Cover page

Image Name: X1.2 Solar Flare on May 14,2024

Image Credit: NASA/SDO

NASA’s Solar Dynamics Observatory captured this image of an X1.2 solar ﬂare peaking at 8:55 a.m. EDT on

May 14, 2024. The image shows a subset of extreme ultraviolet light that highlights the extremely hot material

in ﬂares created from a mixture of SDO’s AIA 171 and 131 channels.

For more information read : https://svs.gsfc.nasa.gov/14589

Managing Editor Chief Editor Editorial Board Correspondence

Ninan Sajeeth Philip Abraham Mulamoottil K Babu Joseph The Chief Editor

Ajit K Kembhavi airis4D

Geetha Paul Thelliyoor - 689544

Arun Kumar Aniyan India

Sindhu G

Journal Publisher Details

Publisher : airis4D, Thelliyoor 689544, India

Website : www.airis4d.com

Email : nsp@airis4d.com

Phone : +919497552476

Editorial

by Fr Dr Abraham Mulamoottil

airis4D, Vol.2, No.6, 2024

www.airis4d.com

This editions starts with Blesson George’s article

“Self Attention Models”. It highlights the transforma-

tive impact of self-attention mechanisms in natural lan-

guage processing (NLP). Self-attention enhances input

embeddings by incorporating context, crucial for tasks

like machine translation and text summarization. It

allows models to dynamically weigh the importance

of diﬀerent sequence elements, improving interpreta-

tion and generation of text. The Transformer model,

leveraging self-attention, revolutionized NLP by over-

coming the limitations of Recurrent Neural Networks

(RNNs), such as sequential processing and diﬃculty

with long-range dependencies. Transformers process

all sequence elements simultaneously, enabling faster

and more eﬃcient training and inference on modern

hardware. Key components include embedding in-

put sentences, using positional encodings, and deﬁn-

ing weight matrices for projecting input embeddings

into queries, keys, and values. Transformers’ ability

to capture context and scale eﬃciently has led to their

widespread adoption, setting new standards in NLP

and rendering RNNs largely obsolete.

”Transforming the Landscape of ML Using Trans-

formers” by Linn Abraham explains how Transformers

have revolutionized machine learning (ML), particu-

larly natural language processing (NLP). Transform-

ers, such as those used in Chat-GPT, leverage self-

attention mechanisms to process data eﬃciently, en-

abling advancements in tasks like machine translation,

text summarization, and sentiment analysis. Trans-

formers outperform earlier models like LSTMs and

GRUs by processing all sequence elements in parallel,

avoiding the sequential limitations of previous meth-

ods. They utilize attention blocks that compute re-

lationships between all input tokens simultaneously,

enhancing the model’s ability to capture context and

dependencies within the data. The architecture’s scal-

ability and eﬃciency have broadened ML applications,

reducing the need for labeled data and making large-

scale training feasible. This has signiﬁcantly advanced

AI capabilities, exempliﬁed by technologies like Chat-

GPT, which can generate coherent and contextually rel-

evant text based on learned patterns from vast datasets.

The article ”X-ray Astronomy: Through Mis-

sions” by Aromal P outlines the evolution of X-ray

astronomy through various satellite missions from the

1950s to the 1970s. Early eﬀorts by the USA with Van-

guard 3 and Explorer 7 failed to detect cosmic X-rays

due to interference from the Van Allen belts. Signiﬁ-

cant progress began with the launch of Uhuru in 1970,

the ﬁrst successful satellite to observe cosmic X-rays,

leading to the discovery of numerous X-ray sources

and diﬀuse X-ray emissions from galaxy clusters. The

1970s saw further advancements with missions like

Copernicus, ANS, Ariel V, and HEAO-1, which con-

tributed to the understanding of X-ray sources and phe-

nomena, including pulsars and X-ray bursts. Each mis-

sion brought incremental improvements in sensitivity

and range, paving the way for future X-ray astronomy

advancements. Despite these successes, imaging X-ray

sources remained a challenge due to the high energy of

X-ray photons, a problem addressed by later missions

in the 1980s.

The article Exploring Stellar Clusters: Insights

from Color-Magnitude Diagrams, Part-1” by Sindhu

G discusses the importance of color-magnitude dia-

grams (CMDs) in studying star clusters. CMDs plot

the apparent magnitude of stars against their color in-

dex, similar to Hertzsprung-Russell diagrams (HRDs),

oﬀering insights into the composition, age, and evo-

lutionary stages of stars within clusters. Star clusters,

assumed to form simultaneously from the same inter-

stellar gas cloud, provide a uniform sample for studying

stellar properties, with mass being the primary vari-

able. Globular clusters, which are among the oldest

structures in the universe, show distinct features on HR

diagrams, such as the absence of hot, massive stars, a

prominent red giant branch, and a horizontal branch.

These features reﬂect the advanced age of globular

clusters. However, constructing accurate HR diagrams

for these clusters is challenging due to issues like stel-

lar crowding, distance determination, and metallicity

variations. The article emphasizes the value of HR

diagrams in reﬁning our understanding of stellar evo-

lution and the ages of clusters and the Galaxy. The

next part will delve deeper into HR and CMD analyses

of various globular clusters.

Precision medicine, also known as personalized

medicine, tailors healthcare to individuals based on

their unique genetic, protein, and other bodily charac-

teristics. It aims to improve treatment eﬀectiveness by

targeting the underlying causes of disease, leading to

more personalized and eﬀective treatments with fewer

side eﬀects compared to traditional medicine. In can-

cer care, precision medicine involves analyzing how

changes in genes or proteins within cancer cells aﬀect

treatment options. By identifying speciﬁc mutations

driving a patient’s cancer, doctors can customize treat-

ments to target those mutations, potentially achieving

better outcomes. Precision medicine is particularly

valuable in cancers like colorectal, breast, and lung

cancers, as well as certain leukaemias, lymphomas,

and melanomas. Precision medicine not only treats

existing cancers but also plays a crucial role in cancer

prevention for high-risk individuals. Genetic testing

can identify inherited mutations that elevate cancer

risk, leading to earlier and more frequent screenings

to catch cancer early. Additionally, medications or

lifestyle changes may be recommended to reduce can-

cer risk proactively. Two primary treatments used in

precision medicine are targeted drug therapy and im-

munotherapy. Targeted therapy identiﬁes and attacks

speciﬁc cancer cells precisely, while immunotherapy

boosts the immune system’s ability to ﬁnd and attack

cancer cells. However, precision medicine in cancer

care faces limitations, including unequal access to ad-

vanced testing and therapies, limited integration into

cancer care, and concerns about aﬀordability and in-

surance coverage for biomarker testing and targeted

therapies. ”Precision Medicine and its Role in Cancer

Care” by Geetha Paul concudes that precision medicine

holds signiﬁcant promise for improving cancer patient

outcomes, but ongoing research and eﬀorts to address

limitations are needed to fully realize its potential in

cancer care.

The article ”Exploring Varied Approaches for DNA

Methylation Detection” by Jinsu Ann Mathew delves

into the methods used to study DNA methylation, an es-

sential epigenetic modiﬁcation crucial for understand-

ing various biological processes and diseases. It high-

lights four major techniques: DNA Methylation Mi-

croarray, Whole-Genome bisulphite Sequencing ( WGBS

), Methylated DNA Immunoprecipitation Sequencing

( MeDIP-seq ), and Reduced Representation bisulphite

Sequencing (RRBS). DNA Methylation Microarrays,

like Illumina’s BeadArrays, act as ﬁngerprint scanners

for DNA methylation patterns across the genome, of-

fering high throughput and cost-eﬀectiveness. WGBS

provides the highest resolution, mapping methylation at

single-nucleotide resolution across the entire genome.

MeDIP-seq enriches and sequences methylated DNA

regions with high speciﬁcity, while RRBS focuses on

speciﬁc genome regions with high resolution and re-

duced complexity. Each technique has its advantages

and limitations, catering to diﬀerent research needs and

resource availability. Microarrays are cost-eﬀective

and standardized, WGBS oﬀers comprehensive cov-

erage, MeDIP-seq provides targeted enrichment, and

RRBS balances cost and detail. Choosing the right

method depends on the research question and avail-

able resources. By understanding the strengths and

limitations of each technique, researchers can unlock

the secrets hidden within DNA methylation patterns,

furthering our understanding of gene regulation, de-

iii

velopment, disease, and epigenetics.

”How to Think Like a Product Manager in Academia”

by Arun Aniyan draws parallels between the role of

a product manager in industry and the approach re-

searchers in academia can take in selecting and devel-

oping research projects. It emphasizes the importance

of starting with the question ”why” to understand the

purpose and direction of the project, akin to identify-

ing the pain points or gaps in a market for a product.

Researchers are encouraged to engage in a detailed lit-

erature review to identify novel research problems and

ensure that their work extends scientiﬁc understanding.

The article suggests adopting a phased approach to re-

search, similar to product development cycles, allowing

for incremental improvements and more frequent pub-

lications. Furthermore, active engagement with stake-

holders or customers in academia, such as advisors or

the scientiﬁc community, is advocated to ensure align-

ment with project goals and avoid wasted eﬀorts. The

article concludes by highlighting the potential beneﬁts

of applying product management principles to scien-

tiﬁc research, ultimately improving the relevance and

impact of research outcomes.

Contents

Editorial ii

I Artiﬁcial Intelligence and Machine Learning 1

1 Self Attention Models 2

1.1 The Superiority of Transformer Models: Revolutionizing NLP and Rendering RNNs Obsolete . 2

1.2 Blocks of Self Attention Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 Transforming the landscape of ML using Transformers 6

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2.2 The Role of Auto-Encoders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2.3 Does Chat-GPT really understand human language? . . . . . . . . . . . . . . . . . . . . . . . . . 6

2.4 Understanding Transformers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.5 The Attention Block . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

II Astronomy and Astrophysics 9

1 X-ray Astronomy: Through Missions 10

1.1 Altius, Tempus: Ages of Satellites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1.2 Satellites in 1970s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2 Exploring Stellar Clusters: Insights from Color-Magnitude Diagrams, Part-1 15

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.2 Star Clusters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.3 The Hertzsprung-Russell Diagram of Globular Clusters . . . . . . . . . . . . . . . . . . . . . . . 16

2.4 Challenges in Constructing HR Diagrams for Globular Clusters . . . . . . . . . . . . . . . . . . . 17

III Biosciences 18

1 Precision Medicine and its Role in Cancer Care 19

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

1.2 Precision Medicine and Gene Changes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

1.3 Gene Changes and Cancer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

1.4 Precision Medicine in Cancer Care: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

1.5 Types of Cancer where Precision Medicine is Used: . . . . . . . . . . . . . . . . . . . . . . . . . 21

1.6 Precision Medicine and Cancer Care Prevention for High Risk Individuals . . . . . . . . . . . . 21

1.7 Limitations to Precision Medicine in Cancer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2 Exploring Varied Approaches for DNA Methylation Detection 23

2.1 DNA Methylation Microarray . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

CONTENTS

2.2 WGBS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

2.3 MeDIP-seq . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

2.4 RRBS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

IV General 27

1 How to Think Like a Product Manager in Academia 28

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

1.2 Why is the Product or Feature Required . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

1.3 Problem Identiﬁcation and Deﬁnition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

1.4 Incremental Improvements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

1.5 Active Customer Engagement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

1.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

Part I

Artiﬁcial Intelligence and Machine Learning

Self Attention Models

by Blesson George

airis4D, Vol.2, No.6, 2024

www.airis4d.com

In previous discussions, we explored the two types

of attention networks: local versus global attention and

hard versus soft attention. Each type has its unique

characteristics and applications. However, the most

important and impactful type of attention network is

the self-attention network.

Self-attention is a mechanism that enhances the

information content of an input embedding by incor-

porating information about the input’s context. In other

words, the self-attention mechanism enables the model

to weigh the importance of diﬀerent elements in an

input sequence and dynamically adjust their inﬂuence

on the output. This is especially important for lan-

guage processing tasks, where the meaning of a word

can change based on its context within a sentence or

document. By leveraging self-attention, each input el-

ement is not only represented by its standalone mean-

ing but is also augmented with contextual informa-

tion from the entire sequence, allowing the model to

understand nuances and relationships that are critical

for accurate interpretation and generation of text. The

mechanism computes attention scores that dynamically

determine how much focus each word should place on

every other word in the sequence, meaning that words

that are more contextually relevant to the current word

will have a higher inﬂuence on its ﬁnal representation.

Self-attention excels at resolving ambiguities by using

the context provided by surrounding words to ﬁne-tune

the interpretation of each word. It can directly link each

word to every other word in the sequence, regardless

of their distance, capturing long-range dependencies

crucial for understanding complex sentences and docu-

ments. Unlike sequential models, self-attention allows

for parallel processing of input sequences, speeding

up training and inference, and making it feasible to

handle very large datasets eﬃciently. Its versatility is

demonstrated across various language processing tasks

such as machine translation, text summarization, and

question answering. The introduction of self-attention

has signiﬁcantly advanced the state-of-the-art in natu-

ral language processing, with models like BERT, GPT,

and their variants leveraging this mechanism to achieve

superior performance on a wide range of benchmarks,

illustrating its profound impact on the ﬁeld.

1.1 The Superiority of Transformer

Models: Revolutionizing NLP

and Rendering RNNs Obsolete

The Transformer model revolutionized natural lan-

guage processing by addressing key limitations in-

herent in Recurrent Neural Networks (RNNs). Un-

like RNNs, which process sequences sequentially and

struggle with long-range dependencies due to the van-

ishing gradient problem, Transformers use a self-attention

mechanism that allows them to process all elements of

a sequence simultaneously. This parallelization signif-

icantly speeds up training and inference, making Trans-

formers more eﬃcient on modern hardware like GPUs

and TPUs. Furthermore, the self-attention mechanism

enables Transformers to model relationships between

all sequence elements directly, regardless of their dis-

tance, thus capturing long-range dependencies more

eﬀectively than RNNs. Additionally, the Transformer

architecture, with its modular design and reliance on

1.2 Blocks of Self Attention Models

attention mechanisms rather than recurrence, is inher-

ently more scalable. It can be easily scaled up by

increasing the number of layers or the size of each

layer, leading to the development of large-scale mod-

els like BERT and GPT-3. This scalability, combined

with the ability to learn richer contextual representa-

tions through attention, makes Transformers superior

for many natural language processing tasks. The sim-

plicity of the Transformer architecture, which eschews

the complexity of RNNs’ gating mechanisms in fa-

vor of straightforward attention and feed-forward lay-

ers, further contributes to its eﬀectiveness and ease

of implementation. These advantages have led to the

widespread adoption of Transformers, eﬀectively re-

placing RNNs in many applications and marking a sig-

niﬁcant advancement in the ﬁeld of NLP.

1.2 Blocks of Self Attention Models

1.2.1 Embedding an Input Sentence

Embedding an input sentence is a crucial step in

natural language processing (NLP) that involves con-

verting words or tokens in the sentence into dense vec-

tors of real numbers, which can then be processed by

machine learning models. This process allows models

to handle textual data in a numerical format, facilitating

various NLP tasks like translation, sentiment analysis,

and text generation. Here are diﬀerent methods through

which embeddings operate:

Tokenization: The ﬁrst step in embedding an

input sentence is to break the sentence into in-

dividual tokens. These tokens can be words,

subwords, or even characters, depending on the

speciﬁc tokenization strategy used.

Mapping to Vectors: Once the sentence is to-

kenized, each token is mapped to a dense vec-

tor. This mapping is typically learned during

the training of the embedding model. Each vec-

tor represents the token in a high-dimensional

space, capturing semantic meaning and contex-

tual information.

Pre-trained Embeddings: Many NLP models

use pre-trained embeddings, such as Word2Vec,

GloVe, or more advanced contextual embed-

dings like those generated by BERT or GPT.

These embeddings are learned from large cor-

pora and can capture complex linguistic patterns

and relationships between words.

Positional Encodings: In models like Transform-

ers, positional encodings are added to the to-

ken embeddings to retain the order of the to-

kens in the sentence. Since Transformers do not

inherently capture positional information, these

encodings are crucial for understanding the se-

quence structure.

1.2.2 Deﬁning the Weight Matrices

In the Transformer architecture, the self-attention

mechanism, speciﬁcally the scaled dot-product atten-

tion, is fundamental. This mechanism relies on three

crucial weight matrices: W

, W

, and W

. These ma-

trices are essential for projecting the input embeddings

into diﬀerent subspaces for queries, keys, and values

respectively, and they are ﬁne-tuned during the training

process.

Each input token x(i), where i is the position in-

dex in the input sequence of length T , is transformed

into query, key, and value vectors through matrix mul-

tiplication with these weight matrices:

Query sequence: q(i) = W

⋅ x(i) for each i ∈

[1, T ]

Key sequence: k(i) = W

⋅ x(i) for each i ∈

[1, T ]

Value sequence: v(i) = W

⋅ x(i) for each i ∈

[1, T ]

The dimensions of these matrices are tailored to ﬁt

the embedding dimensions and the desired complexity

of the model: W

and W

both have dimensions of d

d, which reﬂects their role in mapping input vectors x of

dimension d to the query and key vectors of dimension

. - W

has dimensions of d

×d, mapping the input

vectors to value vectors of dimension d

After deﬁning the weight matrices and project-

ing the input tokens into query, key, and value vectors,

the next crucial step in the Transformer’s self-attention

mechanism is computing the unnormalized attention

1.3 Conclusion

weights, often followed by generating the attention

scores. This process helps to determine how much

focus or ”attention” each part of the input sequence

should receive relative to others. Here’s how it is ac-

complished:

1.2.2.1 Compute Unnormalized Attention

Weights

The unnormalized attention weights are calcu-

lated by performing a dot product between the query

vector and all key vectors. This operation assesses the

degree of alignment or similarity between the query

and each part of the input sequence. The result is a set

of scores that indicates how much each element of the

sequence should be attended to in relation to the query.

For a given query q(i) and all keys (k(1), k(2),

..., k(T )), the dot products are calculated as follows:

scores(i) = q(i)⋅k(1)

, q(i)⋅k(2)

, ..., q(i)⋅k(T )

This results in a vector of scores for each query,

where each score corresponds to a token in the input

sequence.

1.2.2.2 Scale the Attention Scores

To ensure that the softmax function operates in a

region where it has gentle gradients (which helps with

gradient ﬂow during training), the scores are scaled

down by the square root of the dimension of the key

vectors (

√

scaled scores(i) =

scores(i)

√

1.2.2.3 Apply Softmax to Normalize the

Attention Scores

After scaling, a softmax function is applied to the

scores for each query. This step converts the raw scores

into a probability distribution, where each element in-

dicates the relative importance of the corresponding

key-value pair:

attention(i) = softmax(scaled scores(i))

The softmax function is applied across the keys

for each query, ensuring that the attention weights sum

to 1.

1.2.2.4 Compute the Output Vectors

Finally, the attention weights are used to compute

a weighted sum of the value vectors, which produces

the output vector for each position in the input se-

quence:

output(i) =

∑

j=1

attention(i, j)⋅v(j)

Each output vector is a combination of values from

across the input sequence, weighted by their computed

attention scores. This allows the model to dynamically

focus on diﬀerent parts of the input based on the context

provided by the queries and keys.

1.3 Conclusion

The Transformer model, leveraging the self-attention

mechanism, has revolutionized natural language pro-

cessing (NLP) by enabling the simultaneous processing

of sequence elements, thereby capturing long-range de-

pendencies more eﬀectively than its predecessors. This

architecture facilitates a more profound understanding

of context within sequences, which enhances perfor-

mance on tasks such as machine translation, text sum-

marization, and sentiment analysis. Through its ability

to scale and process data in parallel, the Transformer

overcomes limitations like the vanishing gradient prob-

lem common in older models such as RNNs. The

model’s eﬀectiveness and eﬃciency in handling com-

plex language tasks have made it a dominant framework

in NLP, setting new standards and paving the way for

future advancements in the ﬁeld.

References

1. Shaw, Peter, Jakob Uszkoreit, and Ashish Vaswani.

”Self-attention with relative position representa-

tions.” arXiv preprint arXiv:1803.02155 (2018).

1.3 Conclusion

2. Understanding and Coding the Self-Attention

Mechanism of Large Language Models From

Scratch

About the Author

Dr. Blesson George presently serves as

an Assistant Professor of Physics at CMS College

Kottayam, Kerala. His research pursuits encompass

the development of machine learning algorithms,

along with the utilization of machine learning tech-

niques across diverse domains.

Transforming the landscape of ML using

Transformers

by Linn Abraham

airis4D, Vol.2, No.6, 2024

www.airis4d.com

2.1 Introduction

By now you must have heard about “Transform-

ers” at least once. I am not talking about the movie

franchise, but the machine learning model that forms

part of the Chat-GPT acronym. If you don’t know,

GPT stands for Generative Pre-trained Transformer. In

this article, I aim to build the case for transformers.

How they revolutionized, not only the ﬁeld of Natu-

ral Language Processing (NLP) but the whole machine

learning landscape. One of the goals is to give you

an intuition of what “attention” blocks in Transform-

ers actually achieve. Through this I hope you also

get an intuition for how technologies like Chat-GPT

are stretching the boundaries of what AI can currently

achieve.

2.2 The Role of Auto-Encoders

A key idea that has enabled this sudden rise of ca-

pability is a class of ML models called auto-encoders.

The advantage they bring to the table is that, they are a

kind of unsupervised ML technique. Meaning that they

do not require each training sample to be associated

with a “label” that then becomes the ground truth for

teaching the machine. This is because auto-encoders

are designed to learn a model that can reproduce the

input as the output. Think of a CNN which tries to

learn how to accurately generate the input image in the

output. Each pixel of the input image itself then be-

comes the label for that pixel. Such architectures have

two parts - an encoder which forms a bottleneck by dis-

tilling the input image into a lower dimensional space.

And a decoder that tries to generate the ﬁnal output

from the output of the encoder. If trained successfully,

we expect the weights of the network to store infor-

mation about the underlying process that generated the

input data. In the context of an NLP model, the input

can be a sentence and the output can be predicting each

subsequent word from the one preceding it. The fact

that auto-encoders can learn from unlabelled data has a

huge implication, it saves machine learning engineers

from spending countless hours preparing labelled data.

Instead, since the internet is mostly a huge corpus of

text, they have at their disposal a gold mine of training

data for training the machine. The only deciding fac-

tor now would be who has the most resources to train

networks on such huge datasets.

2.3 Does Chat-GPT really

understand human language?

Chat-GPT is a product of research in Natural

Language Processing (NLP), a subdomain of Artiﬁ-

cial Intelligence. The terminology “Natural Language”

comes because there exists “programming languages”

that are man-made languages designed for humans to

communicate with machines. Let’s look at some of the

tasks that NLP models are trained to perform:

Classifying whether a movie review is positive

or negative - this is called sentiment analysis.

2.4 Understanding Transformers

Summarizing a news article in a few words - text

summarization.

Translating a sentence from English to French -

machine translation.

What should be the next word in this (incom-

plete) sentence? - text generation

Chat-GPT itself is called a Large Language Model or

LLM for short. Although, today LLMs are capable of

much more, it is helpful to focus on one of their abilities

- text generation. Let’s think of Chat-GPT and similar

chatbots as systems that can predict the next word given

a seed word or sentence. Once we have a system that

can do this, if the output along with the initial prompt

is fed back to the input we can generate the next word in

the sequence. So this is what basically happens under-

the-hood of Chat-GPT. But to make it possible to do all

those amazing things you have seen it do - it leverages

the power of the “transformer” network together with

the capacity of learning 150 billion model parameters.

2.4 Understanding Transformers

Before transformers, the dominant algorithms were

LSTMs and GRUs which were modiﬁcations of some-

thing called a Recurrent Neural Network (RNN). These

were notoriously diﬃcult to train, especially on large

datasets because of the sequential nature of their algo-

rithms. Which obviously meant they could not take ad-

vantage of the huge parallelization abilities of modern

GPUs. One of the contribution of the original paper,

that introduced the transformers, was the “Attention”

block which helped alleviate this problem. The way

they solved the problem is by introducing an “atten-

tion” block that once trained can produce outputs in

parallel for each token in an input sentence.

To understand transformers we need to recognize

what they were designed for. The transformer archi-

tecture was designed for machine translation tasks. In

the terminology of NLP, it is a kind of sequence-to-

sequence model. It employs the auto-encoder algo-

rithm that we had talked about earlier. But instead of

using LSTMs or such for the encoder-decoder part it

uses “attention” blocks and MLPs along with the usual

mix of residual connections and normalization layers

(see Fig. 1). In the process of learning to reproduce

the input, auto-encoders learn a hidden representation

of the input (feature) space, which is called the embed-

ding (vector). Thus the transformer encoder after being

trained has learnt some useful representation from the

input language. Which it then uses to do the reverse,

convert the embeddings back to a sentence, this time

to another language. At the least, it is helpful to learn

how this transformer encoder works since the encoder

can be used for other tasks such as text classiﬁcation

etc.

2.5 The Attention Block

What made existing models slow, before Trans-

formers came into the picture, was the sequential pro-

cess of processing each input token in a sentence and

updating a state variable. You can think of this state

variable as the memory of a neural network. The trans-

former bypasses the need to do sequential computations

with the “Attention” block, which you can think of as a

smaller neural network that learns three weight matri-

ces from the data, which we call Query(Q), Value(V)

and Key(K) for reasons that might become clear later

on. The Query matrix multiplied by the transpose of

the Key matrix produces a score matrix where each

element shows how much a word (token) should pay

attention to another word (token). You can think of the

word “attention” as asking the following question for

each word in a sentence - Which other words in this sen-

tence are important for understanding this word? That

is, how related each word is to every other word. The

key insight here is that, despite the fact that the Atten-

tion mechanism computes an attention score for each

pair of tokens in a sentence, it achieves parallelization

through the use of matrix operations.

We can build on this intuition in a future article

which would try to understand how the “multi-head

attention” block is diﬀerent from what we just learnt.

Also it would explore the other components like MLP

layers that together complete the transformer encoder

block.

REFERENCES

Figure 1: A sketch of the transformer encoder block.

(Image Credit:Dosovitskiy et. al.)

References

[Chollet(2021)] Franc¸ois Chollet. Deep Learning with

Python. Manning, Shelter Island, NY, second edi-

tion edition, 2021. ISBN 978-1-61729-686-4.

[Vaswani et al.] Ashish Vaswani, Noam Shazeer, Niki

Parmar, Jakob Uszkoreit, Llion Jones, Aidan N

Gomez, Lukasz Kaiser, and Illia Polosukhin. At-

tention is All you Need.

[Dosovitskiy et al.(2021)] Alexey Dosovitskiy, Lucas

Beyer, Alexander Kolesnikov, Dirk Weissenborn,

Xiaohua Zhai, Thomas Unterthiner, Mostafa De-

hghani, Matthias Minderer, Georg Heigold, Syl-

vain Gelly, Jakob Uszkoreit, and Neil Houlsby. AN

IMAGE IS WORTH 16X16 WORDS: TRANS-

FORMERS FOR IMAGE RECOGNITION AT

SCALE. 2021.

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 prediction of solar ﬂares

from images of the Sun, morphological classiﬁca-

tions of galaxies from optical images surveys and ra-

dio galaxy source extraction from radio observations.

Part II

Astronomy and Astrophysics

X-ray Astronomy: Through Missions

by Aromal P

airis4D, Vol.2, No.6, 2024

www.airis4d.com

1.1 Altius, Tempus: Ages of Satellites

Sputnik boom created a high tension in the space

race between the United States of America and the So-

viet Union, with both countries focusing on diﬀerent

science areas to achieve new milestones. Even before

the detection of cosmic X-rays, the USA launched two

satellites to study X-rays. The ﬁrst mission for such

observation was the Vanguard 3 satellite, launched

in September 1959, which was operational for three

months. The satellite’s primary payload was an ion

chamber designed by the Navel Research Laboratory

(NRL), which was supposed to conduct studies of X-

rays coming from the sun. Unfortunately, the detectors

were covered by the radiations coming from the Van

Allen belt, preventing any useful detection of X-rays

by the satellite. The second mission was the Explorer 7

launched by USA in October 1959. This also had an ion

detector developed by NRL among other detectors on

board. But it shared the same fate as that of Vanguard 3.

1.2 Satellites in 1970s

1.2.1 Uhuru

The idea of sophisticated X-ray satellites were in

plan from the early 1960s. The detection of cosmic

X-rays accelerated the eﬀorts to build such satellite.

On December 12 1970, on the 7

Independence day

of Kenya, the Small Astronomical Satellite 1 (SAS-1)

was launched from the Italian San Mariano Launch

platform oﬀ the coast of Kenya. It was the ﬁrst among

the small astronomy satellites developed by NASA.

SAS-1 was the ﬁrst satellite entirely dedicated to the

observation of X-rays coming from outside the solar

system. After the successful launch, the mission was

renamed as Uhuru by the project manager Marjorie

Townsend in honor of the hospitality and support pro-

vided by the people of Kenya. ”Uhuru” is a Swahili

term meaning freedom. Uhuru was placed on nearly

equatorial circular orbit with an apogee of around 560

km and a perigee of 520 km, having an inclination of

○

and an orbital period of 96 minutes. Uhuru’s pay-

load consisted of two sets of large area proportional

counters operating in the energy range of 2-20 keV (we

can discuss details about X-ray detectors in the com-

ing chapters). The mission operated for nearly 2 years,

providing signiﬁcant insights into the X-ray world and

motivating further exploration of X-ray universe. More

details about the Uhuru telescope are in the table.

Major achievements of Uhuru include:

Completion of the ﬁrst all-sky survey of X-rays

up to a sensitivity of 5 ∗10

−4

times the intensity

of the Crab Nebula

Discovery of the diﬀuse X-ray emission from

clusters of galaxies

Detection of 339 X-ray sources

Table 1.1: Uhuru

Instrument Set 1 Set 2

Bandpass (keV) 1.7-18 1.7-18

Eﬀective Area (cm

) 840 840

Field of view (FWHM) 0.52

○

× 5.2

○

5.2

○

× 5.2

○

Timing resolution (s) 0.192 0.384

Sensitivity (ergs cm

−2

sec

−1

) 1.5×10

−11

1.2 Satellites in 1970s

Figure 1: Marjorie Townsend discusses the SAS-

1 X-ray Explorer Satellite’s performance with Bruno

Rossi during preﬂight tests at NASA’s Goddard Space

Flight Center. Marjorie Townsend was the ﬁrst

woman to become a satellite project manager at

NASA(Credit:NASA)

1.2.2 Appollo 15 and Appollo 16

Uhuru was the ﬁrst successful satellite to record

cosmic X-rays from the outer space. After the suc-

cess of Uhuru, many telescopes and payloads were

launched into space. On July 26, 1971, Appollo 15

carried an X-ray ﬂuorescence spectrometer (XRFS) to

study the composition of the lunar surface. Later in

1972, Appollo 16 carried the same instrument to study

the moon’s surface from lunar orbit. On the missions’

return journeys, the XRFS observed the sky and con-

ducted 0.5-1 hour-long observation of already discov-

ered X-ray sources. This was quite a challenging task

at that time; even Uhuru was only able to take obser-

vations for 1-2 minutes continuously.

1.2.3 Copernicus

Next satellite to study extra-solar X-rays was the

Copernicus or the Orbiting Astronomical Observatory-

3 (OAO-3). This was a collaborative eﬀort between the

Her X-1

Cyg A

Super Cluster?

SMC

Cyg X-3

Perseus Cluster

Sco X-1

Crab

LMC

NGC 6624

Cy X-1

M31

3C273

NGC 3783

Cen X-3

Virgo Cluster

Coma cluster

Figure 2: The Map of the X-ray sky after Uhuru, ac-

cording to the fourth Uhuru catalogue.(Credit : Forman

et al. 1978)

USA and the United Kingdom (UK). Copernicus was

launched in 1972, and it was the ﬁrst multi-wavelength

satellite developed by NASA. Its main payload was

a UV telescope(Princeton Experiment Package-PEP)

developed by Princeton University. It also carried an

X-ray instrument developed by Mullard Space Science

Laboratory at the University College London, which

consisted of four X-ray detectors. Copernicus studied

several pulsars, making the ﬁrst-ever long-term obser-

vation of pulsars and other bright X-ray sources known

at that time. It also found the absorption dips in Cyg

X-1.

1.2.4 ANS

Astronomische Nederlandse Satelliet (ANS) was

a collaboration between NASA and Netherlands Insti-

tute for Space Research (NIRV). It was launched in

August 1974 and operated for around 3 years. ANS

consisted of three instruments: Ultra Violet Telescope

spectrometer (UVT), Soft X-ray Experiment (SXX)

and Hard X-ray Experiment (HXX). ANS discovered

X-ray bursts that occur in the X-ray binary systems

consisting of neutron stars. It also marked the ﬁrst

detection of X-ray ﬂares from UV Ceti and YZ Cmi

stars.

1.2.5 Ariel V

Ariel V was another joint venture by the USA

and the UK, which was launched from coasts of Kenya

in October 1974. It operated for around 6 years and

became one of the successful missions of its era. It

1.2 Satellites in 1970s

had 6 major science instruments that worked in the en-

ergy range of 0.3-40 keV. Four of them were aligned

with the spin axis for the detailed study of a small part

of the sky within 10

○

of the satellite pole. Its instru-

ments included a rotation modulation collimator, three

diﬀerent detectors- a photomultiplier, an electron mul-

tiplier and a proportional counter. The remaining two

instruments were perpendicular to the spin axis of the

satellite. These included the All-Sky Monitor (ASM),

a small pinhole camera to study transient events, and

the Sky Survey Instrument, which consisted of pro-

portional counters. This mission discovered several

long-period pulsars and identiﬁed iron line emission in

extragalactic sources.

1.2.6 FILIN

Soviet Union’s ﬁrst payload placed in space to

study X-rays was the instrument FILIN, which was

mounted on board of Salyut-4 space station. It con-

sisted of three gas ﬂow proportional counter detectors

that worked in the energy range of 2-10 keV, and a

smaller proportional counter to study about soft X-rays

(0.2-2keV). FILIN was mounted on the space station in

December 1974 and operated for around 3 months. The

ﬁrst one month was allocated for scanning, while the

remaining 2 were allocated for the pointed observation

of bright X-ray sources and supernova remnant.

1.2.7 SAS-3

The second small astronomical satellite for X-

ray Astronomy, SAS-3, was launched from San Marco

Launch facility in May 1975. SAS-3 was designed as a

spinning satellite to study the polarization of the X-ray

sources. It consisted of four X-ray instruments: ro-

tating modulation collimator systems(RMCS), crossed

slat collimator(SME), each of them with a proportional

counter which designed to view the sky in a wide band

directions, tube collimator (TC) sensitive towards X-

rays in the energy range 0.4-55 keV, and a low energy

detector system(LEDS). SAS-3 provided more pre-

cise positional accuracy for bright X-ray sources and

discovered several bursting sources, including rapid

bursters. This mission also observed X-rays from

Figure 3: Schematic diagram and instruments of SAS-

3.(Credit:NASA)

highly magnetic white dwarf binary systems.

1.2.8 Aryabhata and Bhaskara

Aryabhata and Bhaskara were the ﬁrst and sec-

ond satellite programs of India, respectively. Aryb-

hata, launched in April 1975, had proportional counter

ﬁlled with mixture of Ar, CO

and He, which studied

X-rays in the energy range of 2.5 - 115 keV. Aryb-

hata found hardening in the spectrum of Cyg X-1 and

discovered two X-ray sources. Bhaskara-I (there were

two Bhaskara missions, the Bhaskara -II didn’t use any

X-ray instrument), launched in 1979, consisted of an

X-ray pinhole survey camera that operated in the range

of 2-10 keV. It mainly focused on observing the tran-

sient events and long-term variability of steady sources.

Bhaskara-I worked successfully for one month and was

turned oﬀ. When it was turned on again, it didn’t op-

erated correctly.

1.2.9 HEAO-1

In August 1977, NASA launched a large scientiﬁc

payloads satellite known as the High Energy Astron-

omy Observatories (HEAO) which had a mass of 3000

kg, whereas Uhuru was only 150 kg in mass. HEAO-

1 surveyed the entire sky over three times in the en-

ergy range of 0.2-10 keV. It was also used for detailed

pointed observations lasting for 3-6 hours. HEAO-1

operated until January 1979 and had four major scien-

1.2 Satellites in 1970s

Figure 4: Aryabhata Satellite.(Credit:public domain)

Figure 5: the HEAO-1 A-1 X-ray source catalog.

(Credit: NASA)

tiﬁc instruments on board:

Large Area Sky Survey experiment (LASS), that

consisted of proportional counter array sensitive

to 0.25-25 keV.

Cosmic X-ray Experiment (CXE), a smaller pro-

portional counter array designed to study dif-

fused X-ray background.

Modulation Collimator (MC) experiment, that

covered the energy range of 0.9-13.3 keV to ac-

curately determine the celestial position.

High energy experiment, the Hard X-ray/low en-

ergy Gamma Ray experiment that consisted of

seven inorganic phoswich scitillator detectors.

Major scientiﬁc contributions of HEAO-1 include the

ﬁnding of aperiodic variability in Cyg X-1 down to a

timescale of 3 ms. It also discovered the ﬁrst eclipse in

low mass X-ray binary system and found variability in

X-ray bursts having a time scale of tens of milliseconds.

1.2.10 Hakucho

Hakucho was the ﬁrst Japanese satellite that stud-

ied X-rays. It was launched as the second in the series

of Cosmic Radiation Satellite (CoRSa) in February

1979 and after its successful launch, it renamed as

Hakucho. During its 6 year operational period, Haku-

cho mainly studied about transient phenomena using

three instruments onboard. Its instruments included

the Very Soft X-ray experiment(VSX) which had four

unit proportional counters with a very thin polypropy-

lene windows; Soft X-ray Experiment (SFX) which had

6 units proportional counters with Beryllium windows;

and Hard X-ray detectors (HDX) which consisted of 2

Na(T1) scintillators. Hakucho discovered many burst

sources and soft X-ray transient sources.

1.2.11 Ariel VI

After the success of Ariel V, the USA and the

UK collaboratively launched Ariel VI in June 1979.

However, it was less successful due to the problems

caused by interference with powerful military radar.

The satellite was aﬀected by strong magnetic ﬁelds,

which severely hammered the command encoder and

the pointing operations. Ariel VI had three scientiﬁc

payloads, one among them was dedicated to study cos-

mic rays and the remaining two were X-ray instruments.

These included a soft X-ray telescope consisting of four

grazing-incidence hyperboloid mirrors that reﬂected

X-rays through an aperture to four continuous-ﬂow

propane gas detectors, and a medium X-ray propor-

tional counter (MXPC), that consisted of four multi-

layered Xe-proportional counters. During its three-

year operation mission, it produced some results, such

as the phase variable iron line emission of source GX

1+4.

The 1970s marked a signiﬁcant progress in X-ray

Astronomy, but all the satellites launched during this

period were only able to detect X-rays and not able to

image them. It is challenging to image X-ray sources

because the X-ray photons can penetrate through the

mirrors due to its high energy, making it diﬃcult to fo-

cus them onto a single imaging detector. In the 1980s,

we saw the ﬁrst X-ray images and we can discuss these

1.2 Satellites in 1970s

advancements in X-ray astronomy in the upcoming ar-

ticle.

References

Santangelo, Andrea and Madonia, Rosalia and

Piraino, Santina A Chronological History of X-

Ray Astronomy Missions. Handbook of X-ray

and Gamma-ray Astrophysics.ISBN 9789811645440

Forman W, Jones C, Cominsky L, et al (1978)

The fourth Uhuru catalog of X-ray sources.ApJS

38:357–412

High Energy Astrophysics Observatories

50 Years Ago, NASA’s Copernicus Set the Bar

for Space Astronomy

About the Author

Aromal P is a research scholar in Depart-

ment of Astronomy Astrophysics and Space Engineer-

ing (DAASE) in IIT Indore. His research mainly fo-

cuses on neutron stars and blackholes.

Exploring Stellar Clusters: Insights from

Color-Magnitude Diagrams, Part-1

by Sindhu G

airis4D, Vol.2, No.6, 2024

www.airis4d.com

2.1 Introduction

A color-magnitude diagram serves as a celestial

blueprint, plotting the apparent magnitude of objects

against their color index. Essentially, the CMD pro-

vides a visual representation akin to the Hertzsprung-

Russell diagram, a fundamental tool in stellar astron-

omy. On an HRD, absolute magnitude or luminosity

ascends the vertical axis against spectral type or tem-

perature, reversed to increase to the left. This visu-

alization allows us to eﬀectively group stars by their

evolutionary state.

In observational studies, the apparent magnitude,

which proportional to the logarithm of ﬂux, can act

as a proxy for luminosity within star clusters. This

is possible because all stars in the cluster are posi-

tioned at approximately the same distances and expe-

rience comparable dimming eﬀects from interstellar

extinction. Likewise, the color index, reﬂects the sur-

face temperature of the stars, as magnitudes measured

through diﬀerent ﬁlters sample diﬀerent parts of the

star’s blackbody-like spectrum. A diﬀerence in mag-

nitude corresponds to the ratio of ﬂuxes. Within stellar

clusters, where all stars reside at uniform distances,

this is directly related to the ratio of the two luminosi-

ties across diverse parts of the spectrum. As stars of

varying temperatures exhibit distinct spectra, they also

manifest diﬀerent colors, creating a rich tapestry of stel-

lar diversity within clusters. Hertzsprung-Russell dia-

grams (HRDs) and color-magnitude diagrams (CMDs)

stand as indispensable instruments for unraveling the

composition and evolution of stars. CMDs of star clus-

ters oﬀer valuable insights, aiding in the estimation

of distances and providing indicators of the ages of

distinct stellar populations.

2.2 Star Clusters

Star clusters serve as invaluable resources for study-

ing stars on a broader scale due to their presumed

uniformity in properties. The premise lies in the as-

sumption that all stars within a cluster originated nearly

simultaneously from the same interstellar gas cloud, re-

sulting in remarkable homogeneity among them. This

uniformity implies that the primary distinguishing fac-

tor among cluster stars is their mass. Therefore, by

analyzing the properties of a single star within a clus-

ter—such as its age, distance, and composition—we

can extrapolate similar characteristics to other stars

within the cluster.

Although some stars in a cluster may form earlier

than others, the temporal spread in their formation is

relatively minor compared to their lifetimes, thus neg-

ligible for most analytical purposes. Similarly, while

there may exist a slight disparity in the distances of

stars within a cluster from our vantage point, this vari-

ation is typically insigniﬁcant in relation to the overall

distance of the cluster. For instance, in the case of the

globular cluster M13, although its outermost stars may

be approximately 50 parsecs distant from the cluster’s

center, the cluster itself lies around 7,700 parsecs away

from Earth.

2.3 The Hertzsprung-Russell Diagram of Globular Clusters

Furthermore, the chemical composition of stars

within a given cluster is expected to be highly uniform,

owing to the presumed thorough mixing of elements

and molecules within the parent gas cloud from which

they formed.

During the formation of stars from a molecular

cloud, a broad spectrum of masses is produced, ranging

from very high mass stars to low mass brown dwarfs.

Yet, observations reveal a pronounced decline in the

formation rate of high mass stars, with the vast ma-

jority of stars in our Solar Neighborhood being less

massive than or comparable to the Sun. This trend

persists across various star-forming regions in the uni-

verse, indicating a universal law governing the relative

distribution of high and low mass stars, known as the

stellar initial mass function.

Therefore, insights gleaned from the study of a

single star cluster can be extrapolated to broader con-

texts, allowing for generalized conclusions regarding

stellar formation and distribution across the cosmos.

This discussion pertains speciﬁcally to globular clus-

ters. Globular clusters, dense, spherical aggregations

of stars orbiting the centers of galaxies, stand as some of

the universe’s oldest structures, boasting ages ranging

from 10 to 13 billion years. Within the realm of study-

ing these ancient ensembles, the Hertzsprung-Russell

(HR) diagram emerges as a potent tool, enabling a

profound examination of the characteristics and evo-

lutionary trajectories of stars nestled within globular

clusters. Initially, we will delve into the HR diagram

of globular clusters, followed by an examination of the

color-magnitude diagram.

2.3 The Hertzsprung-Russell

Diagram of Globular Clusters

The HR diagrams of globular clusters manifest

several remarkable features that distinguish them from

those of younger open clusters. Some of them are given

below.

Absence of Hot, Massive Stars: Notably absent

in globular clusters are stars of spectral types O, B, A,

and early F, given their short lifespans and departure

Figure 1: Globular star cluster Messier 55 is located

at a distance of about 17000 light-years from Earth.

(Image Credit: ESO/J. Emerson/VISTA )

Figure 2: Globular star cluster NGC 6362. Image

Credit: ESA/Hubble & NASA

2.4 Challenges in Constructing HR Diagrams for Globular Clusters

from the main sequence.

Prominent Red Giant Branch: A striking fea-

ture within HR diagrams of globular clusters is the

well-deﬁned red giant branch, housing stars that have

depleted hydrogen in their cores and are presently un-

dergoing hydrogen fusion in a shell enveloping an inert

helium core.

Horizontal Branch: Numerous globular clusters

exhibit a horizontal branch, showcasing stars engaged

in helium fusion within their cores alongside hydrogen

fusion in a surrounding shell.

Asymptotic Giant Branch: Certain globular clus-

ters showcase an asymptotic giant branch, accommo-

dating stars involved in helium and hydrogen fusion

within shells encasing an inert carbon-oxygen core.

Main Sequence Turnoﬀ: Situated at lower lumi-

nosities and temperatures compared to younger clus-

ters, the main sequence turnoﬀ point, where stars de-

part from the main sequence, reﬂects the advanced age

of globular clusters.

Globular clusters oﬀer an exceptional setting for

testing hypotheses concerning stellar evolution. Uni-

form in age, distance, and chemical composition, stars

within these clusters primarily vary in mass. By scru-

tinizing observed HR diagrams of globular clusters

alongside theoretical frameworks, astronomers can re-

ﬁne estimates of both cluster and Galactic ages.

2.4 Challenges in Constructing HR

Diagrams for Globular Clusters

Constructing precise HR diagrams for globular

clusters poses several challenges:

Crowding: The high stellar density within glob-

ular clusters may lead to blending and confusion, com-

plicating the acquisition of accurate photometric data

for individual stars.

Distance Determination: The accurate determi-

nation of distances to globular clusters is crucial for

converting apparent magnitudes to absolute magni-

tudes, yet it remains subject to uncertainty.

Metallicity Inﬂuence: Given their lower metal-

licities relative to the Sun, globular clusters exhibit

distinct star positions on HR diagrams, necessitating

consideration when comparing observations to theo-

retical frameworks.

Despite these obstacles, the HR diagrams of glob-

ular clusters persist as invaluable resources, oﬀering

profound insights into the formation and evolution of

these ancient stellar systems and the Galaxy as a whole.

In the upcoming article, we will explore the Hertzsprung-

Russell diagram and the color-magnitude diagram of

several globular clusters.

References:

Star Clusters: Inside the Universe’s Stellar Col-

lections

Star Clusters

What are star clusters?

Star Clusters

Star cluster

Globular cluster

Measuring the Age of a Star Cluster

HR Diagram, Star Clusters,and Stellar Evolution

The Colour-Magnitude-Diagram of Globular Clus-

ters and Stellar Evolution

Colour-Magnitude Diagrams of Star Clusters:

Determining Their Relative Ages

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

Precision Medicine and its Role in Cancer

Care

by Geetha Paul

airis4D, Vol.2, No.6, 2024

www.airis4d.com

1.1 Introduction

Precision medicine is a cutting-edge approach to

healthcare that tailors disease prevention, diagnosis,

and treatment to the speciﬁc characteristics of each

patient based on the particular genes, proteins, and

other substances in a person’s body. This approach is

also sometimes called personalised medicine or per-

sonalised care. Imagine it like a tailor crafting a suit

- instead of a generic oﬀ-the-rack option, the suit is

designed to ﬁt the individual’s unique measurements

perfectly. In contrast to traditional medicine, which

often relies on a one-size-ﬁts-all approach, precision

medicine considers factors like Genetics, Environment,

Lifestyle, etc. By considering these individual factors,

precision medicine aims to improve treatment eﬀec-

tiveness by targeting the underlying cause of disease in

a speciﬁc patient; precision medicine can lead to more

eﬀective treatments with fewer side eﬀects. Focus-

ing on treatments that are most likely to be successful,

precision medicine can reduce unnecessary test proce-

dures and reduce costs. While still a developing ﬁeld,

precision medicine holds great promise for the future

of healthcare, oﬀering a more personalised and prac-

tical approach to preventing, diagnosing, and treating

diseases.

In the future of medicine, doctors are going be-

yond a one-size-ﬁts-all approach. With special tests,

they can gain valuable insights into your unique body

and how it responds to things. This information helps

them create a personalised healthcare plan just for you,

complete with speciﬁc recommendations tailored to

your needs. These tests can be beneﬁcial for diagnosing

illnesses more precisely and ﬁnding the most eﬀective

treatments. But that’s not all! They can also empower

you to make informed decisions about your health. By

understanding your body better, you can choose healthy

habits, schedule earlier screenings when needed, and

take other proactive steps to stay well. So, while your

doctor might not use the term precision medicine di-

rectly, you might hear them talk about genes, DNA, or

unique markers in your body. These puzzle pieces help

them create the best healthcare plan for you.

Concerning cancer, precision medicine often means

looking at how changes in certain genes or proteins in

a person’s cancer cells might aﬀect their care, such as

their treatment options. But it can have other uses as

well.

1.2 Precision Medicine and Gene

Changes

Understanding how genes and proteins work within

our cells are vital in tailoring treatments for each in-

dividual. Precision medicine is primarily based on

knowing the eﬀects of changes in speciﬁc genes (pro-

teins) and inside cells. Genes carry instructions within

the DNA of our cells, and they tell the cell how to make

the proteins it needs to function. When cells divide

to make new cells, the genes inside the cell are copied.

A gene change (a variant or mutation) happens when

1.3 Gene Changes and Cancer

(Image courtesy:https://www.cancer.gov/about-cancer/causes-prevention/genetics/genetic-

changes-infographic)

Figure 1: How genetic information creates proteins

through the transcription process.

there’s a mistake in the copying process. Individual

gene variations can inﬂuence how a person responds

to diseases and medications. We inherit our genes

from our parents, like a blueprint for building proteins

(inherited gene change). Sometimes mistakes happen

during cell division, even after birth, creating new gene

changes throughout life (acquired gene change). The

impact of these changes can vary - some might be

harmful, while others may not aﬀect our health.

1.3 Gene Changes and Cancer

Cancer begins when some of the genes in a cell

become abnormal, causing the cell to grow and di-

vide out of control. Here, you can learn more about

how cell gene changes can lead to cancer. The ‘code’

or ‘blueprint’ for each gene is contained in chemicals

called nucleotides. DNA is made up of 4 nucleotides

(A, T, G, and C), which act like the letters of an alpha-

bet. Each gene comprises a long chain of nucleotides,

the order of which tells the cell how to make a speciﬁc

protein. While we all have the same set of genes, we

also have diﬀerences in our genes that make each of us

unique.

Cancer is fundamentally a disease driven by changes

in genes. These genetic alterations disrupt the normal

growth and division of cells, leading to uncontrolled

cell proliferation and tumour formation. Here’s a closer

look at the connection between gene changes and can-

cer:

(Image courtesy:https://www.cancer.gov/about-cancer/causes-prevention/genetics/genetic-

changes-infographic)

Figure 2: Shows the Gene mutation or gene change.

A missense mutation is when Tryptophan is mutated to

Cysteine, and a nonsense mutation is another change

of a single DNA.

Oncogenes: Normally, these genes promote cell

growth and division. When mutated, they become

overly active, causing cells to divide uncontrollably.

An oncogene is a mutated gene that has the potential to

cause cancer. Before an oncogene becomes mutated, it

is called a proto-oncogene and plays a role in regulating

normal cell division. Cancer can arise when a proto-

oncogene is mutated, changing it into an oncogene and

causing the cell to divide and multiply uncontrollably.

Some oncogenes work like an accelerator pedal in a

car, pushing a cell to divide repeatedly. Others work

like a faulty brake in a vehicle parked on a hill, causing

the cell to divide unchecked.

Tumour Suppressor Genes: These genes act as

brakes on cell division. Mutations in these genes

can disable their function, allowing cells to divide

unchecked.

DNA Repair Genes: Mistakes occur during cell

division, and these genes help ﬁx them. Mutations in

DNA repair genes can lead to an accumulation of er-

rors, increasing the risk of cancer-causing mutations.

Some gene mutations can be inherited from parents, in-

creasing a person’s risk of developing certain cancers

(e.g., BRCA mutations and breast cancer). However,

most cancer-causing mutations occur throughout a per-

son’s lifetime due to factors like exposure to carcino-

gens (smoking, radiation) or errors during cell division.

Multiple Mutations: Cancer usually develops from

a combination of several gene mutations accumulating

over time. A single mutation might not be enough to

cause cancer, but multiple mutations can work together

1.4 Precision Medicine in Cancer Care:

to create a perfect storm for uncontrolled cell growth.

Developing Targeted Therapies: By identifying

the speciﬁc mutations driving a patient’s cancer, doc-

tors can tailor treatments to target those mutations and

potentially achieve better outcomes with fewer side ef-

fects.

Improved Cancer Risk Assessment: Genetic

testing can help identify individuals with inherited mu-

tations that increase their cancer risk, allowing for early

detection strategies like increased screening.

Although gene changes play a major role in cancer,

it’s important to remember that cancer is a complex dis-

ease inﬂuenced by various factors. Genetic alterations

can aﬀect or intensify the normal function of proteins,

which could cause uncontrollable growth and division

of cells, leading to cancer.

1.4 Precision Medicine in Cancer

Care:

Precision medicine can analyse a person’s genes

and family history to identify those with a higher risk of

developing certain cancers. This allows for early inter-

vention strategies like lifestyle changes or preventative

medications to lower their risk potentially. By look-

ing for speciﬁc genetic mutations or protein markers,

precision medicine can help detect cancers at earlier

stages when they’re often more treatable. This can sig-

niﬁcantly improve a patient’s prognosis. Traditional

methods may not always diﬀerentiate between diﬀer-

ent types of cancer. Precision medicine uses tumour

proﬁling to identify the speciﬁc genetic makeup of the

cancer cells. This leads to a more accurate diagnosis,

crucial for choosing the most eﬀective treatment. This

is the most impactful aspect of precision medicine.

By understanding cancer’s speciﬁc mutations, doctors

can choose targeted therapies that directly attack the

vulnerabilities of those cancer cells. This approach of-

fers potentially better outcomes and fewer side eﬀects

than traditional therapies like chemotherapy. Precision

medicine allows doctors to monitor a patient’s response

to treatment by tracking changes in the cancer cells’ ge-

netic makeup. This helps determine if the therapy is

working and allows for adjustments if needed.

1.5 Types of Cancer where Precision

Medicine is Used:

While precision medicine isn’t a one-size-ﬁts-

all solution for every cancer, researchers are making

signiﬁcant strides. Several types of cancer are al-

ready beneﬁting from this approach, including col-

orectal, breast, and lung cancers. Precision medicine

is also valuable in certain leukaemias, lymphomas,

melanomas, and cancers of the oesophagus, stomach,

ovaries, and thyroid. If you have one of these cancers,

your doctor will likely order tests to identify speciﬁc

gene or protein changes in your tumour cells. These

tests help determine the most eﬀective treatment course

for you. Don’t hesitate to ask your doctor if such testing

was performed on your cancer.

1.6 Precision Medicine and Cancer

Care Prevention for High Risk

Individuals

Precision medicine goes beyond just treating ex-

isting cancers. It can also play a crucial role in cancer

prevention for high-risk individuals. Family history or

a doctor noticing a concerning pattern might prompt

tests to identify inherited genetic mutations that ele-

vate cancer risk. Genetic counsellors can guide pa-

tients through this process. Positive test results can

lead to earlier and more frequent screenings to catch

cancer early. Additionally, doctors might recommend

medications or lifestyle changes to reduce the patient’s

cancer risk proactively.

The two types of treatment most often used in

precision medicine are Targeted drug therapy and Im-

munotherapy. Targeted drug therapy is a type of cancer

treatment that uses drugs or other substances to identify

and attack speciﬁc cancer cells precisely. Targeted ther-

apy can be used by itself or in combination with other

treatments, such as traditional or standard chemother-

apy, surgery, or radiation therapy. If your treatment

plan includes targeted therapy, knowing how it works

1.7 Limitations to Precision Medicine in Cancer

and what to expect can help you prepare for treatment

and make informed decisions about your care.

Immunotherapy is a treatment that uses a person’s

immune system to ﬁght cancer. Immunotherapy can

boost or change how the immune system works to ﬁnd

and attack cancer cells. If the treatment plan includes

immunotherapy, knowing how it works and what to

expect can help us to prepare for treatment and make

informed decisions about your care.

1.7 Limitations to Precision

Medicine in Cancer

Precision medicine oﬀers a revolutionary approach

to cancer care, but limitations still exist.

Only some have equal access to the latest precision

medicine approaches. Geographic location can restrict

access to advanced testing and targeted therapies. Re-

searchers are working to understand how to integrate

precision medicine into cancer care. Clinical trials play

a crucial role in advancing precision medicine. Spe-

ciﬁc gene mutations are often required to participate,

and trials might be concentrated at larger cancer cen-

tres, making participation diﬃcult for some patients.

Even when available, precision medicine might not be

fully utilised. Family history, a critical factor in cancer

risk assessment, might not be thoroughly evaluated.

Genetic testing might not be performed, or the results

might not be eﬀectively used to guide preventive mea-

sures. Biomarker testing and targeted therapies can

be expensive, raising concerns about aﬀordability and

insurance coverage.

Overall, precision medicine in cancer care is a

powerful tool with the potential to improve patient out-

comes signiﬁcantly. It’s an ongoing ﬁeld of research,

with discoveries constantly emerging.

References:

https://www.cancer.org/cancer/managing-cancer/

treatment-types/immunotherapy.html

https://www.cancer.gov/about-cancer/causes-prevention/

genetics/genetic-changes-infographic

https://www.cancer.org/cancer/managing-cancer/

treatment-types/precision-medicine.html

https://www.cancer.org/cancer/managing-cancer/

treatment-types/targeted-therapy.html

About the Author

Geetha Paul is one of the directors of

airis4D. She leads the Biosciences Division. Her

research interests extends from Cell & Molecular Bi-

ology to Environmental Sciences, Odonatology, and

Aquatic Biology.

’

Exploring Varied Approaches for DNA

Methylation Detection

by Jinsu Ann Mathew

airis4D, Vol.2, No.6, 2024

www.airis4d.com

Following our previous exploration of DNA methy-

lation and its pivotal role in gene regulation, we now

delve into the diverse array of techniques developed

to study this essential epigenetic modiﬁcation. Accu-

rately mapping DNA methylation patterns is crucial for

understanding numerous biological processes, includ-

ing development, diﬀerentiation, and disease mecha-

nisms. To this end, researchers have developed sev-

eral sophisticated methods to analyze DNA methyla-

tion with varying degrees of coverage, resolution, and

speciﬁcity.

In this continuation, we focus on four major DNA

methylation analysis techniques: DNA Methylation

Microarray, Whole Genome Bisulphite Sequencing (

WGBS ), Methylated DNA Immunoprecipitation Se-

quencing ( MeDIP seq ) and Reduced Representation

bisulphite Sequencing ( RRBS ). Each of these methods

oﬀers unique insights and advantages, catering to dif-

ferent research needs and experimental designs. This

article will explore the underlying principles, applica-

tions, strengths, and limitations of each technique, pro-

viding a comprehensive overview to guide researchers

in selecting the most suitable method for their speciﬁc

studies. Through this examination, we aim to enhance

our understanding of the epigenetic modiﬁcations that

govern gene expression and contribute to various bio-

logical phenomena.

(Image courtesy:https://www.slideshare.net/slideshow/dna-methylation-from-array-to-

sequencing/266860264)

Figure 1: Illumina’s 450K and 850K BeadArrays.

2.1 DNA Methylation Microarray

DNA methylation microarrays, like Illumina’s pop-

ular 450K and 850K BeadArrays, are workhorses in the

ﬁeld of epigenetics. These microarrays act like tiny ﬁn-

gerprint scanners for DNA methylation patterns across

the genome. Here’s how they work:

Tiny Targets: The microarray surface holds thou-

sands of oligonucleotide probes (Figure 1), each de-

signed to bind to a speciﬁc location on the DNA called

a CpG site, where methylation can occur.

Bisulphite Magic: DNA samples undergo a chem-

ical treatment (bisulphite conversion) that diﬀerentiates

between methylated and unmethylated DNA.

Shining a Light: The treated DNA is then applied

to the microarray. Each probe can distinguish between

the modiﬁed (methylated) and unmodiﬁed DNA based

on the bisulphite treatment. Fluorescent labels attached

to the DNA allow researchers to measure the intensity

of the signal, reﬂecting the methylation level at each

CpG site.

Data Delving: After processing and adjustments

2.2 WGBS

to account for variations, researchers use bioinformat-

ics tools to interpret the biological meaning of the

methylation patterns.

Advantages

These microarrays oﬀer several advantages, in-

cluding high throughput, standardized protocols, and

cost-eﬀectiveness. They can analyze hundreds of thou-

sands to millions of CpG sites simultaneously, provid-

ing a comprehensive view of the methylome. Standard-

ized platforms like the 450K and 850K BeadArrays en-

sure reproducibility and comparability across studies.

Additionally, microarrays are more cost-eﬀective than

sequencing-based methods, making them accessible to

many research labs. Their user-friendly protocols and

interfaces further enhance accessibility, even for re-

searchers without extensive bioinformatics expertise.

Limitations

Despite their utility, DNA methylation microar-

rays have limitations. Their coverage is predeﬁned,

potentially missing important methylation sites else-

where in the genome. They provide lower resolution

compared to sequencing-based methods , which may

overlook novel methylation sites or rare methylation

events. Additionally, probe design and selection can in-

troduce bias, focusing on known regions of importance

while potentially disregarding other relevant areas of

the genome.

2.2 WGBS

Whole Genome Bisulphite Sequencing (WGBS)

is a cutting-edge technique used to comprehensively

map DNA methylation patterns across the entire genome

at single-nucleotide resolution. Here’s how WGBS

works:

Bisulphite conversion: Genomic DNA is treated

with bisulphite, a chemical that converts unmethylated

cytosines (C) to uracil (U), while leaving methylated

cytosines unchanged. This conversion process pre-

serves the methylated cytosines, allowing them to be

distinguished from unmethylated cytosines during se-

quencing.

Sequencing: After bisulphite treatment, the DNA

is then chopped into smaller pieces and subjected to

high-throughput sequencing. During sequencing, uracil

gets read as thymine (another building block).

Bioinformatics analysis: By comparing the orig-

inal DNA sequence with the bisulphite-converted and

sequenced one, researchers can identify the locations

where cytosines remained unchanged (methylated) be-

cause they didn’t get converted by bisulphite.

Advantages

Unlike microarrays or MeDIP-seq, WGBS oﬀers

the highest resolution, pinpointing the exact location

of every methylated cytosine in the entire genome.

WGBS doesn’t miss any spots. It analyzes methylation

patterns across the entire genome, providing a compre-

hensive picture. Since it relies on chemical conversion

and sequencing, WGBS avoids any potential biases in-

troduced by probe design in microarrays or antibody

selection in MeDIP-seq.

Limitations

WGBS is a more expensive and technically de-

manding method compared to other techniques. The

sheer amount of data generated by WGBS requires

powerful computational tools for analysis and inter-

pretation. bisulphite treatment can damage DNA, re-

quiring high-quality starting material.

2.3 MeDIP-seq

Methylated DNA Immunoprecipitation Sequenc-

ing (MeDIP-seq) is a powerful technique used to study

DNA methylation patterns across the genome. This

method combines immunoprecipitation of methylated

DNA with high-throughput sequencing, allowing re-

searchers to map methylation marks with high speci-

ﬁcity and coverage. Here’s a detailed explanation of

the MeDIP-seq process:

Antibody Attraction: Much like a magnet drawn

to metal, MeDIP-seq utilizes antibodies speciﬁcally

2.4 RRBS

designed to recognize and bind methyl groups attached

to DNA.

DNA Fragmentation: The DNA sample is ﬁrst

broken down into smaller, more manageable pieces.

Immunoprecipitation: The fragmented DNA is

then mixed with the methylation-targeting antibodies.

These antibodies act like ﬁshing hooks, speciﬁcally

pulling out the methylated DNA fragments from the

solution.

Sequencing Power: The captured, methylated

DNA fragments undergo high-throughput sequencing,

revealing the precise locations of methylation within

the genome.

Advantages

Unlike microarrays that analyze everything at once,

MeDIP-seq enriches for speciﬁcally methylated DNA,

allowing researchers to delve deeper into these regions.

Compared to some whole-genome sequencing meth-

ods, MeDIP-seq is a less complex technique and re-

quires smaller amounts of DNA as a starting point. The

use of antibodies speciﬁc to 5-methylcytosine allows

for precise enrichment of methylated DNA, making

MeDIP-seq highly sensitive and speciﬁc. MeDIP-seq

provides comprehensive coverage of the methylome,

enabling the identiﬁcation of methylated regions across

the entire genome.

Limitations

MeDIP-seq has lower resolution compared to bisul-

phite sequencing methods. It identiﬁes regions of

methylation rather than single-nucleotide resolution,

which may miss ﬁne-scale methylation patterns. The

eﬃciency of immunoprecipitation can vary depending

on the density and context of methylated cytosines,

potentially introducing biases in the data. MeDIP-

seq provides relative, not absolute, quantiﬁcation of

methylation levels, making it challenging to determine

the exact percentage of methylation at speciﬁc sites.

2.4 RRBS

Reduced Representation Bisulphite Sequencing (

RRBS ) is a specialized technique used to study DNA

methylation patterns in speciﬁc regions of the genome

with high resolution and coverage, while simultane-

ously reducing the complexity and cost associated with

whole-genome sequencing.

Fragmentation by restriction enzymes: RRBS

starts by chopping the DNA sample into smaller frag-

ments using restriction enzymes. These enzymes act

like molecular scissors, cutting DNA at speciﬁc recog-

nizable sequences

Adapter ligation: Short adapter sequences are

then attached to the ends of the fragmented DNA.

These adapters are like identiﬁcation tags that allow

the speciﬁc regions of interest to be isolated later.

Bisulphite conversion: Similar to WGBS, the

DNA fragments undergo bisulphite treatment, convert-

ing unmethylated cytosines to uracil.

Enrichment and Sequencing: Only the frag-

ments containing the adapter sequences (representing

the targeted regions) are selectively ampliﬁed and then

sequenced.

Advantages

Compared to WGBS, RRBS is a more budget-

friendly option as it focuses on analyzing only speciﬁc

regions of the genome. Researchers can choose speciﬁc

areas of interest for methylation analysis, making it

ideal for focused studies. By analyzing only targeted

regions, RRBS generates a smaller dataset compared

to WGBS, simplifying analysis and interpretation. By

focusing on speciﬁc regions, RRBS can achieve higher

sequencing depth within those regions compared to

WGBS.

Limitations

Unlike WGBS, RRBS doesn’t analyze the en-

tire genome, potentially missing important methyla-

tion events outside the targeted areas. Choosing the

appropriate regions for analysis is crucial, as impor-

tant methylation sites outside these regions might be

2.5 Conclusion

missed.

2.5 Conclusion

In conclusion, the methodologies of DNA methy-

lation analysis, including Microarray, Whole-Genome

bisulphite Sequencing (WGBS), Methylated DNA Im-

munoprecipitation Sequencing (MeDIP-seq), and Re-

duced Representation bisulphite Sequencing (RRBS),

each oﬀer unique insights into the intricate landscape

of epigenetic regulation.

Microarrays provide a cost-eﬀective and standard-

ized approach for genome-wide methylation proﬁling,

making them ideal for large-scale studies. WGBS

stands as the gold standard for obtaining the most com-

prehensive view of DNA methylation patterns, oﬀering

single-nucleotide resolution across the entire genome.

MeDIP-seq oﬀers a targeted approach to enrich and

analyze methylated regions, providing a valuable mid-

dle ground between microarrays and WGBS. Finally,

RRBS strikes a balance between cost and detail, allow-

ing researchers to focus on speciﬁc regions of interest

for in-depth methylation analysis.

The choice of technique ultimately depends on the

research question and available resources. By under-

standing the strengths and limitations of each method,

researchers can select the most appropriate tool to un-

lock the secrets hidden within the methylation patterns

of our DNA, ultimately furthering our understanding

of gene regulation, development, disease, and the in-

tricate world of epigenetics.

References

Comprehensive coverage for genome-wide DNA

methylation studies

DNA methylation detection: bisulphite genomic

sequencing analysis

bisulphite sequencing

Overview of Methylated DNA Immunoprecipi-

tation Sequencing (MeDIP-seq)

Principles and Workﬂow of Whole Genome bisul-

phite Sequencing

An Introduction to Reduced Representation bisul-

phite Sequencing (RRBS)

Complete Guide to Using Reduced Representa-

tion bisulphite Sequencing (RRBS) for Genome-

Wide DNA Methylation Analysis

About the Author

Jinsu Ann Mathew is a research scholar

in Natural Language Processing and Chemical Infor-

matics. Her interests include applying basic scientiﬁc

research on computational linguistics, practical appli-

cations of human language technology, and interdis-

ciplinary work in computational physics.

Part IV

General

How to Think Like a Product Manager in

Academia

by Arun Aniyan

airis4D, Vol.2, No.6, 2024

www.airis4d.com

1.1 Introduction

The products that we use daily, staring from a sim-

ple tooth brush to the mobile phone we use, have gone

through rigorous design and thought process. Even a

simple needle used for sewing clothes has undergone a

long design process. A lot of us take for granted that

the small things that we use for daily applications were

a simple idea that was manufactured immediately.

Every industry has a designated product specialist

called the product manager. They’re maybe separate

product managers for diﬀerent products. The product

manager will be a specialist in terms of the product

itself and also the consumer base related to it.

The principles that a product manager follows are

strict and methodological, such that it decides the scope

of a project. For example, Microsoft Word, which is a

product, has functionalities and features that were de-

signed and proposed by a product manager. The prod-

uct manager understands the customer requirements

and pain points well enough, which lays out the plan

for a product design. The methods that a product man-

ager follows are good to follow in academic research as

well. It will help in doing focussed work and producing

sharp results.

1.2 Why is the Product or Feature

Required

The ﬁrst principle and ﬁrst question a product

manager asks is “why” should this product be devel-

oped even before venturing in to asking “what” and

“how”. The “why” will set the future direction of the

product and project. When asking “why”, a detailed

investigation of the domain or pain point that the prod-

uct may address will need to be explored. The initial

research will be around what is the pain point that

customers have in the speciﬁc domain that needs a so-

lution. Is the pain point new which was introduced or

has been around for a long time is part of the question?

If the pain point is a new one, the question would

be the source and how it was introduced. There will

be a market study on if already a solution is available

and what would be the scope for another solution and

its possibility of adoption.

Similarly, if the pain point has been around for

a while, the thinking would be either there is already

a solution that addresses it but not fully, or it has not

been addressed at all. If solutions already exist, then the

product team would study the deﬁciencies of existing

solutions and ﬁnd if there is still viable scope for a new

solution.

In academia, this can be analogous to choosing a

new research problem to work on. One will need to

do a detailed literature review and ﬁnd a deﬁcit in the

area which will need a novel solution that can push the

envelope of scientiﬁc understanding. The scope would

be decided by how much would be required to make the

research publishable, and also what would be a good

result that extends our scientiﬁc understanding.

1.3 Problem Identiﬁcation and Deﬁnition

1.3 Problem Identiﬁcation and

Deﬁnition

Another way of issue identiﬁcation is called “un-

derstanding your customer’s requirements”. Once the

product team has identiﬁed the pain point, there maybe

several angles to view it. But the foremost step is to see

it from a customer perspective. The product team may

have their own favourite recipe to solve the problem,

but it may not be the solution the customer requires.

Developing a solution that is not customer-centric will

make the product fail in terms of adoption.

Therefore, it is important to see how the customer

would like to treat the problem and have the solution

cater to their requirements. The product team needs to

sit with the customer and understand the issue from

their perspective and have it deﬁned. Initially, the

core issue needs to be identiﬁed from the customer

requirements and then the scope and solution design is

deﬁned.

In science, what is often done is that many solu-

tions are designed but may not be widely accepted by

the community. This is mostly because the solution not

bad, but because it may not be catering the community

requirements.

1.4 Incremental Improvements

Depending on the issue, it may not be possible to

solve the challenges in one go within the given time. In

such cases, the product team will realise the time con-

straint and will opt to deliver the solution in a phased

approach. In this manner, the ﬁrst solution will solve

the issue to the minimum agreed by the customer. This

agreement with the customer has to be reached before

the project commences. The initial delivery solution is

designed at minimum scale.

Thereafter, a timeline is agreed upon with the

customer to deliver updates that will further add into

the complete solution. The advantage of this phased

delivery approach is that it allows for incremental im-

provements. There maybe certain patches in the ﬁrst

phase the customer likes certain small changes. Having

a delivery timeline which is phased will allow updating

the overall solution having continued improvement.

This iterated approach to solution delivery is a

common methodology taken by many product man-

agers. This helps to do periodic retrospective of the

project and make amendments in an agile fashion while

releasing the next phase.

In science, this incremental approach is rarely

used and often many groups spend too much time pol-

ishing a solution and by that time they may get scooped

by another publication which possibly is a B-class re-

sult. Packaging solutions and taking a shorter time

to publish is always advantageous for young scientists

who are under constant pressure of publishing. Phased

approaches give room for more publications.

1.5 Active Customer Engagement

If a project needs to fail, it must fail early and

avoid a catastrophic failure. In industry, product man-

agers actively engage with their customers to keep the

solution aligned to the customer needs. This involves

constant communication with the customers, sharing

the project updates, giving early peeks and demos of

the solution. The initial feedbacks will help the project

align in the correct direction and also avoid catastrophic

failure at the end.

In academia, a common mistake that young re-

searchers do is they fail to align with the project goal

and tend to go in the wrong direction. They will have

spent large amounts of time and eﬀort and ﬁnally ﬁnd

the work was wasted. This requires sharing updates

with the project advisor frequently and receiving their

timely inputs.

1.6 Conclusion

Product managers in industry perform a vital role

in the success of products and their future. The princi-

ples that they practice are ones that could be followed in

scientiﬁc research as well. Every paper can be thought

of a product that the science community wants to bene-

ﬁt from. In such regard, if principles of product design

1.6 Conclusion

and management are applied, it would beneﬁt both the

community and the researcher.

Reference

Essential Design Principles for Product Man-

agers

Seven Product Management Principles

How to Think Like a Product Manager

Product Principles

About the Author

Dr.Arun Aniyan is leading the R&D for Arti-

ﬁcial intelligence at DeepAlert Ltd,UK. He comes from

an academic background and has experience in design-

ing machine learning products for diﬀerent domains.

His major interest is knowledge representation and com-

puter vision.

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.