emergent is called the “Imago”.

1.5 Common Dragonﬂy species

(a) Stage1

(b) Stage2

female with its Exuvia

Figure 1.3: Stages of emergence in Trithemis aurora female

1.5 Common Dragonﬂy species

(a) Male (b) Female (c) Nymph

Figure 1.4: Trithemis aurora (Burmeister , 1839 )

Common name: Crimson marsh glider

Family: Libellulidae

The dragonﬂies in Figure 1.4 are commonly known as Crimson marsh gliders. Males and females are

morphologically distinct in colouration. The male is crimson red, while the female is yellowish with black

stripes.

Male: Eyes are crimson red above and brown laterally. Reddish thorax with purple pruinoscence. The

abdomen is crimson red with a violet tinge, the base of which is swollen. Wings are transparent with a crimson

tint and basal amber patch. Wing Spot dark - reddish brown & legs black.

Female: Eyes with purple-brown above & greyish below. Olivaceous thorax with median & lateral stripes,

which are brown & black, respectively. The abdomen is reddish brown with median & later black stripes. The

1.6 Dragonﬂies as Ecological indicators

wing spot is dark brown & legs are greyish with yellow markings. Wings are transparent with bright yellow to

brown tinge. The basal portion is with pale amber colouration & tip is brown.

(a) Male (b) Female (c) Nymph

Figure 1.5: Rhyothemis variegata ( Linnaeus, 1763 )

Common name: Picture wing

Family: Libellulidae

The dragonﬂies in Figure 1.5 are commonly known as picture wing. They are medium-sized dragonﬂies

with dark bodies and reddish-brown eyes. Has an iridescent green thorax and black abdomen. Legs and wing

spots are black. They appear like butterﬂies because of their variegated markings of yellow and black on the

wings. In the case of the male, their forewings are transparent except for black spots in the nodes and borders.

The hindwings are also transparent, but the basal portion of the wings has black and yellow markings.

Females are similar to males except for the markings in their wings, and they have deep black markings.

The apical half portion of the wings is transparent, but the later basal half is opaque with black and yellow

markings. The hindwing is variegated with black and yellow, but the apical tip is transparent.

(a) Male (b) Female (c) Nymph

Figure 1.6: Pantala ﬂavescens (Fabricius, 1798)

Common name: Global skimmer or Global skimmer

Family: Libellulidae

The dragonﬂies in Figure 1.6 are commonly known as Wandering Glider/ Global skimmer.

Male: Body is reddish brown in colour, eyes, neck and thorax light brown, Abdomen is orangish brown

with a longitudinal line in the middle. Wings are transparent and are brown-coloured pterostigma.

Female: The body is yellowish orange in colour, and the abdomen is orange in colour. The wings are

transparent and are brown-coloured pterostigma.

1.6 Dragonﬂies as Ecological indicators

According to Ecologist Pankaj Koparde “Dragonﬂies and damselﬂies are predatory insects. They play a

signiﬁcant role in controlling the insect population, especially pests such as mosquitoes and agricultural pests.

They are freshwater insects, showing a semi-aquatic life cycle. Their larvae are underwater, and adults are

terrestrial and aerial predators.” The presence of dragonﬂies can reveal changes in the water ecosystems more

quickly than studying other animals or plants. In Fact, from the nymph to the adult stage, the dragonﬂy has

a signiﬁcant, positive ecological impact. Dragonﬂy eggs are laid and hatched in or near water, so their lives

impact water and land ecosystems. Once hatched, dragonﬂy nymphs can breathe underwater, which enables

them to eat mosquito larvae, other aquatic insects and worms and even tadpoles and small ﬁshes, while adult

dragonﬂies in the air capture and eat adult mosquitoes. Community-wide mosquito control programs that spray

insecticides to kill mosquitoes also kill dragonﬂies. Dragonﬂies indicate a positive impact on the ecosystem.

1.7 Conservation goals

The well-being of both dragonﬂies and humans is entwined through healthy freshwater systems. While

certain dragonﬂy species have survived much environmental, often stressful, change in the past, today, they

are experiencing novel adverse impacts from human activity. They are susceptible to these novel eﬀects, so

conservation action is required to save them.

To do this, we ﬁrst must value healthy landscapes and waterscapes, both systems required to protect

dragonﬂies. After identifying the key threats, conservation goals are put in place, followed by conservation

action focused on removing the threats.

References

Predatory Luring Behavior of Odonates, Michael Edgehouse,& Christopher P. Brown,Journal of Insect

Science, Volume 14, Issue 1, 2014, 146

Dragonﬂy, ScienceDirect, Journals & Books, 2021

Introduction to Odonata, Society for Odonate Studies, Ver.1, 2020

Dragonﬂies, the forgotten indicators of the ecosystem’s good health, Prajakta Joshi, The Bridge Chronicle,

27 April, 2020

About the Author

Geetha Paul is one of the directors of airis4D. She leads the Biosciences Division. Her research

interests extends from Cell & Molecular Biology to Environmental Sciences, Odonatology, and Aquatic Biology.

Healthcare developments during the Covid

era

by Dr Philip Mathew

airis4D, Vol.1, No.1, 2023

www.airis4d.com

Global pandemic COVID-19 is an infectious disease caused by the novel severe acute respiratory syndrome

coronavirus 2 (SARS-CoV-2). Globally, the COVID-19 epidemic has resulted in a devastating loss of life and

posed an unparalleled threat to public health. Although the pandemic has taken a toll on all the countries in the

world, we saw humankind tackling every hurdle that came in the way in the best possible manner. There has

been immense development in the ﬁeld of vaccines and healthcare technology (1,2).

In the initial phase of the pandemic, isolation and mitigation had been the mainstay of pandemic manage-

ment due to high disease transmission, a case fatality rate of more than 1%, and the lack of an eﬃcient antiviral

medication or vaccine. Pandemic preparedness and response were facilitated by digital health technology in

ways that are challenging to accomplish manually (1). Monitoring and tracking individuals became a national

or global necessity and new technologies were developed to aid this.

Many countries were able to monitor the movement of people using tools like migration maps, which

capture real-time position information on individuals using social media, mobile payment apps, and mobile

phones. With this data, machine learning models were created to estimate geographical SARS-CoV-2 emergence

and spread, and surveillance (3).

In addition to identifying the pattern of disease, early diagnosis and quantiﬁcation of the severity of the

condition were done to plan critical resource allocation like hospital beds and manpower. Rapid diagnosis

and risk assessment of COVID-19 were facilitated by artiﬁcial intelligence (AI). A cloud-based AI-assisted CT

service was used to ﬁnd COVID-19 pneumonia cases. This innovation quickly analyzes CT scans, distinguishing

COVID-19 from other lung conditions and greatly accelerating the diagnostic procedure (4).

Blockchain, one of the newer digital technologies, oﬀers distinctive qualities (including immutability,

decentralization, and transparency) that can be helpful in a variety of ﬁelds such as handling electronic medical

records and mobile health) (5).

Beyond the sick patients, there were thousands of chronic patients who required regular consultations and

care from clinicians but were unable to access their doctors due to restrictions and fear of being exposed to

the virus. To lessen patients’ exposure to Covid virus in healthcare facilities, virtual care services have been

employed globally. These platforms use video conferencing and digital monitoring to provide healthcare care

to patients. Around the world, general practitioner and hospital outpatient sessions have signiﬁcantly changed

as up to 75 percent of all consultations were conducted digitally (1, 6).

Amongst all these developments, the one that brought the greatest relief was the development of the

vaccine against covid-19 infection. Early in 2020, researchers started looking for a SARS-CoV-2 coronavirus

vaccine, but they were careful not to make any promises of speedy success. The mumps vaccine, created in

the 1960s, took four years to develop from viral sample to approval. Two mRNA COVID-19 vaccines, the

Pﬁzer-BioNTech and the Moderna COVID-19 vaccines, received emergency use authorization from the FDA in

2021. The FDA ﬁrst granted emergency use authorisation to COVID-19 vaccines based on minimal data than

is typically necessary because there is a pressing need for them. Before the FDA can authorize or approve a

vaccine for emergency use, the data must demonstrate that the vaccine is both safe and eﬃcacious. Extensive

safety monitoring of vaccines has occurred and is still occurring. Millions of COVID-19 vaccinations have been

administered during this pandemic (7).

India did prove her prowess by developing COVAXIN. In conjunction with the National Institute of Virology

of the Indian Council of Medical Research (ICMR), Bharat Biotech developed COVAXIN, an indigenous

Covid-19 vaccine. The BSL-3 (Bio-Safety Level 3) high containment facility at Bharat Biotech serves as the

development and manufacturing site for this locally produced, inactivated vaccine (8).

In summary, a variety of digital technologies have been adopted for large-scale activities including mass

testing, quick contact tracing, managing supply chains for vaccines and medications, telemedicine consultations,

etc.Some of these developments continue to be part of our lives while some have been reduced to just a fad once

society opened up to physical meetings. Covid-19 has taught us that no country can survive as an island and we

have learned something or other to manage this pandemic from every country in the world.

References

1. Whitelaw S, Mamas MA, Topol E, Van Spall HGC. Applications of digital technology in COVID-19

pandemic planning and response. The Lancet Digital Health. 2020;2(8):e435-e440. doi:10.1016/S2589-

7500(20)30142-4

2. Renu N. Technological advancement in the era of COVID-19. SAGE Open Med. 2021;9:20503121211000910.

doi:10.1177/20503121211000912

3. Wu JT, Leung K, Leung GM. Nowcasting and forecasting the potential domestic and international

spread of the 2019-nCoV outbreak originating in Wuhan, China: a modelling study. The Lancet.

2020;395(10225):689-697. doi:10.1016/S0140-6736(20)30260-9

4. Liu, J. (2020). Deployment of IT in China’s ﬁght against the Covid-19 pandemic. ITN Online.

5. Ng WY, Tan TE, Movva PVH, et al. Blockchain applications in health care for COVID-19 and be-

yond: a systematic review. The Lancet Digital Health. 2021;3(12):e819-e829. doi:10.1016/S2589-

7500(21)00210-7

6. COVID-19: Ten healthcare reaction and resilience trends - KPMG Global (home.kpmg)

7. The lightning-fast quest for COVID vaccines — and what it means for other diseases (nature.com)

8. Covid-19 vaccines under trials in India.

About the Author

Dr. Philip Mathew is the Head of Critical Care at Believers Church Medical College Hospital

with 15+yrs of experience in anaesthesia & Critical Care in India & abroad. He also serves as the Deputy

Director-Innovation with more than 5 years experience in mentoring start-ups in addition to being named mentor

at international innovation platforms.

Part IV

Computer Programming

Mastering R: A Challenge for Beginners

by Nevin Thomas

airis4D, Vol.1, No.1, 2023

www.airis4d.com

Learning a programming language can be challenging and overwhelming and R is no exception. The

steep learning curve and vast array of functions and packages can make it diﬃcult for beginners to get started

with R, but the rewards of learning this powerful language are well worth the eﬀort. It is widely used in

data analysis, visualization, statistical modelling and machine learning with a large and active community of

users and developers who contribute to its development and share resources and knowledge about how to use it

eﬀectively. In addition to that, it is also used in a variety of ﬁelds, such as sports, biology, and social science.

In the early 1990s, Ross Ihaka and Robert Gentleman developed the R programming language at the

university of Auckland in New Zealand and the name “R” is derived from the ﬁrst initials of the developers’

names. R was designed to be a free, open-source alternative to commercial software packages such as SPSS

and SAS, which are commonly used for statistical analysis. R was initially designed to be used in academic and

research contexts, but it has subsequently achieved signiﬁcant use across a variety of disciplines and sectors.

In this paper, we will explore the challenges that beginners face when learning R, and provide tips and

resources to help make the learning process more manageable and rewarding.

During my fourth semester of undergraduate course, I got introduced to R programming, amidst the covid,

learning a new programming language through online classes was challenging and diﬃcult. But the features of

“R” struck me because never in my life thought that a programming language could draw statistical visualization

based on your needs. This made me curious and started to explore but all these build-ups ended in the same

semester itself and moved on thinking that I would never ever come across it. After graduation, I had some

time to prepare for the next move of my career. So as to build my skills I joined AIRIS4D, under Dr. Sajeeth

Ninan Philip I got reintroduced to “R” and then the journey started. So, along the way, I got exposed to

diﬀerent packages, functions and resources of R. From the knowledge gained, I have summed up all into the

next paragraph.

Firstly, installing R and R studio, latest version of R can be downloaded from the CRAN(Comprehensive R

Archive Network) website whereas RStudio can be downloaded from posit website. The Integrated development

Environment, RStudio is a user-friendly programming platform that simpliﬁes script writing, helps you manage

your environment and working directory, and makes it easier to access and interact with ﬁles on your computer.

It also provides user-friendly graphics capabilities. There are hundreds of websites that can help you learn

the language and as a beginner I started with TechVidvan R tutorial – “Coz you wished to be a data scientist”

and GeeksforGeeks. From there you can learn about the basics, fundamentals, variables, Data types, Data

Visualizations and Machine Learning with R.

As for a beginner I ﬁnd RStudio the best way to start learning. In RStudio, we can write code inside

scripts which are called RScripts. A program in R is made up of three things: Variables, Comments, and

Keywords. Working with variables require operators and R have Arithmetic, Logical, Relational, Assignment

and Miscellaneous operators. R has several built-in data structures that are useful for storing and manipulating

data, including vectors, matrices, arrays, data frames, and lists. For further details refer the website mentioned

above. Moving to data analysis sector R can be used eﬃciently and eﬀectively.

As for a data analyst data cleaning is a crucial part of the analytical process. Using basic tools of R can

be intimidating for any beginner to data analysis, so to replace all those hurdles R have numerous packages and

one of them is “tidyverse” package is an opinionated collection of R packages designed for data science. All

packages share an underlying design philosophy, grammar, and data structures where you can ﬁnd packages like

“ggplot2”, “dplyr”,”readr”,”purr”, “tibble”.

Figure 1.1: Packages of Tidyverse(Source: Tidyverse Website)

You can install tidyverse package with:

install.packages (“tidyverse”).

One of my commonly used packages from tidyverse are “dplyr” and “ggplot2”. I ﬁnd ’dplyr’ particularly

useful because it provides a set of simple, eﬃcient functions for manipulating data.

Some of the basic functions used in dplyr are:

mutate() adds new variables that are functions of existing variables

select() picks variables based on their names.

ﬁlter() picks cases based on their values.

summarise() reduces multiple values down to a single summary.

arrange() changes the ordering of the rows.

Example:

#>starwars

select(name,ends with(”color”))

#> name hair color skin color eye color

#> <chr> <chr> <chr> <chr>

#> 1 Luke Skywalker blond fair blue

#> 2 C-3PO <NA> gold yellow

#> 3 R2-D2 <NA> white, blue red

#> 4 Darth Vader none white yellow

#> 5 Leia Organa brown light brown

starwars

desc(mass)

1 Jabba De. . . 175 1358 <NA> green-tan. . . orange 600 herm. . . mascu. . .

#> 2 Grievous 216 159 none brown, wh. . . green, y. . . NA male mascu. . .

#> 3 IG-88 200 140 none metal red 15 none mascu. . .

#> 4 Darth Va. . . 202 136 none white yellow 41.9 male mascu. . .

#> 5 Tarﬀul 234 136 brown brown blue NA male mascu. . .

You can visit the website to gather more information on dplyr

Secondly, ”ggplot2” is a system for declaratively creating graphics, based on The Grammar of Graphics. It’s

hard to succinctly describe how ggplot2 works because it embodies a deep philosophy of visualisation. However,

in most cases you start with ggplot(), supply a dataset and aesthetic mapping (with aes()). You then add on

layers (like geom point() or geom histogram()), scales (like scale colour brewer()), faceting speciﬁcations (like

facet wrap()) and coordinate systems (like coord ﬂip()).

If you are new to ggplot2 you are better oﬀ starting with a systematic introduction, rather than trying to

learn from reading individual documentation pages. Currently, there are three good places to start:

The Data Visualisation and Graphics for communication chapters in R for Data Science. R for Data

Science is designed to give you a comprehensive introduction to the tidyverse, and these two chapters

will get you up to speed with the essentials of ggplot2 as quickly as possible.

If you’d like to take an online course, try Data Visualization in R With ggplot2 by Kara Woo.

If you’d like to follow a webinar, try Plotting Anything with ggplot2 by Thomas Lin Pedersen.

If you want to dive into making common graphics as quickly as possible, I recommend The R Graphics

Cookbook by Winston Chang. It provides a set of recipes to solve common graphics problems.

Example:

ggplot(mpg, aes(displ, hwy, colour = class)) +

geom point()

Figure 1.2: A scatter plot between total volume of the cylinders in the engine(cc) vs highway mileage (Source:

Tidyverse Website

Tidyverse is built around the idea of tidy data, where each row represents a single observation and each

column represents a single variable. This allows for easy subsetting and reshaping of data, as well as the use of

a wide range of functions for data manipulation and visualization. Overall, the tidyverse provides a powerful set

of tools for working with data in R, and has become a popular choice for data scientists and statisticians around

the world.

In addition to the tools for data manipulation and visualization provided by the tidyverse, the Shiny package

allows users to easily create interactive web applications with R. Shiny is built on top of the tidyverse and

makes it easy to create dynamic, reactive applications that can be used to explore and analyze data in real time.

With Shiny, users can build applications that allow others to interact with their data and analysis, making it an

important tool for data communication and collaboration. Shiny applications are built using a combination of

R code and HTML, CSS, and JavaScript. The Shiny package provides a set of functions and tools that make it

easy to build interactive applications, including functions for creating user interfaces, handling user input, and

building reactive components that automatically update in response to changes in the data or user input.

Shiny apps are contained in a single script called app.R. The script app.R

lives in a directory (for example, newdir/) and the app can be run with

runApp(”newdir”).

app.R has three components:

a user interface object

a server function

a call to the shinyApp function

The user interface (ui) object controls the layout and appearance of your app. The server function contains

the instructions that your computer needs to build your app. Finally, the shinyApp function creates Shiny app

objects from an explicit UI/server pair.

Example:

Figure 1.3: A simple R shiny App with a histogram of R’s ”faithful” dataset with a conﬁgurable number of

bins. (Source: R Shiny Website)

Figure 1.4: Recently I was working with R shiny package to create a plot based on COVID-19 datasets. This

app has a drop-down feature which helps to switch between two bar plots.(Source: Nevin Thomas)

In summary, Shiny is a valuable addition to the R ecosystem that allows users to create interactive web

applications with ease. Whether you are looking to build a dashboard, data visualization, or other type of

interactive application, Shiny has the tools and features you need to get the job done. If you haven’t already, we

encourage you to give Shiny a try and see for yourself how it can help you build interactive data applications

with R.For more information about Shiny and how it can be used to create interactive web applications with R,

visit the Shiny website (Shiny)

Mastering R can be a challenging task for beginners, but it is also a rewarding and valuable skill to have.

While there are many concepts and techniques to learn, the beneﬁts of using R for data analysis and visualization

are well worth the eﬀort. To succeed as an R beginner, it is important to be persistent, ask questions, and seek

out resources and support when needed. With practice and determination, anyone can master the fundamentals

of R and begin using it to solve real-world problems.

References

1. KThe oﬃcial intro, ”An Introduction to R”, available online in HTML and PDF

2. Statistical Inference via Data Science: A ModernDive into R and the tidyverse by Chester Ismay and

Albert Y. Kim.

3. Solutions and notes for R4DS by Jeﬀrey B. Arnold.

4. Mastering Shiny by Hadley Wickham

About the Author

Nevin Thomas is a student at AIRIS 4D, where he is studying Data Science, specializing in the R

programming language and using it to create data visualizations and perform data analysis. Recently, he has

been focused on working with COVID-related datasets, using his skills to explore and understand the impact of

the pandemic.

Part V

Fiction

Curves and Curved Trajectories

by Ninan Sajeeth Philip

airis4D, Vol.1, No.1, 2023

www.airis4d.com

This article is a thought experiment on a virtual world with imaginary creatures capable of doing ﬁctitious

activities.

Characters:

1. Dot: a zero-dimensional creature

2. Angel the Light: a one-dimensional Angel

3. Mittu: a two-dimensional ant

1.1 Episode 1

Meet Mr Dot, a dimensionless species pasted to a cross ﬁxed at the centre of a black box, completely

isolated from the rest of the world. The box is tied to a string that can be pulled along a ﬁxed line. A train

engine is attached to the string that pulls the string along a straight line. Since the dimensionless creature has no

sense of dimensions, it would always consider itself at rest at a ﬁxed point inside the box on the cross. However,

due to its constrained movement along the straight line, it will experience an initial jerk that gradually vanishes

oﬀ. In fact, with no sense of dimension, the creature won’t realise the backward nor forward directions and

hence would articulate the experience as an invisible entity, a demon, with some psychological impact on him.

Though it does not recognise the direction, it recognises the change in the jerk and hence develops the notion

of time. According to Dot, the unit for time is the interval between two impacts that corresponds to the starting

and stopping jerks produced by the engine.

One night, an angel from one dimension, a beam of light, appears to Dot in a dream and informs it of the

existence of dimensions and the possibility of an extra degree of movement. Though Dot is unable to visualise

the existence of such an entity, it becomes apparent that if what the angel Light told is correct, the ﬁctitious

experience it had can be represented by a force that can be related to the rate of change of movement along that

imaginary degree of freedom. However, with no sense of dimension, it becomes impossible for Dot to judge

whether the engine is pushing or pulling the box. The only logical interpretation he has is that this force is

always positive or unidirectional, emphasising the notion that time is unidirectional.

1.2 Episode 2

Mr Dot and Angel the Light used to have regular conversations and debates for several months. One day,

a stranger untied the box from the engine and tied it to a turning wheel that turns about a ﬁxed point. Since a

dimensionless creature has no sense of dimensions, Mr Dot would always consider himself idling at the ﬁxed

point inside the box on the cross. However, due to the constrained rotation of the box about a ﬁxed point, Dot

could experience an unknown stress within.

In fact, with no sense of dimension, Dot could only articulate his experience as an invisible entity, a demon,

with some psychological inﬂuence on him. The striking diﬀerence it had compared to the previous experience

was the persisting constancy of the torture instead of the systematic torture he had experienced earlier.

Dot shared his new experience of constant torture with Light. He expected that Light would come up with

a proper explanation. However, Light aﬃrmed the experience of the strange constant torture that has also been

troubling him.

While they are still debating and exploring, a six-legged species with two horns appear in the dream

and introduces itself as Mittu, an ant who has experienced an extra dimension in addition to what Light had

conﬁrmed. Mittu clariﬁed that it is because of their rotational motion along a surface that imposed a constraint

on the string attached to a ﬁxed point on that surface that was responsible for the strange experience. According

to Mittu, they are experiencing the centripetal force, and no demon exists. Since Dot nor Light could visualise

the argument made by Mittu, they responded to it as an unrealistic model of their world.

Mittu narrated the inﬁnite possibilities to move along this extra dimension that would result in a lot more

other psychological experiences, some of which could be radically diﬀerent from the constant torture they

experience from the constrained rotation about a ﬁxed point. Mittu then told them about some discontinuities

in that world called “wormholes”. If one accidentally crosses its event horizon, they disappear from this world

forever and enter a new world. The new world is similar but would be inhabited by a completely diﬀerent

population of ants. A species that thus gets trapped in the wormhole can, however, return to its previous world

by going through another “wormhole”.

Having heard Mittu this far, Dot and Light became convinced that he is either cooking up stories or is

out of his mind. This new dimension is a fascinating concept to explain the ﬁctitious tortures they undergo.

However, instead of solving it, Mittu has told them even more ﬁctitious entities like a world and the existence

of many worlds! To make it even more incomprehensible, Mittu has told them about wormholes that allow it to

traverse the Worlds. Impossible! Unanimously declared both of them.

According to Dot, both the angel and the ant are liars. Light is a ﬁrst-order liar speaking about an impossible

entity called dimension and moving along it as the cause for the systematic torture. Light argued that he and

Mr Dot speak the truth while the ant is a liar. He tried to convince Dot that zero dimension is a special case,

an instant of the one-dimensional world. One-dimension is an inﬁnite stretch of zero-dimensional points from

minus inﬁnity to plus inﬁnity. Dot laughed in return, arguing dimension itself is a ﬁctitious concept, inﬁnity is

imaginary, and making arbitrary additions to his universe is impossible.

The ant was rather pious throughout the argument, aﬃrming that both are correct, except that their

understanding is incomplete. He explained that the ﬁrst dimension is an inﬁnite stretch of zero-dimensional

entities, two dimensions mean an inﬁnite stretch of one-dimensional entities along a second dimension and the

resulting entity is called a surface.

Although Light could see a point in the narration of the ant, it could not visualise how to extend one

dimension to get anything with a higher dimension. For him, adding a line to another line would only produce

an overlapping line. How can it become another entity, as claimed by the ant?

1.2 Episode 2

Figure 1.1: Mittu’s explanation for the constant torture experienced by others.

Mittu realised that it would be impossible to convince Light with simple “logical” arguments. You should

generalise on your logic, instructed Mittu. Extend the argument you shared with Dot to generate a dimension by

combining an inﬁnite chain of dimensionless entities. A surface, similarly, is an inﬁnite array of one-dimensional

lines put side by side from minus inﬁnity up to plus inﬁnity.

Mind-blowing, said Light.

Well, said Mittu. You know that dimensionless entities are the building blocks of lines.

Light nodded in agreement. Mittu continued. The same dimensionless entities form the building block of

everything, including the surface. When you reduce a line along the length of your axis, you get a dimensionless

point. Likewise, you will get a line when you reduce a surface along its invisible axis.

Okay, that’s interesting, said Light.

A line allows you to move forward or backwards only, said Mittu.

Yes, said Light. However, Dot had no clue where the discussion was heading.

Since you can move forward and backwards along a line, continued Mittu, if you have another line along

the set of inﬁnite lines on the surface, you can move forward and backwards in that direction. However, your

location along the other dimension need not change, and your one-dimensional friends would say that you are

not moving!.

Wow! That’s too much to handle with my brain, said Light. Mittu smiled and continued. You can put

an inﬁnite number of lines in the direction of each dimensionless point on your line along the direction of the

invisible dimension.

Okay, said Light. Mittu paused for a while, took a deep breath and continued. There is no need to put all

lines in the exact direction of the invisible dimension. Some of them can start from a point on your line and end

at a diﬀerent point. You know that when a string along your dimension pulls you, you experience a foce and

Dot experiences some torture. The torture Dot experience and the force you experience sync with the engine’s

start and stop. Since you can visualise the direction of the train, and you know it is when the train starts and

stops, for you, it is all well understood. In the case of our dimensionless friend Dot, since he has no sense of this

dimension, it is just torture that appears periodically with a diﬀerence in the polarity of the torture. Dot might

also realise that a negative polarity will always complement a positive polarity or that they appear in pairs. But

for you who can visualise the forward and backward movement of the train, it is just the consequence of the

constraint to move along the line. Yes, Light agreed.

1.2 Episode 2

This, you call the degree of freedom that our dimensionless friend Dot can’t visualise. Mittu posed for a

minute to let Light think it over.

Now, look under your foot, Mittu suggested. Both Dot and Light looked down. “Nothing here” exclaimed

Dot while Light exclaimed, “Oh my God! The land below me is moving!” Mittu smiled and told Light that it

is not the land, but he himself is moving. “Me?” exclaimed Light. “According to the equations of motion, I

should be exerting force in the direction of motion to move. I haven’t spent any. How can then ?”

“Don’t panic”, told Mittu. “You are moving because the box on which you and Dot are sitting is tied to a

string and is going round.”

“Going round? Nonsense”, said Light. “We are going straight! What is round?”

“Oh yes”, Mittu agreed. “Because you can visualise only one dimension, every path appears as a straight

line for you, and we call it the geodesic”

”But there are only straight lines, whatever you call it:, aﬃrmed Light. Mittu just smiled and continued,

“Whether you believe it or not, the torture you experience, which we call the centripetal force, is always

proportional to the speed you move relative to the ground.”

“Weird!” exclaimed Light. “The torture is the penalty I pay for this free trip!” He said sarcastically.

Everyone laughed. Dimensionless torture for a free ride! After all, nothing comes for free!

[Will be continued. . . .]

About the Author

Professor Ninan Sajeeth Philip is a Visiting Professor at the Inter-University Centre for Astronomy

and Astrophysics (IUCAA), Pune. He is also an Adjunct Professor of AI in Applied Medical Sciences [BCMCH,

Thiruvalla] and a Senior Advisor for the Pune Knowledge Cluster (PKC). He is the Dean and Director of Artiﬁcial

Intelligence Research and Intelligent Systems and has a teaching experience of 33+ years in Physics. His area

of specialisation is AI and ML.

About airis4D

Artiﬁcial Intelligence Research and Intelligent Systems (airis4D) is an AI and Bio-sciences Research Centre.

The Centre aims to create new knowledge in the ﬁeld of Space Science, Astronomy, Robotics, Agri Science,

Industry, and Biodiversity to bring Progress and Plenitude to the People and the Planet.

Vision

Humanity is in the 4th Industrial Revolution era, which operates on a cyber-physical production system. Cutting-

edge research and development in science and technology to create new knowledge and skills become the key to

the new world economy. Most of the resources for this goal can be harnessed by integrating biological systems

with intelligent computing systems oﬀered by AI. The future survival of humans, animals, and the ecosystem

depends on how eﬃciently the realities and resources are responsibly used for abundance and wellness. Artiﬁcial

intelligence Research and Intelligent Systems pursue this vision and look for the best actions that ensure an

abundant environment and ecosystem for the planet and the people.

Mission Statement

The 4D in airis4D represents the mission to Dream, Design, Develop, and Deploy Knowledge with the ﬁre of

commitment and dedication towards humanity and the ecosystem.

Dream

To promote the unlimited human potential to dream the impossible.

Design

To nurture the human capacity to articulate a dream and logically realise it.

Develop

To assist the talents to materialise a design into a product, a service, a knowledge that beneﬁts the community

and the planet.

Deploy

To realise and educate humanity that a knowledge that is not deployed makes no diﬀerence by its absence.

Campus

Situated in a lush green village campus in Thelliyoor, Kerala, India, airis4D was established under the auspicious

of SEED Foundation (Susthiratha, Environment, Education Development Foundation) a not-for-proﬁt company

for promoting Education, Research. Engineering, Biology, Development, etc.

The whole campus is powered by Solar power and has a rain harvesting facility to provide suﬃcient water supply

for up to three months of drought. The computing facility in the campus is accessible from anywhere through a

dedicated optical ﬁbre internet connectivity 24×7.

There is a freshwater stream that originates from the nearby hills and ﬂows through the middle of the campus.

The campus is a noted habitat for the biodiversity of tropical Fauna and Flora. airis4D carry out periodic and

systematic water quality and species diversity surveys in the region to ensure its richness. It is our pride that

the site has consistently been environment-friendly and rich in biodiversity. airis4D is also growing fruit plants

that can feed birds and provide water bodies to survive the drought.