Cover page
Super Nova 2023ixf
The supernova, dubbed SN 2023ixf, (the bright blue patch seen on the left of the airis4D logo at bottom) first
seen in May 21,2023 is located in one of the spiral arms of the Pinwheel Galaxy, also known as Messier 101,
which is 25 million light-years away from Earth. You may read about it and watch an animation on it’s discovery
here: https://www.space.com/supernova-closest-earth-m101-galaxy
Managing Editor Chief Editor Editorial Board Correspondence
Ninan Sajeeth Philip Abraham Mulamootil K Babu Joseph The Chief Editor
Ajit K Kembhavi airis4D
Geetha Paul Thelliyoor - 689544
Arun Kumar Aniyan India
Jorunal Publisher Details
Publisher : airis4D, Thelliyoor 689544, India
Website : www.airis4d.com
Email : nsp@airis4d.com
Phone : +919497552476
i
Editorial
by Fr Dr Abraham Mulamoottil
airis4D, Vol.1, No.5, 2023
www.airis4d.com
The 6th Edition of AIRIS4D Journal: Unveiling the Potential of AI and Advancements in Science As we
introduce the highly anticipated 6th edition of the AIRIS4D journal, it is only fitting for our AI research group
to discuss the fierce competition among major players in the field of artificial intelligence. Microsoft’s Bing
AI and Copilot, Google’s Bard, and OpenAI’s ChatGPT-4 are revolutionizing the world and shaping our daily
lives. These powerful AI language models, including ChatGPT, Bard, and Bing, have the remarkable ability
to enhance natural language understanding, facilitate human-computer interactions, and optimize information
retrieval. With applications ranging from customer service and virtual assistants to content generation and
search engine technology, these AI models are poised to redefine the way we interact with technology. The
advancements in AI are not limited to language models alone. Exciting progress is being made in domains
such as autonomous vehicles, healthcare, robotics, and more. These developments aim to improve efficiency,
decision-making processes, and overall user experiences across various aspects of life. The competition
between these influential players fosters innovation and pushes the boundaries of AI technologies. It stimulates
the development of increasingly sophisticated AI models, algorithms, and applications. As a result, we can
expect continuous improvements in search engine capabilities, conversational AI agents, code development
tools, and other AI-powered services. Ultimately, these advancements aim to enhance user experiences, provide
more accurate and context-aware responses, and boost productivity in numerous domains. Moreover, this
competition among AI giants has the potential to democratize AI technologies, making them accessible to a
wider range of users and industries. As these technologies become more refined and integrated into different
aspects of our lives, they have the power to transform our interactions with machines, information retrieval,
software development, and our overall navigation of the digital landscape. It is important to note that the specific
impact on the world and our lives will depend on the continuous advancements, adoption, and integration of
these AI technologies in various sectors. As AI becomes more prevalent, it is crucial to address ethical
considerations, data privacy concerns, and potential biases to ensure the responsible and beneficial deployment
of these technologies. Now, turning our attention to the content of this 6th edition, we are thrilled to present
a lineup of insightful articles that delve into diverse areas of scientific research. Blesson George’s article,
”Difference Boosted Neural Network (DBNN),” introduces the Difference Boosting Neural Network (DBNN),
a Bayesian neural network that incorporates conditional independence within a Naive Bayes network. This
approach aims to improve the accuracy of the Naive Bayes classifier in classification tasks across different data
types. The article also promises a forthcoming issue where the process of updating prior values using boosting
methods will be discussed. In ”Decoding Methods for Language Models,” Jinsu Ann Mathew explores various
decoding methods employed in language models to generate high-quality and contextually appropriate text.
The article covers decoding techniques such as Greedy Decoding, Beam Search, Top-K Sampling, Nucleus
Sampling, and Random Sampling. Understanding these methods is crucial for effectively utilizing language
models in applications such as machine translation, text summarization, and conversational agents. Sindhu G’s
article, ”Eruptive Variable Stars,” provides an overview of the different types of eruptive variable stars and their
characteristics. Eruptive variables experience eruptions on their surfaces, resulting in changes in brightness
and activity. The article highlights the significance of studying eruptive variable stars in understanding stellar
dynamics and showcases examples of different types of these stars. Geetha Paul’s article, ”Mitochondrial DNA
(mtDNA) - The Small Circular Genome,” offers a comprehensive overview of mitochondrial DNA, its structure,
functions, and applications in various fields. The unique characteristics of mtDNA, such as its high copy
number within cells, make it valuable for genetic diversity monitoring, identifying distinct populations, studying
evolution in endangered species, and contributing to fields like genetics, evolution, forensic investigations, and
conservation efforts. ”Unveiling Earth’s Atmosphere: Layers, Composition, and Lifeline” by Robin Jacob Roy
provides readers with an overview of Earth’s atmosphere, including its composition and the different layers that
compose this vital part of our planet. The article also explores various atmospheric phenomena, such as weather
patterns, the greenhouse effect, atmospheric optics, atmospheric pressure, auroras, and atmospheric acoustics.
Understanding and preserving Earth’s atmosphere is essential as we face the challenges of climate change and
strive for a sustainable future. Finally, in ”The Ten Years of Science - Part II,” Ninan Sajeeth Philip takes us
on a captivating journey through the remarkable advancements in genetics, microbiomes, neuroscience, and
astronomy over the past decade. Philip’s storytelling prowess and comprehensive exploration of these fields
inspire wonder, critical thinking, and a deeper appreciation for the wonders of science. This article serves as
a timely reminder of the power of scientific exploration and the importance of responsible stewardship as we
continue pushing the boundaries of knowledge in the years to come. As we embark on this exciting edition of
AIRIS4D, we invite readers to dive into these captivating articles, explore the frontiers of science, and embrace
the transformative potential of AI technologies. Together, let us celebrate the relentless pursuit of knowledge
and innovation that drives us toward a future shaped by scientific progress and responsible AI integration.
iii
Contents
Editorial ii
I Artificial Intelligence and Machine Learning 1
1 Difference Boosted Neural Network(DBNN) - Part 1 2
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Conditional Independence in Naive Bayes classifier . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Imposed conditional independence in DBNN . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Decoding methods for language models 5
2.1 Greedy Decoding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Beam Search . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Top-K Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4 Nucleus Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.5 Random Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
II Astronomy and Astrophysics 13
1 The Galaxy Revelation and a Growing Universe 14
1.1 What are Nebulae? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2 Galaxies Within Galaxies: The Island Universe Hypothesis . . . . . . . . . . . . . . . . . . . 14
1.3 Edwin Hubble Settles the Dust . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.4 Galaxies that are Moving Away from Us . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.5 The Mystery Continues.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2 Eruptive Variable Stars 20
2.1 What Are Eruptive Variable Stars? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2 Some Examples Of Eruptive Variable Stars . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
III Biosciences 25
1 Mitochondrial DNA (mtDNA) - The Small Circular Genome 26
1.1 Primary function of mitochondrial DNA . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.2 Mitochondrial DNA in species identification and phylogenetic studies . . . . . . . . . . . . . 27
1.3 Barcoding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.4 Mitochondrial DNA in Forensic Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.5 Mitochondrial DNA in Conservation and Evolutionary Biology: . . . . . . . . . . . . . . . . 29
1.6 Abundance and High Copy Number of mtDNA within each cell: . . . . . . . . . . . . . . . . 30
1.7 Improved Extraction Efficiency: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
CONTENTS
1.8 Amplification Reliability: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
IV Climate 32
1 Unveiling Earth’s Atmosphere: Layers, Composition, and Lifeline 33
1.1 Layers of the Earth’s Atmosphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
1.2 Formation of the Atmosphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
1.3 Physical Phenomenons caused by Earth’s Atmosphere. . . . . . . . . . . . . . . . . . . . . . 35
V General 38
1 The Ten Years of Science -Part II 39
1.1 Developments in Genetics Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
1.2 Neuroscience . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
1.3 Threats? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
v
Part I
Artificial Intelligence and Machine Learning
Difference Boosted Neural Network(DBNN) -
Part 1
by Blesson George
airis4D, Vol.1, No.5, 2023
www.airis4d.com
1.1 Introduction
The Difference Boosting Neural Network (DBNN) is a type of Bayesian neural network that integrates
the principle of imposed conditional independence within a Naive Bayes network. This integration generates a
classifier that has consistently produced positive results across numerous data types. In addition, this network
employs the novel technique of ”difference boosting” to update the prior classifier weight. It turns out that
DBNN is effective in various classification problem domains.
1.2 Conditional Independence in Naive Bayes classifier
A Bayesian classifier is a form of machine learning algorithm that makes predictions based on Bayes’
theorem. Frequently, Bayesian classifiers are used for text classification, spam filtering, and medical diagnosis.
In addition to being employed in natural language processing and computer vision, they are used as well in other
fields.The three primary categories of Bayesian classifiers are the Naive Bayes Classifier, the Bayesian Belief
Network (BBN) Classifier, and the Bayesian Logistic regression classifier.
The Naive Bayes classifier is a simple yet highly effective Bayesian classifier algorithm that can be applied
to a variety of classification tasks, including text classification, spam filtering, and image classification.
Bayes’ theorem is also known as Bayes’ Rule or Bayes’ law is the basic building block of Naive Bayesian
classifier. Bayes theorem is used to determine the probability of a hypothesis with prior knowledge. It depends
on the conditional probability. The formula for Bayes’ theorem is given as:
P (A|B) =
P (B|A) P (A)
P (B)
Where,
P (A|B) is Posterior probability: Probability of hypothesis A on the observed event B.
P (B|A) is Likelihood probability: Probability of the evidence given that the probability of a hypothesis
is true.
P (A) is Prior Probability: Probability of hypothesis before observing the evidence.
P (B) is Marginal Probability: Probability of Evidence.
1.3 Imposed conditional independence in DBNN
In Naive Bayes, the assumption of conditional independence is made, stating that the presence or absence
of a specific feature in a data point, given the class of the data point, is independent of the presence or absence
of any other feature. This assumption enables the Naive Bayes classifier to calculate the probability of a data
point belonging to a particular class by multiplying the probabilities of each feature of the data point belonging
to that class.
In terms of probability, the conditional independence is given as
P (A ∩ B|C) = P (A|C).P (B|C)
Conditional independence is basically the concept of independence, P (A ∩ B) = P (A) × P (B), applied to the
conditional model.
In real data, conditional independence fails for a number of reasons. The following factors are primarily
responsible for this failure:
1. Not all data is truly independent. In the actual world, data independence is rarely absolute.
2. The data is poorly represented. Even if the data are genuinely independent, it is possible that it is
not adequately represented in the training set. This can occur if the training set is too limited or not
representative of the target population.
3. Data is noisy. Typically, real-world data is chaotic. This indicates that the data may include errors or
outliers. The presence of noise can make it difficult to determine the underlying relationships between
features.
When conditional independence fails, the accuracy of the Naive Bayes classifier may decrease. Nonetheless, it
can be a useful classification tool, particularly when the data set is large and the features are complex.
1.3 Imposed conditional independence in DBNN
Conditional independence is established by generating a new set of features based on the existing features
in the dataset. In many cases, these derived features represent the probability density distribution of the original
features, revealing how the values of the features are distributed. The probability density distribution is obtained
through a combination of two processes: binning and histogram construction.
1.3.1 Binning of features
To perform binning, the feature is divided into intervals or bins, with the user specifying the size of each
bin. There are several approaches to determine the bin size. One common method is to use the square root
of the number of values (denoted as N) as the bin size. Alternatively, the bin size can be adjusted iteratively
based on specific criteria. It is also possible to have different bin sizes for different features, allowing for more
flexibility in the binning process.
1.3.2 Histogram of the distribution
After partitioning the feature into distinct bins, the histogram of the feature is constructed. This process
involves incrementing the count in each bin whenever an instance associated with that particular bin is encoun-
tered in the dataset. Finally, we will obtain a matrix structure, C
J,M
, with the rows representing the other J
features and the columns representing the histogram of their distributions in M bins for the given bin C
in
3
1.3 Imposed conditional independence in DBNN
1.3.3 Likelihood Estimation
For each class, we construct separate probability density distributions, resulting in distinct distributions
corresponding to each class. These distributions serve as the likelihoods for each class, which are essential for
computing Bayes’ theorem. The likelihood for each class C
k
to produce a feature U
m
with a value in the bin m
is given by the intersection
P (U
m
∩ C
k
)
The prior, on the other hand, is typically assumed by the user based on the domain knowledge and can take the
form of a uniform prior or a Gaussian distribution, among other possibilities.
With all the required components in place, we can now move forward with calculating the posterior
probabilities using Bayes’ theorem. The initially assumed prior values are updated using boosting methods, and
this process will be elaborated on in the forthcoming issue of the article.
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.
References
1. Conditional Independence — The Backbone of Bayesian Networks
2. INa
¨
ıve Bayes Classifier Algorithm
3. Properties of Naive Bayes
4. A Learning Algorithm based on Primary School Teaching Wisdom
5. Boosting the Differences: A Fast Bayesian classifier neural network
4
Decoding methods for language models
by Jinsu Ann Mathew
airis4D, Vol.1, No.5, 2023
www.airis4d.com
In the realm of artificial intelligence, language models have emerged as incredibly powerful tools, rev-
olutionizing various natural language processing tasks. These models, fueled by vast amounts of text data
and sophisticated algorithms, have the remarkable ability to generate coherent and contextually appropriate
human-like text. However, the true potential of a language model lies not only in its ability to generate text but
also in its decoding methods that determine how it selects and organizes that text.
Decoding methods are the crucial mechanisms by which language models convert learned patterns into
meaningful and intelligible output. Just like a storyteller carefully selects the right words and weaves them
together, decoding methods determine how the language model crafted its narratives. Decoding methods
determine the sequence of words, grammar structures, and semantic understanding employed by the language
model to produce output that aligns with the given input and desired task.
In this article, we delve into the world of decoding methods for language models, exploring various
techniques employed to generate high-quality and contextually appropriate text. By gaining an understanding
of these decoding methods, we unlock the immense potential of language models and effectively utilize them
across diverse applications, including machine translation, text summarizing, conversational agents, and many
other exciting domains.
2.1 Greedy Decoding
Greedy decoding is a straightforward strategy where the model selects the word with the highest probability
at each step. It prioritizes the most probable word at each stage without considering the global context. This
strategy prioritizes immediate likelihood, aiming to produce the most probable word without considering the
broader context or potential future implications.
The appeal of greedy decoding lies in its computational efficiency, as it requires minimal processing and
decision-making at each step. However, this approach can result in suboptimal outcomes, as the model may
overlook more suitable or coherent word choices in favor of those with higher immediate probabilities.
As an example, assume the model has been trained on a dataset of food-related phrases. Here’s a simplified
tree diagram representing the decoding process using greedy decoding:
In this example (Figure 1) , the model evaluates the available options and chooses the word ”Pizza” with
a probability of 0.6. It selects this word because it has the highest probability among the given choices. The
model generates the output sentence by appending the selected word to the input text.
However, it’s important to note that greedy decoding doesn’t consider the potential impact of the chosen
2.2 Beam Search
Figure 1: Example of word probabilities predicted from a language model, highlighting the choice made by
greedy search at the first timestep.
word on the overall coherence or future word choices. In this example, the model didn’t explore other possibilities
like ”Fruits” or ”Burger” which might have resulted in different, potentially more interesting, or contextually
appropriate sentences.
Limitations
Lack of Exploration: Greedy decoding prioritizes the most probable word at each step without consid-
ering alternative options. It lacks exploration of different paths or word choices, potentially leading to repetitive
or less diverse output.
Local Optimization: By focusing on immediate likelihood, greedy decoding may lead to sub optimal
overall sequence optimization. It prioritizes short-term probabilities without considering the global context or
long-term coherence.
Lack of Uncertainty Handling: Greedy decoding doesn’t account for uncertainty or probabilistic
variations in word selection. It ignores the probabilistic nature of language generation, resulting in deterministic
output that may not capture the full range of possibilities.
Insensitivity to Length or Structure: Greedy decoding does not consider the impact of generated text
length or sentence structure. It may produce outputs that are too short or lack appropriate grammatical structure.
Greedy decoding, despite its drawbacks, offers some advantages that make it a useful strategy in certain
scenarios.Its simplicity and efficiency make it computationally lightweight and well-suited for real-time text
generation applications. The deterministic nature of greedy decoding ensures consistent output, which can be
beneficial in contexts where reproducibility or consistency is important. By prioritizing high-probability word
choices, greedy decoding generates text that aligns well with the training data and follows observed patterns,
making it suitable for tasks that require text generation within the training distribution. Additionally, in situations
where immediate likelihood needs to be prioritized over long-range dependencies or diversity, greedy decoding
proves effective. Overall, while it has its limitations, greedy decoding remains a valuable and practical decoding
strategy in specific contexts where simplicity, efficiency, and immediate likelihood are prioritized.
6
2.2 Beam Search
(image courtesy:https://towardsdatascience.com/foundations-of-nlp-explained-visually-beam-search-how-it-works-1586b9849a24)
Figure 2: Beam Search example, with width = 2
2.2 Beam Search
Beam search algorithm selects multiple tokens for a position in a given sequence based on conditional
probability. The basic idea behind beam search is to maintain a set of partial output sequences, called ”beams,”
at each step of the search. The number of beams is a parameter that determines the breadth of the search. Instead
of exhaustively exploring all possible output sequences, beam search focuses on the most promising ones by
selecting the top-k sequences based on their probabilities.
Consider a straightforward example with a beam width of 2(Figure 2), using characters for simplicity.
Here, the algorithm starts with the ”¡START¿” token and generates probabilities for each word. It selects
the top choices for the first position, let’s say ”A” and ”C”. Moving to the second position, it runs the model
again, considering the previous choices (”A” and ”C”) and generating probabilities for the next character. It
selects the best character pairs based on combined probabilities, resulting in choices like ”AB” and ”AE”. This
process continues for subsequent positions, constraining the previous choices and generating probabilities for
the next character. It selects the best combinations based on the combined probabilities, like ”ABC” and ”AED”.
The algorithm repeats this until it picks an ”¡END¿” token, which concludes that branch of the sequence. It
keeps track of the best sequences throughout and predicts the one with the highest overall probability.
Beam Search improves upon Greedy Search in two key ways:
Expanding the Search Space: While Greedy Search only considers the single best word at each position,
Beam Search expands this by considering the best ’N’ words. Instead of committing to a single choice, Beam
Search maintains multiple candidate sequences, allowing for a broader exploration of possibilities.
Considering Context and Combination: Greedy Search treats each position in isolation, focusing only
on the best word for that position without considering the surrounding context. In contrast, Beam Search takes
the ’N’ best sequences so far and evaluates the probabilities of the combination of preceding words along with
the current word. This contextual consideration enables Beam Search to capture dependencies and produce
7
2.3 Top-K Sampling
more coherent and meaningful sequences.
Limitations:
Fixed Beam Width: The performance of beam search is highly dependent on the chosen beam width. A
narrower beam may limit exploration and lead to suboptimal results, while a wider beam increases computational
complexity. Selecting an optimal beam width can be challenging and may require experimentation.
Preference for Shorter Sequences: Beam search tends to favor shorter sequences due to the nature of
the pruning step. Longer sequences may have lower probabilities initially, leading to their elimination from the
beam. This bias towards shorter sequences can result in the generation of less descriptive or incomplete outputs.
Sensitivity to Language Model Biases: Beam search relies on a language model to calculate probabilities
and make predictions. If the language model itself is biased or has limitations, such biases or limitations can be
propagated in the generated sequences. Beam search alone does not address these biases and limitations.
2.3 Top-K Sampling
In top-k sampling, instead of considering all possible tokens, only the top-k most likely tokens are
considered for selection at each step. The value of k is a hyperparameter that determines the size of the set
of tokens to consider. Typically, a smaller value of k leads to more focused and deterministic outputs, while a
larger value of k allows for more randomness and diversity in the generated text.
The steps involved in top-k sampling are as follows:
1. First, the language model generates the probabilities for each possible token at a given step based on the
context and the model’s internal state.
2. The tokens are then sorted based on their probabilities, and the top-k tokens with the highest probabilities
are selected.
3. If the desired token (e.g., an end-of-sentence token) is among the top-k tokens, the sampling process
stops, and that token is chosen as the output.
4. If the desired token is not in the top-k tokens, the probabilities of the remaining tokens are renormalized
to ensure they sum up to one.
5. Finally, one token is randomly sampled from the remaining set of tokens based on their re-normalized
probabilities. The chosen token becomes the output for that step, and the process continues for the next
step.
By using top-k sampling, the model has a higher chance of selecting tokens that are likely to appear in the
generated text while still allowing for some level of randomness and diversity. It helps avoid the model getting
stuck in repetitive or deterministic patterns and produces more varied and interesting outputs. The value of the
hyperparameter k can be dynamically determined based on specific criteria. By adjusting the top-k parameter,
the size of the shortlist from which the model samples during token generation can be modified (Figure 3).
Setting top-k to 1 gives us greedy decoding.
Limitations
Lack of long-range coherence: Top-k sampling focuses on local probabilities and can sometimes result
in a lack of long-range coherence in the generated text. Since it only considers a subset of tokens at each step,
it may miss out on important dependencies or global structures in the language.
8
2.4 Nucleus Sampling
(image courtesy:https://docs.cohere.com/docs/controlling-generation-with-top-k-top-p)
Figure 3: Adjusting to the top-k setting
Fixed threshold for diversity: Top-k sampling uses a fixed threshold (k) to determine the number of
tokens to consider at each step. This fixed threshold may not always capture the optimal balance between
diversity and coherence. Some parts of the generated text may become repetitive or overly deterministic, while
other parts may lack diversity.
Inability to capture tail probabilities: Top-k sampling tends to focus on the most probable tokens and
may not fully capture the tail probabilities of the distribution. This means that rarer or less likely tokens have
reduced chances of being selected, potentially limiting the generation of unique or unexpected outputs.
Difficulty in controlling output length: Top-k sampling alone does not provide a straightforward way
to control the length of the generated text. It may result in varying text lengths, making it challenging to ensure
consistent output lengths.
2.4 Nucleus Sampling
The difficulty of determining the optimal top-k value in text generation has led to the adoption of a dynamic
approach known as Nucleus Sampling or top p sampling. This method addresses the challenge by shortlisting
tokens based on a cumulative probability threshold rather than a fixed number. Nucleus sampling selects the
smallest possible set of top-v tokens, where v represents the tokens whose cumulative probability does not
exceed a predefined value, often denoted as ”p.”
Nucleus sampling allows for a more flexible selection of tokens, as the number of tokens in the nucleus can
vary depending on the probabilities. This addresses the issue of fixed thresholds in top-k sampling and helps
capture a wider range of likely tokens, including both frequent and rare ones.
The threshold parameter ”p” controls the diversity of the generated text. A smaller value of ”p” results in
a narrower set of tokens, leading to more focused and deterministic outputs. On the other hand, a larger value
of ”p” expands the set of tokens, allowing for more randomness and diversity in the generated text (Figure 4).
9
2.5 Random Sampling
(image courtesy:https://docs.cohere.com/docs/controlling-generation-with-top-k-top-p)
Figure 4: A toy example with a top-p value of 0.15
Limitations
Sensitivity to threshold selection: Nucleus sampling relies on a threshold parameter ”p” to determine
the size of the nucleus, which can significantly impact the generated output. The choice of the threshold is
subjective and can be challenging to determine optimally. Small variations in ”p” can lead to significant changes
in the generated text’s diversity and coherence.
Potential loss of control: Nucleus sampling aims to generate diverse outputs by including a wider range
of tokens. However, this increased diversity can come at the cost of reduced control over the content and
coherence of the generated text. The inclusion of lower probability tokens from the nucleus may introduce less
relevant or less coherent outputs.
Difficulty in controlling output length: Similar to other sampling methods, nucleus sampling does not
provide a straightforward way to control the length of the generated text. The output length can vary, making it
challenging to ensure consistent text lengths in applications that require specific constraints.
2.5 Random Sampling
In random sampling, tokens are selected at each step purely based on their probabilities without any explicit
filtering or constraint. It is a simple and straightforward method of generating text that introduces a high level
of randomness and unpredictability into the output. The steps involved in random sampling are as follows:
1. The language model generates probabilities for each possible token at a given step based on the context
and the model’s internal state.
2. Each token is assigned a probability based on its likelihood of being the next token in the sequence.
3. Random sampling involves selecting one token at random from the set of possible tokens, using their
probabilities as weights. The selection process follows a probability distribution, where tokens with
higher probabilities have a higher chance of being selected, but there is still a possibility for tokens with
lower probabilities to be chosen.
4. The selected token becomes the output for that step, and the process continues for the next step, generating
the subsequent tokens based on the newly updated context.
Random sampling allows for maximum randomness in the generated text since it does not impose any
constraints or filtering based on probability thresholds or set sizes. It can produce diverse outputs and capture
10
2.5 Random Sampling
unexpected patterns or variations in the language model’s responses.
Limitations
Lack of control: Random sampling does not provide explicit control over the content, style, or quality of
the generated text. The generated outputs can vary widely and may include irrelevant or nonsensical content.
This lack of control makes it challenging to ensure the desired characteristics or constraints in the generated
text.
Coherence and quality issues: Random sampling can result in incoherent or low-quality outputs. Without
any filtering or constraints, the generated text may lack logical flow, contain grammatical errors, or produce
nonsensical combinations of words. This can impact the readability and usefulness of the generated content.
Unpredictable output: The randomness introduced by random sampling makes the generated output
highly unpredictable. Even with the same input and model state, different runs of random sampling can
produce vastly different results. This unpredictability can be problematic when consistency or reproducibility
is desired. Lack of sensitivity to probability distribution: Random sampling treats all tokens as equally
likely candidates. It does not consider the relative probabilities assigned to each token by the language model.
As a result, it may overlook the more likely or meaningful tokens in favor of less probable ones, leading to
suboptimal or less coherent outputs.
Summary
Greedy decoding: Selects the token with the highest probability at each step, resulting in locally optimal
choices but potentially sacrificing global coherence.
Beam search: Maintains a fixed-size set of top-scoring partial sequences, exploring multiple paths to
mitigate the limitations of greedy decoding.
Top-k sampling: Considers only the top-k most likely tokens at each step, striking a balance between
diversity and coherence in the generated text.
Nucleus sampling: Dynamically determines the set of tokens to consider based on a cumulative probability
threshold, allowing for flexible selection and controlled diversity.
Random sampling: Selects tokens at each step purely based on their probabilities without constraints,
introducing high randomness but lacking explicit control over the generated text.
In conclusion, this article explored various decoding strategies for text generation, including greedy
decoding, beam search, top-k sampling, top-p (nucleus) sampling, and random sampling. Using a good model
with bad decoding strategies or a bad model with good decoding strategies is not enough. To truly enhance the
quality and appeal of the generated content, it is crucial to strike a well-calibrated balance between the model’s
capabilities and the chosen decoding strategies.
References
Foundations of NLP Explained Visually: Beam Search, How It Works, Ketan Doshi, towardsdatascience,
April, 2021
Two minutes NLP — Most used Decoding Methods for Language Models, Fabio Chiusano, medium,
January, 2022
The Three Decoding Methods For NLP, James Briggs, medium, February, 2021
11
2.5 Random Sampling
Understanding greedy search and beam search, Jessica L
´
opez, medium, February, 2021
Decoding Strategies that You Need to Know for Response Generation, Vitou Phy, medium, July,2020
Greedy Search Decoding For Text Generation In Python, Jason LZP, medium, May, 2022
What is Beam Search? Explaining The Beam Search Algorithm, Matt Payne, September, 2021
Beam Search Decoding For Text Generation In Python , Jason LZP, medium, July, 2022
Top-k & Top-p, cohere
5 Text Decoding Techniques that every “NLP Enthusiast” Must Know , Prakhar Mishra, medium, De-
cember, 2021
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.
12
Part II
Astronomy and Astrophysics
The Galaxy Revelation and a Growing
Universe
by Linn Abraham
airis4D, Vol.1, No.5, 2023
www.airis4d.com
1.1 What are Nebulae?
Imagine looking out into the night sky with just your eyes and seeing hundreds upon hundreds of small
and big stars spread out over the night sky. By consistently observing some of these over several weeks or
months we can identify the planets which have a different motion in the sky. On closer observation we can find
several hazy patches of light in the sky. Andromeda (M31) is one such example of a fuzzy and extended object,
the Large and Small Magellanic clouds being the other. All such objects were termed ’nebulae’ by the early
astronomers. What are these strange fuzzy objects seen in the sky? What makes them different?
On days that are clear or with a long exposure photograph something else becomes visible. A band of light
in the sky which the ancient greeks called the ”Milky way”. The term ”galaxy” is greek for milky circle. When
our telescopes improved we could see that rather than being a diffuse cloud the milky way consisted of a vast
collection of stars flattened into a disk like structure seen edge-on.
Returning to our discussion of nebulae, in modern astronomy the term nebulae is now reserved for vast
clouds of gas and dust in space. Thus all of the nebulae that were visible to early astronomers were not in fact
nebulae. Although there are a few nebulae that are visible in the night sky like the Orion nebulae. Of the other
type of nebulae that were visible to early astronomers, the most common were the spiral shaped nebulae.
1.2 Galaxies Within Galaxies: The Island Universe Hypothesis
There is a story about how our understanding of the size of the visible universe changed sharply after the
1920’s. Until then the Milky-way galaxy was thought to be the entire universe. It all started with the well
known philosopher Immanuel Kant who speculated that some of the faint cloudy patches observed in the night
sky might be separate ”island universes” distinct from our own. This hypothesis turned from just a hypothesis
into a fully blown debate among astronomers as more data and observations poured in.
The Great Debate in astronomy also called the Shapley-Curtis debate of 1920 was precisely on this topic.
Harlow Shapley is the astronomer who is credited with discovering the true shape and size of our own Galaxy.
Shapley believed that these nebulae were relatively small and lay within the bounds of our own universe, while
Curtis held that they were in fact independent galaxies, very large but distant.
1.3 Edwin Hubble Settles the Dust
Figure 1: Messier 82 (also known as NGC 3034, Cigar Galaxy or M82) is a starburst galaxy approximately
12 million light-years away in the constellation Ursa Major. The figure is a mosaic image taken by the Hubble
Space Telescope of Messier 82, combining exposures taken with four colored filters that capture starlight from
visible and infrared wavelengths as well as the light from the glowing hydrogen filaments.
15
1.3 Edwin Hubble Settles the Dust
Figure 2: The entire Orion Nebula in a composite image of visible light and infrared; taken by Hubble Space
Telescope in 2006
16
1.3 Edwin Hubble Settles the Dust
Figure 3: The ”Great Spiral Nebula” in the constellation Andromeda (1902 photograph). The Debate was over
whether this was a cloud of gas and dust or a distant galaxy.
17
1.3 Edwin Hubble Settles the Dust
1.3 Edwin Hubble Settles the Dust
Henrietta Levitt’s discovery of the period-lumniosity relationship for Cepheid variables in the Small
Magellanic Clouds was critical to solving the debate. In the later part of the year 1920, Edwin Hubble used the
100-inch Hooker Telescope at Mount Wilson Observatory, which was the largest operational telescope at the
time to observe Cepheids in the Andromeda nebulae. He used them as distance indicators and measured the
distance to the nebulae using those. He found that the Andromeda nebulae was many times more distant than
the majority of stars visible in the night sky. He thus established that Curtis was right and that Immanuel Kant’s
speculations were justified. We now know that our Milky Way is just one out of the 2 trillion or more galaxies
in the observable universe.
1.4 Galaxies that are Moving Away from Us
Although the redshift of starlight was observed before Hubble, Edwin Hubble systematically observed
galaxies and measured their distances using the same technique that he used for establishing the size of the
Universe. He then analyzed their redshift and derived the famous Hubble Law which showed that all galaxies
are moving away from us at an accelerating pace propotional to the distance towards them. The Andromeda
galaxy is an exception to this and scientists now believe that the Andromeda galaxy is on a collision course
towards our own galaxy.
1.5 The Mystery Continues..
Today we know that the nebulae that early astronomers observed are actually galaxies. Vast cities of stars,
gas and dust floating in space just like the one in which we are living. However even after 100 years since
Hubble’s discoveries and more detailed observations of galaxies, there is still a lot more to be known about
them. The wide variety of shapes that they come in puzzle astronomers to this day. Even though Hubble showed
that the millions of galaxy shapes can be reduced to a few categories (The Hubble Classification system), there
is still no underlying theory that can explain all the features that are observed in galaxies.
Figure 4: The Hubble Space Telescope Galaxy collection showing the variety of galaxy shapes. Image Credit:
NASA
18
1.6 References
Figure 5: The Hubble sequence: classification of galaxies. Image Credit: Wikipedia.org
1.6 References
1. Shu, Frank H. The Physical Universe: An Introduction to Astronomy. 9. print. A Series of Books in
Astronomy. Sausalito, Calif: Univ. Sience Books, 1982.
2. Rai, Choudhuri Arnab. Astrophysics for Physicists.
3. Whitrow, G. J. “Kant and the Extragalactic Nebulae” 8 (March 1967): 48.
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
sources from PanSTARRS optical images. He has used data from a several large astronomical surveys including
SDSS, CRTS, ZTF and PanSTARRS for his research.
19
Eruptive Variable Stars
by Sindhu G
airis4D, Vol.1, No.5, 2023
www.airis4d.com
2.1 What Are Eruptive Variable Stars?
Eruptive variables are stars that undergo eruptions on their surfaces, which can manifest as flares or mass
ejections. These eruptions release tremendous amounts of energy and material into space, leading to significant
changes in the star’s brightness and activity. In eruptive variables the variations in brightness are due to violent
processes and flares occurring in their chromospheres and coronae. These variations are often accompanied
by shell events, mass outflows in the form of stellar winds, and interactions with the surrounding interstellar
medium. Eruptive variable stars exhibit irregular or semi-regular brightness variations, which can be attributed
to material being lost from the star or, in some cases, being accreted onto it. Contrary to their name, these
variations are not necessarily associated with explosive events.
Eruptive variables, with their flares and mass ejections, are of great interest to astronomers. They provide
insights into the dynamic nature of stars, the physics of magnetic fields, and the mechanisms driving stellar
activity. Studying these events helps us better understand stellar evolution, the influence of magnetic fields on
stellar atmospheres, and the impact of stellar eruptions on the surrounding environment. In recent years, space
telescopes and advanced observation techniques have allowed for more detailed studies of eruptive variables
across different wavelengths, from X-rays to radio waves.
2.2 Some Examples Of Eruptive Variable Stars
2.2.1 Flare stars
Flare stars (Figure 1 and Figure 2) are a type of eruptive variable that are characterized by sudden and
intense increases in brightness. They are most commonly found among low-mass stars, particularly red dwarfs.
Flares on flare stars are similar in nature to solar flares. Just like on the Sun, the flares on flare stars are caused
by the release of magnetic energy stored in their atmospheres. These flares can cause a sudden increase in
brightness by up to two magnitudes, which corresponds to a brightness increase of approximately six times,
within a matter of seconds. After the rapid increase in brightness, flare stars gradually fade back to their normal
brightness levels within a period of typically half an hour or even less. This rapid rise and fall in brightness
make flare stars highly variable objects. Flare stars are also known as UV Ceti variables. Flare stars are typically
classified as late M through late K spectral types. These correspond to relatively cool temperatures ranging
from about 2500 to 4000 K. The spectral features of flare stars often exhibit emission lines of hydrogen and
2.2 Some Examples Of Eruptive Variable Stars
Figure 1: Right: a flare on the fainter component of the visual binary star system Kr uger 60. The images are
taken over an interval of a few minutes. Source: Sproul Observatory photograph.
Figure 2: UV Ceti in X-rays, observed with the ROSAT High Resolution Imager. Source: NASA.
calcium, indicating active chromospheres. Some of the nearby red dwarf flare stars include Proxima Centauri,
the closest star to our solar system, and Wolf 359.
2.2.2 FU Orionis variables
FU Orionis variables (Figure 3 and Figure 4) are a specific type of eruptive variable stars characterized by
their distinct behavior. These stars undergo gradual increases in brightness, typically by about 6 magnitudes,
over the course of several months. This initial brightening phase is followed by either a period of nearly
constant maximum brightness that can be sustained for long periods or a slow decline in brightness by 1-2
magnitudes. After the initial brightness increase, FU Orionis variables may remain at maximum brightness
for extended periods, ranging from years to decades. Alternatively, FU Orionis variables may slowly fade or
decline in brightness after reaching maximum. The fading process can occur over a significant span of time,
and the decline in brightness is typically more gradual compared to the initial brightening phase. At maximum
brightness, the spectral types of FU Orionis variables range from Ae(alpha) to Gpe(alpha). However, following
an outburst, there is a gradual development of an emission spectrum, and the spectral type becomes later. It is
notable that all presently known FU Orionis variables are associated with reflecting cometary nebulae.
2.2.3 RS Canum Venaticorum variables
RS Canum Venaticorum (RS CVn) variable type consists of close binary stars with active chromospheres
that can lead to the formation of large stellar spots. The variability in RS CVn binaries can occur on various
21
2.2 Some Examples Of Eruptive Variable Stars
Figure 5: R Coronae Borealis. Source: Project VS-COMPAS.
timescales. On shorter timescales, the variability is often associated with the orbital period of the binary system.
As the stars orbit each other, their mutual gravitational influence can cause changes in the observed brightness.
In some cases, eclipses may occur when one star passes in front of the other, leading to additional variability
and characteristic light curve patterns. On longer timescales, RS CVn binaries exhibit variability due to changes
in the activity level of the primary star, including fluctuations in sunspot activity. These variations can occur
over months to years and are associated with the magnetic cycle of the primary star. The typical brightness
fluctuation of RS CVn variables is around 0.2 magnitudes.
2.2.4 R Coronae Borealis variables
R Coronae Borealis (RCB)(Figure 5) stars are hydrogen-poor, carbon- and helium-rich, high-luminosity
stars that exhibit characteristics of both eruptive and pulsating variables. They belong to the spectral types
Bpe-R. RCB stars display slow and non-periodic fading in brightness, with magnitudes decreasing by 1-9 in
the visible band (V). These fading events can last for varying durations, ranging from a month or more to
several hundred days. In addition to the fadings, RCB stars also exhibit cyclic pulsations. These pulsations
are characterized by variations in brightness with amplitudes ranging from several tenths of a magnitude. The
periods of these pulsations fall within the range of 30-100 days.
2.2.5 WolfRayet variables
Classic Population I Wolf-Rayet (WR) (Figure 6)stars are massive, hot stars that are known to exhibit
variability in their brightness. The causes of this variability can be attributed to several factors, including binary
interactions and the presence of rotating gas clumps around the star. These WR stars are characterized by their
broad emission line spectra, which prominently feature lines of helium (He), nitrogen (N), carbon (C), and
oxygen (O). Wolf-Rayet stars display irregular light changes with amplitudes up to 0.1 magnitudes in the V
band. These variations are likely caused by physical processes, specifically non-stable mass outflows from their
atmospheres.
23
2.2 Some Examples Of Eruptive Variable Stars
Figure 6: Wolf-Rayet Star 124. Source: Hubble Legacy archive, NASA, ESA.
References:
Understanding Variable Stars, John R Percy, Cambridge University Press.
Variable Stars
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
Variable star
UV Ceti and the flare stars
Variable Star Classification and Light Curves
FU Orionis
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.
24
Part III
Biosciences
Mitochondrial DNA (mtDNA) - The Small
Circular Genome
by Geetha Paul
airis4D, Vol.1, No.5, 2023
www.airis4d.com
Mitochondrial DNA (mtDNA) is a small circular genome exclusively found within the mitochondria,
which are energy-producing organelles in cells. A genome refers to an organism’s complete genetic material
(DNA or RNA). It contains all the instructions and information necessary for that organism’s development,
functioning, and reproduction. The genome blueprint determines an organism’s traits, characteristics, and
biological processes. It is organized into chromosomes and consists of a sequence of nucleotides (A, C, G, and
T in DNA) that form the genetic code. Unlike nuclear DNA, which is inherited from both parents, mtDNA is
solely inherited from the mother. The role of mtDNA is essential for the proper functioning of the mitochondria
and, consequently, for the overall functioning of the cell..
Image courtesy: https://en.wikipedia.org/wiki/Mitochondrial DNA
Figure 1: Shows the entire cell, the mitochondrion and the circular mitochondrial DNA.
1.1 Primary function of mitochondrial DNA
1.1 Primary function of mitochondrial DNA
Mitochondrial DNA (mtDNA) is the genetic material within mitochondria, producing ATP, the cell’s
energy source. Electrons are transferred in electron transport chains through oxidative phosphorylation, leading
to ATP production. mtDNA encodes a subset of proteins, including those involved in the electron transport
chain, which is crucial for ATP generation. The absence of mtDNA would impair ATP production and result
in cellular and organismal deficiencies. The length of mtDNA is approximately 16,500 base pairs, distinct
from nuclear DNA. Within mtDNA, 37 genes contribute to mitochondrial function. Among these genes, 13 are
protein-coding genes that provide instructions for enzymes involved in oxidative phosphorylation. This process
is responsible for ATP synthesis, the primary energy source for cells. mtDNA also contains genes encoding
transfer RNA (tRNA) and ribosomal RNA (rRNA), which aid in protein synthesis within mitochondria.
Image courtesy:https://www.sciencedirect.com/science/article/pii/S0925443909002427
Figure 2: Shows fundamental role of mitochondria in the generation of energy (ATP), these organelles are also
the main producers of oxygen free radicals.
1.2 Mitochondrial DNA in species identification and phylogenetic studies
Mitochondrial DNA (mtDNA) has become a valuable tool for species identification and phylogenetic
studies due to its unique properties. Compared to nuclear DNA, mtDNA is maternally inherited and has a
higher mutation rate, which makes it a valuable marker for resolving relationships within and between species.
27
1.3 Barcoding
Image courtesy: Folia Biologica (Krak
´
ow), vol. 69 (2021), No3 http://www.isez.pan.krakow.pl/en/folia- biologica.html
Figure 3: Stages of mtDNA analysis: A – electrophoretic separation of PCR products, B – results of sequencing
purified PCR products, C – bioinformatic analysis involving a comparison of the sequences obtained with
database resources and analysis of polymorphisms between the test sequence and reference sequences, D –
phylogenetic analysis based on the sequences obtained (our own results).
1.3 Barcoding
DNA barcoding is a molecular technique that uses a short DNA sequence to identify species. In animals,
the mitochondrial cytochrome c oxidase subunit I (COI) gene has become the standard barcode for species
identification. The COI gene has a high degree of sequence variation between species and a low degree of
variation within species, which makes it a reliable marker for species identification. By comparing the COI
sequence of an unknown specimen to a reference database of known species, one can identify the species to
which the specimen belongs. An unquestionable advantage of molecular methods over traditional ones is that
identification can be based on trace quantities of material, including highly processed or degraded material
(bone fragments, teeth, or fur.
Moreover, bioinformatic tools supporting molecular analyses limit the need to test reference material, as
the results can be compared to sequences deposited in databases. The routine analysis includes:
DNA isolation.
Amplification of target fragments.
Sequencing of PCR products.
A bioinformatic analysis of the results.
Phylogenetics: mtDNA has been used extensively in phylogenetic studies to reconstruct evolutionary relation-
ships among species. By comparing the mtDNA sequences of different species, one can infer the evolutionary
relationships between them. Phylogenetic analyses can also help resolve taxonomic uncertainties and identify
cryptic species, which are morphologically similar but genetically distinct species.
1.4 Mitochondrial DNA in Forensic Analysis
Forensic analysis: mtDNA can be used in forensic investigation to identify the species of biological
samples. For example, mtDNA analysis determines the species of hair, bone, and blood samples found at crime
scenes. mtDNA is also helpful in identifying the origin of seized animal products, such as ivory and rhino horn.
Mitochondrial DNA analysis offers a significant advantage to the forensic geneticist when it has not been
possible to obtain a standard nuclear DNA profile, such as in severely degraded DNA, bones and hair shafts.
28
1.5 Mitochondrial DNA in Conservation and Evolutionary Biology:
Mitochondria are inherited through the maternal line and will be present in both male and female living children
offering tools to assist in familial links over many generations, such as in missing person cases where there is
no direct family reference. Sequencing of mitochondria requires a good knowledge of bioinformatic tools to
ensure correct sequence and haplotype information. Ancient and degraded DNA provides many challenges in
mitochondrial sequencing because of the competing presence of contaminants, amplification product verification
challenges and interfering sequence copies from other parts of the genome, all requiring specialist provision
and expertise.
Image courtesy: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8457767/
Figure 4: Diagram of the basic structure of the mitochondrial genome to highlight the hypervariable regions
(HV1, HV2 and HV3) that lie within the ‘control’ or D-loop area of this genome that are of particular interest
to forensic scientists.
Mitochondrial DNA has a region which is a non-coding stretch of 1121 base pairs comprising the D-loop
and a transcription promoter region — the ’control region’. The D-loop consists of three strands of DNA, one
complementary to one of the strands, holding it apart and forming the displacement (D) loop. The D-loop area
of this genome is particularly interesting to forensic scientists.
1.5 Mitochondrial DNA in Conservation and Evolutionary Biology:
Conservation biology: mtDNA can be used to monitor the genetic diversity of endangered species and
identify genetically distinct populations. By comparing the mtDNA sequences of different people, one can infer
the evolutionary history of the species and identify populations that need conservation measures. mtDNA has
been used to study the evolution of various traits, such as body size, colouration, and behaviour. By comparing
the mtDNA sequences of different species or populations, one can infer the evolutionary history of these traits
and how they have evolved.
29
1.6 Abundance and High Copy Number of mtDNA within each cell:
Image courtesy: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8457767/
Figure 5: Illustration of MtDNA inheritance over four generations indicates ancestral females that share the
same MtDNA haplotype as the children at the base.
1.6 Abundance and High Copy Number of mtDNA within each cell:
Mitochondria contain multiple copies of mtDNA within each cell, making it readily available for analysis.
This high copy number makes mtDNA easier to extract and amplify, even from degraded or limited biological
samples. In fields such as forensic analysis, where DNA samples may be compromised or limited in quantity,
the abundance of mtDNA can be advantageous.
1.7 Improved Extraction Efficiency:
Multiple copies of mtDNA within each cell increase the chances of successful DNA extraction. When
extracting DNA from a sample, having a higher starting copy number of mtDNA increases the likelihood of
obtaining sufficient DNA for downstream analysis. The mtDNA is particularly useful in cases where the DNA
material is limited, degraded, or mixed with inhibitors that can interfere with the extraction process.
1.8 Amplification Reliability:
The high copy number of mtDNA makes it more amenable to amplification techniques, such as polymerase
chain reaction (PCR). PCR is a commonly used method to amplify specific DNA regions for analysis. With
mtDNA, the chance of successful amplification is higher due to the increased number of target molecules
available. It is especially advantageous when dealing with samples that contain a low amount of DNA or
degraded DNA, as the higher starting copy number compensates for potential DNA damage or degradation. In
situations where the amount of DNA is limited or degraded, the high copy number of mtDNA enhances the
sensitivity of detection methods. Even if only a tiny fraction of the mtDNA molecules in a sample are intact
or accessible, the multiple copies can still provide enough templates for successful analysis. It is precious in
forensic investigations, where DNA samples recovered from crime scenes are often scarce, mixed, or fragmented.
Heteroplasmy refers to different mtDNA variants within an individual due to the coexistence of multiple mtDNA
populations. The high copy number of mtDNA allows for detecting low-frequency mtDNA variants in a sample.
By analyzing many mtDNA molecules, it becomes possible to identify and study the heteroplasmic nature of
30
1.8 Amplification Reliability:
mtDNA. It is relevant in forensic analysis when investigating mixed DNA samples or tracing maternal lineages.
mtDNA’s higher resistance to degradation than nuclear DNA is an additional advantage. Mitochondria have
protective structures and mechanisms that help maintain the integrity of mtDNA even in challenging conditions.
Consequently, mtDNA can be more readily preserved in samples that have been subjected to degradation factors
such as environmental exposure, extreme temperatures, or ageing. This attribute makes mtDNA valuable for
analyzing ancient DNA samples, archaeological remains, or samples with compromised DNA quality.
Conclusion
In conclusion, mitochondrial DNA is important in species identification, phylogenetics, forensic analysis,
conservation, and evolutionary biology. The unique properties of mtDNA make it a valuable tool for resolving
relationships within and between species and understanding the evolutionary history of life on Earth.
References
https://www.sciencedirect.com/topics/biochemistry-genetics-and-molecular-biology/mitochondrial-dna
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6562384/
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8457767/
Hebert P.D.N, Cywinska A, Ball S.L& DeWaard J.R. 2003 Biological identifications through DNA bar-
codes. Proc. R. Soc. B. 270, 313–321.doi:10.1098/rspb.2002.2218. .
Folia Biologica (Krak
´
ow), vol. 69 (2021), No3 http://www.isez.pan.krakow.pl/en/folia-biologica.html
https://doi.org/10.3409/fb
69-3.12
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.
31
Part IV
Climate
Unveiling Earth’s Atmosphere: Layers,
Composition, and Lifeline
by Robin Jacob Roy
airis4D, Vol.1, No.5, 2023
www.airis4d.com
Earth’s atmosphere is a thin layer of gasses that surrounds our planet. It is held in place by the Earth’s
gravity. The atmosphere is made up of about 78% nitrogen, 21% oxygen, and 1% other gases, such as argon,
carbon dioxide, and water vapor. It is essential for life on Earth, as it provides us with oxygen to breathe,
protects us from harmful radiation, and helps to regulate the planet’s temperature. This invisible blanket extends
thousands of kilometers above the Earth’s surface and is divided into several distinct layers, each with its
unique characteristics and crucial roles. Exploring the layers of Earth’s atmosphere offers us insights into the
complexity and fragility of our planet’s life-sustaining environment.
1.1 Layers of the Earth’s Atmosphere
The atmosphere is divided into layers based on the following criteria:
Temperature: The temperature of the atmosphere decreases with altitude in the troposphere, increases
with altitude in the stratosphere, and then decreases again with altitude in the mesosphere. The thermo-
sphere and exosphere are very thin and the temperature varies greatly.
Density: The density of the atmosphere decreases with altitude. The troposphere is the densest layer,
followed by the stratosphere, mesosphere, thermosphere, and exosphere.
Composition: The composition of the atmosphere also varies with altitude. The troposphere is made
up of about 78% nitrogen, 21% oxygen, and 1% other gases. The stratosphere is made up of about 90%
nitrogen and 10% oxygen. The mesosphere is made up of about 80% nitrogen and 20% oxygen. The
thermosphere is made up of about 99% hydrogen and 1% helium. The exosphere is made up of very thin
gases, such as hydrogen, helium, and oxygen.
Chemical reactions: The chemical reactions that occur in the atmosphere also vary with altitude.
The troposphere is where most of the Earth’s weather occurs. The stratosphere is where the ozone
layer is located. The mesosphere is where meteors burn up. The thermosphere is where the aurora
borealis (northern lights) and aurora australis (southern lights) occur. The exosphere is where the Earth’s
atmosphere meets space.
The atmosphere is divided into five major layers:
1. Troposphere: The troposphere is the lowest layer of Earth’s atmosphere, extending from the surface up
to an average height of 12 kilometers. It is where weather phenomena occur, including the formation of
1.2 Formation of the Atmosphere
clouds, precipitation, and wind patterns. As we ascend within the troposphere, the temperature generally
decreases. The troposphere also contains the majority of Earth’s atmospheric mass and is essential for
sustaining life on the planet.
2. Stratosphere: Above the troposphere lies the stratosphere, extending from approximately 12 to 50
kilometers above the Earth’s surface. The stratosphere is characterized by the presence of the ozone
layer, a region rich in ozone molecules that absorb and shield us from harmful ultraviolet (UV) radiation
emitted by the Sun. This protective layer plays a vital role in safeguarding living organisms from potential
damage and contributes to maintaining the planet’s overall climate.
3. Mesosphere: Beyond the stratosphere lies the mesosphere, extending from around 50 to 85 kilometers
above the Earth’s surface. The mesosphere is the layer where meteoroids burn up upon entry, creat-
ing beautiful streaks of light known as meteors. It is also the coldest region in Earth’s atmosphere,
with temperatures decreasing as altitude increases. Additionally, the mesosphere provides an essential
boundary for the ionosphere, a layer responsible for reflecting radio waves and enabling long-distance
communication.
4. Thermosphere: Situated above the mesosphere is the thermosphere, extending from approximately 85
kilometers up to 600 kilometers or more. In this layer, solar radiation significantly heats the sparse air
molecules, resulting in high temperatures. However, despite the intense heat, the thermosphere would not
feel warm to us due to the extremely low density of particles. The International Space Station (ISS) and
many satellites orbit within this layer, benefiting from its low atmospheric drag.
5. Exosphere: The exosphere represents the outermost layer of Earth’s atmosphere, gradually transitioning
into the vacuum of space. It extends beyond the thermosphere and merges with the interplanetary medium.
The exosphere is sparsely populated with gas molecules, and collisions between particles are rare. Some
atoms and molecules can reach escape velocity and escape into space, contributing to the continuous loss
of Earth’s atmosphere over vast timescales.
Figure 1 illustrates the different layers of Earth’s atmosphere.
1.2 Formation of the Atmosphere
The Earth’s atmosphere formed over billions of years through a variety of processes. The first atmosphere
was likely created by the outgassing of gases from the Earth’s interior. These gases included water vapor, carbon
dioxide, nitrogen, and methane. As the Earth cooled, the water vapor condensed to form the oceans. The carbon
dioxide was absorbed by the oceans and rocks, and the nitrogen and methane remained in the atmosphere.
The atmosphere continued to evolve over time as the Earth was bombarded by meteorites and comets.
These impacts released gases into the atmosphere, including water vapor, carbon dioxide, and nitrogen. The
water vapor condensed to form the oceans, and the carbon dioxide was absorbed by the oceans and rocks. The
nitrogen remained in the atmosphere.
The first life on Earth evolved about 3.5 billion years ago. These early life forms were able to photosynthe-
size, which means they used sunlight to convert carbon dioxide and water into oxygen and carbohydrates. This
process released oxygen into the atmosphere, which gradually replaced the carbon dioxide.
1.3 Physical Phenomenons caused by Earth’s Atmosphere.
The Earth’s atmosphere is not only a protective shield and vital for sustaining life, but it also gives rise to a
multitude of fascinating physical phenomena that shape our planet’s environment and influence various natural
35
1.3 Physical Phenomenons caused by Earth’s Atmosphere.
processes. Let’s delve into some of these remarkable phenomena caused by the effects of the atmosphere:
Weather Patterns: The atmosphere plays a central role in the formation of weather patterns. The
interaction between air masses with different temperatures, humidity levels, and pressure systems leads
to the creation of diverse weather conditions such as storms, hurricanes, tornadoes, and monsoons. These
atmospheric dynamics are responsible for the distribution of rainfall, the formation of clouds, and the
circulation of winds on both regional and global scales.
Greenhouse Effect: The greenhouse effect is a natural phenomenon enabled by certain gasses in the
Earth’s atmosphere, such as carbon dioxide (CO2) and methane (CH4). These gasses allow sunlight
to penetrate the atmosphere and reach the Earth’s surface, where it is absorbed and re-emitted as heat.
However, they also trap some of the heat, preventing it from escaping back into space. This process
helps to regulate the planet’s temperature, making it suitable for life. However, human activities have
significantly increased greenhouse gas concentrations, leading to enhanced global warming and climate
change.
Atmospheric Optics: The atmosphere gives rise to a range of optical phenomena that captivate and
inspire us. The scattering of sunlight by air molecules and suspended particles leads to the blue color
of the sky during the day. When sunlight passes through water droplets in the air, it can result in the
formation of rainbows, halos, and glories. Additionally, atmospheric conditions can create stunning
phenomena like sunsets and sunrises, where the scattering of light by particles and the bending of light
rays create a tapestry of vibrant colors.
Atmospheric Pressure: The weight of the Earth’s atmosphere exerts pressure on the surface, resulting
in atmospheric pressure. As altitude increases, the density of air decreases, leading to a decrease in
atmospheric pressure. This pressure gradient is responsible for driving winds and air currents, which play
a crucial role in redistributing heat and moisture around the planet. Variations in atmospheric pressure
also influence weather patterns and can be used to forecast changes in weather conditions.
Aurora Borealis and Aurora Australis: The dazzling light displays known as the auroras, or the
Northern and Southern Lights (Aurora Borealis and Aurora Australis), are awe-inspiring phenomena
caused by the interaction between charged particles from the Sun and Earth’s magnetic field. When these
particles enter the atmosphere near the poles, they collide with atoms and molecules, releasing energy in
the form of colorful lights. The result is a breathtaking display of shimmering green, red, purple, and
blue hues dancing across the night sky.
Atmospheric Acoustics: The properties of the atmosphere also impact the transmission of sound waves.
The density and composition of air affect the speed and intensity of sound. For example, sound travels
faster in denser mediums, such as in colder air near the ground. Atmospheric conditions, such as
temperature inversions, can cause sound to be refracted or trapped, leading to the phenomenon of sound
carrying over long distances, known as ”acoustic mirages” or ”Fata Morgana.”
The Earth’s atmosphere is a complex system that governs the physical and chemical processes on our planet. It
comprises a complex system of layers, each serving a distinct purpose and collectively interacting to sustain life
on our planet. From the dynamic troposphere, where weather phenomena occur, to the protective stratosphere
with its ozone layer, and the thermosphere that facilitates satellite operations, each layer plays a critical role
in maintaining the delicate balance necessary for life’s existence. Understanding and preserving the health of
Earth’s atmosphere are crucial endeavors as we navigate the challenges of climate change and seek to ensure a
sustainable future for generations to come.
36
1.3 Physical Phenomenons caused by Earth’s Atmosphere.
References:
Earth’s atmosphere
Parts of the Atmosphere
Layers of the Atmosphere.
Atmosphere Formation.
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.
37
Part V
General
The Ten Years of Science -Part II
by Ninan Sajeeth Philip
airis4D, Vol.1, No.5, 2023
www.airis4d.com
1.1 Developments in Genetics Research
Genetics has significantly transformed in the past decade, with discoveries and technological advancements
revolutionising the field. These developments have greatly expanded our understanding of genetics and have
provided us with new tools for studying and manipulating DNA. These developments offer exciting prospects
for future research and treatment of genetic diseases. With continued progress in genetics, we aim to unlock
new insights into the role of genetics in health and disease and develop new therapies to improve human health.
One of the most significant genetic developments in the past decade was the discovery of CRISPR/Cas9 gene
editing technology. This revolutionary tool allows scientists to edit DNA precisely, opening up new possibilities
for treating genetic diseases, developing new crops, and even creating designer babies. The technology can edit
the genomes of various organisms, including humans, mice, and plants. It has already shown great promise in
treating genetic diseases like sickle cell anaemia.
The past decade has also seen significant progress in our understanding of epigenetics, which refers to
the study of changes in gene expression primarily due to environmental or habitual changes and not related
to alterations in the DNA sequence itself. Such changes include modifications to the histone proteins that
package DNA, as well as changes in the methylation of DNA. These modifications can significantly affect gene
expression, and their study has shed light on the role of environmental factors in gene regulation and disease
susceptibility. Factors such as diet, stress, and exposure to toxins can result in epigenetic changes in individuals,
which can propagate to offspring, making it a trait characteristic. Epigenetic changes seem to play a role in
various diseases, including cancer, heart disease, and diabetes. Some examples of epigenetic changes are:
DNA methylation: DNA methylation is the mechanism that adds a methyl group to a DNA molecule.
Methyl groups can silence genes, making them less likely to be expressed. Ageing is one example of
methylation.
Histone modification: Histone modification adds or removes chemical groups to histone proteins spooled
around DNA. Histone modifications can affect how tightly DNA is wound, affecting gene expression.
RNA interference: RNA interference is the process of blocking the production of a protein by blocking
its RNA message. Environmental factors, such as exposure to toxins, can trigger RNA interference.
The good thing about Epigenetic changes is that most of them are reversible, meaning following good
habits and healthy environmental conditions can slow down such changes.
Epigenetic changes are adaptation methods nature discovered to protect lifeforms on the planet. A good
example is the coloured skin. Why are people in the tropics coloured while those closer to the poles are
1.1 Developments in Genetics Research
fair-skinned? In the tropics, as you move closer to the equator, the exposure to Ultraviolet (UV) radiation from
the sun is much higher than near the poles. It can damage the DNA and kill the species. To protect against any
damage, the species’ body produces enzymes that methylate DNA that prevent the genes from being expressed.
The genes in the skin silenced by methylation produce melanin, the pigment that gives skin its colour. More
melanin means darker skin and better protection from UV. The colour of the iris of the eyes is also determined
by the type and amount of melanin produced. Eumelanin and Pheomelanin pigments make them blackish or
red to yellow coloured. Fascinating science, right?
Another significant development in genetics has been the explosion of genomic data. Completing the
Human Genome Project in 2003 marked an important milestone in the field. Though it used to be extremely
expensive, advances in sequencing technology have dramatically reduced the cost of genome sequencing in the
past decade, making it more accessible to researchers and clinicians. Investing in genome sequencing has led to
a wealth of genomic data, which could identify genetic risk factors for various diseases and develop personalised
medicine approaches.
Researchers published the first-ever synthetic genome in 2012, created by stitching together pieces of DNA
in the laboratory. This technology has the potential to revolutionise the field of synthetic biology, allowing
scientists to develop entirely new organisms or modify existing ones for various purposes, such as the production
of biofuels or pharmaceuticals.
Research in microbiomes, the communities of microorganisms that live in and on our bodies, have gained
significant attention in the past decade. Research has shown that microbiomes are critical in maintaining our
health and well-being. Advances in microbiome research have provided insights into how microbes interact
with our immune system and influence our overall health. Our body is home to trillions of microbes, most of
which live in our gut. They outnumber our cells and contribute much more to our immunity and overall health.
Along with their number, their composition related to the environment, dietary and genetic features
of the host decides how well they protect and maintain the host’s health. Their role in regulating obesity,
diabetes, allergies and autoimmune diseases is phenomenal. Probiotics, live bacterial seeds in capsules, are
an emerging branch of medical treatment to manage the population and composition of gut microbiota. Like
probiotics, prebiotics research is also progressing significantly. Prebiotics are substances that promote the
growth of beneficial bacteria. Hence, it ensures that the probiotic treatments would have lasting results. These
interventions have shown promise in treating irritable bowel syndrome and allergies. The past decade has also
seen significant progress in understanding the role of the microbiome in mental health. Studies have linked
changes in the gut microbiome to depression and anxiety, and interventions such as probiotics have shown
promise in improving mental health outcomes.
The study of microbiomes, which refers to the collective microbial communities within a particular
environment, has grown tremendously in the past decade. Advances in sequencing technology, computational
analysis, and imaging techniques have allowed us to understand the complexity and diversity of microbial
ecosystems and their role in human health and disease.
One of the most significant developments in microbiome research has been the Human Microbiome Project,
launched in 2008 by the National Institutes of Health (NIH). This ambitious project aimed to characterise the
microbial communities that inhabit the human body and to understand their role in health and disease. The
project provided an unprecedented look into the human microbiome, identifying thousands of microbial species
that coexist within us.
In recent years, there has also been significant progress in understanding the role of the microbiome in brain
function. Studies have linked changes in the gut microbiome to neurodegenerative diseases such as Alzheimer’s
and Parkinson’s, as well as to conditions such as autism and depression. This research has opened up new
40
1.2 Neuroscience
avenues for developing therapies that target the microbiome to improve brain health.
Advances in sequencing technology have also led to the discovery of previously unknown microbial species
and functions. For example, researchers discovered a new class of bacteria that can break down polyurethane,
a standard plastic, in 2016. This discovery has exciting implications for the development of new biodegradable
materials.
Finally, advances in imaging technology have allowed us to visualise microbial communities in new ways.
For example, in 2019, researchers developed a technique for imaging the gut microbiome in living animals,
providing new insights into how the microbiome functions in real-time.
Discoveries and technological advancements have transformed our understanding of microbial communities
and their role in health and disease. From the Human Microbiome Project to the development of probiotics
and prebiotics, these developments provide exciting prospects for future research and treatment of microbiome-
related conditions. With continued progress in microbiome research, we aim to unlock new insights into the
microbial world and develop new therapies to improve human health.
1.2 Neuroscience
Significant advances have been made in understanding brain function in the past decade. Brain-computer
interfaces (BMI) have made it possible to restore communication to patients with severe paralysis. Research
has also shed light on the mechanisms underlying brain disorders like Alzheimer’s disease and depression. The
BMIs, allow direct communication between the brain and a computer or other external device. They have also
been used to treat depression and anxiety by modulating neural activity. The latest news in this line is the FDA’s
approval of Elon Musk’s Neuralink to put computer chips in the human brain. BMI may bring unprecedented
dimensions to how paralysis and backbone injuries are treated. For example, BMI has shown promise in treating
paralysis, allowing patients to control prosthetic limbs with their thoughts.
Much of the research in these lines was carried out in the past decade, and neuroscience has seen significant
advancements that have transformed our understanding of the brain and its functions. From the development
of new imaging techniques to the discovery of new neural networks, these developments have provided new
insights into the workings of the brain and have exciting implications for diagnosing and treating neurological
disorders.
One of the most significant developments in neuroscience has been the development of optogenetics.
This technique allows researchers to control the activity of specific neurons using light. Optogenetics has
revolutionised the field, allowing researchers to explore the neural circuits underlying behaviour and to develop
new therapies for neurological disorders such as Parkinson’s disease and epilepsy.
Advances in imaging technology have also allowed researchers to visualise the brain in unprecedented
detail. Functional magnetic resonance imaging (fMRI) and diffusion tensor imaging (DTI) have allowed us to
study the brain in action, mapping neural activity and the connections between different brain regions.
The past decade has also seen significant progress in understanding the genetic basis of neurological disor-
ders. The development of genome sequencing technology has allowed researchers to identify genetic mutations
associated with conditions such as autism and schizophrenia, providing new targets for drug development.
Another breakthrough in neuroscience has been the development of deep brain stimulation (DBS). This
technique involves implanting electrodes in the brain to modulate neural activity. DBS has been used to treat
various neurological disorders, including Parkinson’s disease, epilepsy, and obsessive-compulsive disorder.
In recent years, there has also been growing interest in the role of neuroplasticity, the brain’s ability to
adapt and change in response to experience. This research has shown that the brain can rewire itself in response
41
1.3 Threats?
to environmental cues, offering new possibilities for treating neurological disorders.
Advances in artificial intelligence and machine learning have allowed researchers to analyse large datasets
and develop predictive models of brain function. These models could revolutionise the diagnosis and treatment
of neurological disorders, allowing for more personalised and effective therapies.
The past decade has witnessed significant progress in neuroscience, with discoveries and technological
advancements transforming our understanding of the brain and its functions. From optogenetics and brain-
machine interfaces to advances in imaging technology and the study of neuroplasticity, these developments
provide exciting prospects for future research and treatment of neurological disorders. With continued progress
in neuroscience, we aim to unlock new insights into the brain and develop new therapies to improve human
health.
1.3 Threats?
Each of the branches described earlier also has potentially negative consequences. Here are some examples:
Astronomy: The growing number of satellites and space debris in orbit around the Earth is a concern
for astronomers, as they can interfere with observations and increase the risk of collisions. Additionally, there
are concerns about the environmental impact of space exploration, including the potential for contamination of
other planets and the potential use of resources on Earth.
Genetics: The increasing availability of genetic information raises concerns about privacy and the potential
for discrimination. There are also ethical questions surrounding gene editing technologies such as CRISPR,
particularly concerning altering traits such as intelligence or physical appearance.
Microbiomes: While research on the microbiome has shown promise in improving human health, there are
concerns about the potential for unintended consequences. For example, altering the microbiome could disrupt
the delicate balance of microbial communities in the body, leading to unexpected health problems. Additionally,
there are concerns about the environmental impact of large-scale microbiome interventions, such as the use of
genetically modified bacteria.
Neuroscience: While neuroscience developments have brought about many positive changes, there are
also some potentially adverse consequences. One concern is the potential for misusing technologies such as
brain-machine interfaces and deep brain stimulation. While these techniques have shown great promise in
treating neurological disorders, there is a risk that they can be used for non-medical purposes, such as enhancing
cognitive abilities or altering mood.
Another concern is the ethical implications of research on the brain. For example, optogenetics raises
questions about the use of animals in research and the potential risks to human subjects. Similarly, using machine
learning algorithms to analyse brain data raises concerns about privacy and the potential for discrimination.
There are also broader social and cultural implications to consider. For example, the growing use of
technology to monitor and manipulate brain activity raises questions about autonomy and individual freedom.
The increasing focus on brain-based explanations for behaviour and mental health also reinforces a reductionist
view of human nature.
Finally, there is the issue of access and equity. The high cost of many of these technologies means that
they may only be available to a privileged few, creating disparities in healthcare and exacerbating existing social
inequalities.
42
1.3 Threats?
Conclusion
The past ten years have seen remarkable scientific developments. Astronomy has revealed new insights into
the universe, genetics has made significant strides in gene editing and gene therapy, microbiome research has
provided insights into the importance of our microbiomes, and neuroscience has advanced our understanding
of the brain. These developments offer exciting prospects for the future of science and technology.
In conclusion, while the developments in neuroscience and other related branches have brought about
many positive changes, there are also potential negative consequences that must be carefully considered. It’s
important to note that the potential negative consequences do not negate the importance of scientific research and
technological development. Instead, they highlight the need for responsible and ethical use of these technologies
and ongoing evaluation and consideration of their potential risks and benefits.
Addressing these concerns will require ongoing dialogue and collaboration between scientists, policymak-
ers, and the public. By doing so, we can work to ensure that these technologies are used responsibly and
ethically and that their benefits are shared equitably.
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 airis4D
and has a teaching experience of 33+ years in Physics. His area of specialisation is AI and ML.
43
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.
