Cover page
Image Name: A Wheel within a Wheel.
opo0221a, also known as Hoag’s Object is a Ring Galaxy about 550 million light years from us. The outer ring
is a ring of young stars formed in the shockwave of two colliding galaxies and the inner one is the older stars
in the galaxies themselves after merging together. Image Credit: NASA/ESA and The Hubble Heritage Team
(STScI/AURA)
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.11, 2023
www.airis4d.com
Welcome to the 11th edition of AIRIS4D Journal, where we continue to unravel the fascinating frontiers
of technology and science. In this edition, we delve into a diverse range of subjects, from quantum computing
to ornamental fish parasites, taking you on an intellectual journey that spans the cosmos to the depths of our
own planet.
Quantum Computing: A Quantum Leap for AI
We kick off with a comprehensive two-part exploration of ”Introduction to Quantum Machine Learning”
by Blesson George. In Part 1, the article illuminates the mesmerizing realm of quantum computing, with its
foundational building blocks—qubits and quantum superposition. We are taken on a journey through quantum
circuits and the boundless possibilities of exponential speedup in specific tasks. George’s insights shine a light
on the quantum future of AI.
Navigating the Landscape of NLP with Evaluation Metrics
Next, we venture into the uncharted forest of natural language processing (NLP) with ”Evaluation Metrics:
A Tale of Measuring Success” by Jinsu Ann Mathew. Drawing parallels between data analysis and wilderness
navigation, Mathew guides us through a plethora of NLP evaluation metrics. Accuracy, precision, recall, F1
score, AUC, Mean Reciprocal Rank, and more come to life, each with its unique role in assessing NLP model
performance. Mathew reminds us that choosing the right metric is the compass that leads to successful AI
models.
Demystifying AI: Unraveling the Black Box
In ”AI Models as Black Boxes: Not for Long” by Linn Abraham, we embark on a quest to demystify
AI’s black boxes. Abraham articulates the significance of model interpretability, shedding light on the need
to satisfy human curiosity, advance scientific knowledge, and foster trust, fairness, and privacy in AI. The
article introduces ”Interpretable Machine Learning” and various methods to make AI models more transparent.
Abraham’s work underscores the importance of building bridges between the technical complexity of AI and
its real-world applications.
Journey Through Gravitational Forces
Professor Ajit Kembhavi guides us through celestial orbits in ”Black Hole Stories-4: Particle Paths in
Newton’s Theory of Gravitation.” This illuminating article takes us on a celestial journey, exploring the motion
of particles in gravitational fields, revealing the inner workings of planetary orbits and effective potentials.
We are reminded of the fundamental principles underpinning our understanding of celestial bodies, offering a
glimpse into the future of gravitational studies.
Discovering Cosmic Powerhouses: Supergiant High Mass X-ray Binaries
Sindhu G. takes us on a cosmic adventure in ”Shedding Light On The Mysteries Of Supergiant High
Mass X-ray Binaries: Cosmic Powerhouses Unveiled.” These powerful binary star systems, housing massive
supergiant stars and compact objects, illuminate our understanding of massive stars and extreme gravitational
conditions. From Cygnus X-1 to Cen X-3, we journey through these cosmic powerhouses and their significance
in astrophysical research.
Parasites in Aquatic Elegance: A Peek into Ornamental Fish Health
In ”The Common Ectoparasite in Ornamental Fishes,” Geetha Paul uncovers the hidden world of ectopar-
asites, specifically Argulus foliaceus and Argulus japonicus, known as ’fish lice.’ These tiny parasites disrupt
the elegance of ornamental freshwater fish, causing a range of health issues. Paul discusses control measures,
underlining the importance of managing these parasites to ensure the well-being of these aquatic companions.
LiDAR: Illuminating Our World with Precision
Our journey concludes with ”LiDAR - A Versatile Imaging Technology” by Ninan Sajeeth Philip. This
enlightening article explores the world of LiDAR, a remote sensing technology that creates precise 3D maps
of our environment. From autonomous vehicles to environmental monitoring, LiDAR plays a crucial role in
numerous industries. Despite its advantages, Philip addresses the challenges and presents low-cost LiDAR
systems, emphasizing the enduring significance of this technology.
As you embark on this intellectual voyage through the pages of AIRIS4D Journal, we encourage you to
embrace the wonder of discovery, the pursuit of knowledge, and the boundless potential of technology. We hope
that the diverse insights shared in this edition ignite your curiosity and spark new ideas.
Thank you for joining us on this journey, and we look forward to continuing to explore the ever-evolving
landscape of AI and science with you.
iii
Contents
Editorial ii
I Artificial Intelligence and Machine Learning 1
1 Introduction to Quantum Machine Learning - Part 1 2
1.1 Basics of Quantum Computing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Evaluation Metrics: A Tale of Measuring Success 6
2.1 Accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Precision, recall, and F1 score . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Area under the Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4 Mean Reciprocal Rank: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.5 Root Mean Squared Error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.6 BLEU Score . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.7 ROUGE Score . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.8 Perplexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3 AI Models as Black Boxes: Not for Long 12
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.2 Towards Interpretable Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.3 A survey of existing interpretability methods . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
II Astronomy and Astrophysics 18
1 Black Hole Stories-4
Particle Paths in Newtons’ s Theory of Gravitation 19
1.1 Planetary Orbits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.2 Orbits in Classical Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.3 The Effective Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.4 Types of Orbits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.5 Orbital Shapes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.6 Approximations Made in the Two-Body Problem . . . . . . . . . . . . . . . . . . . . . . . . 24
2 Shedding Light On The Mysteries Of Supergiant High Mass X-ray Binaries : Cosmic
Powerhouses Unveiled 26
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.2 Supergiant High Mass X-ray Binaries (sgHMXBs) . . . . . . . . . . . . . . . . . . . . . . . 26
2.3 Classical Supergiant High-Mass X-ray Binaries . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.4 Obscured Supergiant High-Mass X-ray Binaries . . . . . . . . . . . . . . . . . . . . . . . . . 29
CONTENTS
2.5 Some Examples Of sgHMXBs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.6 Importance Of The Study Of sgHMXBs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
III Biosciences 31
1 The common Ectoparasite in Ornamental Fishes 32
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.2 Symptoms of Infection: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.3 Mode of Propagation: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.4 Something Fishy in your Aquarium? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
1.5 Control measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
IV General 37
1 LiDAR - A versatile Imaging Technology 38
1.1 Do it Yourself LiDAR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
1.2 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
v
Part I
Artificial Intelligence and Machine Learning
Introduction to Quantum Machine Learning
- Part 1
by Blesson George
airis4D, Vol.1, No.11, 2023
www.airis4d.com
1.1 Basics of Quantum Computing
In the ever-expanding realm of computing, where data flows like a digital river, a fundamental unit known
as the ”bit” emerges as the cornerstone of all data processing. This seemingly modest binary entity, capable of
representing either a 0 or a 1, underpins the entire digital universe. However, in the revolutionary landscape
of quantum computing, a remarkable entity takes center stage: the ”qubit.” Diverging significantly from the
classical bit, the qubit possesses a mesmerizing attribute – quantum superposition. This phenomenon enables
a qubit to exist in multiple states simultaneously, transcending the boundaries of classical computing. In
this transformative era of quantum computation, the qubit’s unique property has the potential to reshape our
understanding of computation as we know it.
1.1.1 Qubits
In the expansive landscape of computing, where data flows like a digital river, an elemental cornerstone
emerges to underpin all data processing: the ”bit.” This seemingly unassuming binary unit, capable of repre-
senting either a 0 or a 1, serves as the very bedrock upon which the entire digital universe is constructed. The
bit’s elegant simplicity conceals its profound significance, as it stands as the atomic unit of information fueling
not just traditional programming but also the most intricate machine learning algorithms.
However, in the realm of quantum computing, a truly revolutionary entity steps into the spotlight: the
”qubit.” Diverging radically from its classical counterpart, the bit, which is confined to the binary realm, the
qubit possesses a mesmerizing quality—quantum superposition. This unique attribute allows a qubit to inhabit
multiple states simultaneously, transcending the limits of classical computing. The qubit, with its quantum
wonder, is poised to reshape the boundaries of computation as we understand them today.
The remarkable characteristic of qubits stems from the quantum mechanical phenomenon known as
superposition. While a qubit, like a classical bit, can represent two fundamental states, its true power lies in the
ability to exist in a combination of 0 and 1 states simultaneously. This implies that a portion of the qubit can be
in the 0 state while another portion is in the 1 state.
Mathematically a qubit’s state can be represented as
Ψ = α|0 > +β|1 >
1.1 Basics of Quantum Computing
Figure 1: A classical computer operates with bits, each of which can exist in one of two states: 0 or 1, and it
can be only in one state at a time. Unlike bits, a qubit, which leverages the principles of quantum mechanics,
possesses a unique property called superposition. This enables a qubit to exist in a state that is a combination
of 0 and 1 simultaneously, expanding its capacity to represent and process information in a profoundly different
way. While a classical bit can only be at one of the two extremes, 0 or 1, a qubit can inhabit any point within
the sphere. Image Courtesy: https://medium.com/@mark.rethana/a-beginners-guide-to-the-quantum-computing-and-superposition-
536e4fc040a2
where α and β are complex numbers that determine the probability amplitudes of being in states 0 and 1
A qubit can be effectively represented by specific physical properties of particles, such as the spin of an
electron or the polarization of a photon.
In the case of electrons, a qubit is associated with the intrinsic angular momentum, or ”spin,” of the electron.
This spin can be in one of two possible states: ”spin-up” or ”spin-down,” serving as the quantum counterparts
to classical 0 and 1. Similarly, photons, particles of light, can serve as qubits by encoding information in their
polarization states, which can be either horizontal or vertical. Just as with electron spin states, these polarization
states allow for the representation of quantum information as either 0 or 1t.
1.1.1.1 Bloch Sphere
The Bloch sphere serves as a geometric representation of the pure state space in a two-level quantum
mechanical system, commonly referred to as a qubit. Within this framework, the ’poles’ of the sphere symbolize
the two fundamental states, typically denoted as zero and one. This innovative concept was originally introduced
by Felix Bloch, a distinguished figure in the field of solid-state physics renowned for his pivotal discovery in
nuclear magnetic resonance (NMR).
In terms of computational basis, |0 > and |1 >, the general pure state of a qubit can be represented as
|ψ >= α|0 > +β|1 >. Since |α
2
| + |β
2
| = 1, the state can be written as
|ψ >= cos
θ
2
|0 > +e
iϕ
sin
θ
2
|1 > with 0 < θ < π and 0 < ϕ < 2π. With the constraints on θ and ϕ,
these two numbers define a point on the surface of the unit sphere in three dimensions. This sphere is called the
Bloch sphere.
3
1.2 Summary
Figure 2: A representation of the Bloch Sphere and a state vector(Ψ). Image Courtesy: Wikipedia
1.1.2 Quantum Circuits
Quantum circuits are a fundamental concept in quantum computing, similar to classical digital circuits
in conventional computing. These circuits are constructed using quantum gates to manipulate and process
quantum bits, or qubits, which are the quantum counterparts of classical bits. Quantum gates, like the NOT
gate (X-gate), Hadamard gate (H-gate), and Z-gate, perform various operations on qubits, such as flipping their
states, creating superpositions, or entangling them.
NOT gate is the most elementary single qubit operation which exchanges the state. ie, it takes |0 > to |1 >
and vice versa. If we denote NOT gate by a matrix, it may be represented as X =
0 1
1 0
!
.
The Hadamard gate transforms elements of the computational basis to “halfway” between the elements. It
takes |0 > to (|0 > +|1 >)/
√
2 and |1 > to (|0 > −|1 >)/
√
2. The matrix denoting Hadamard gate (H Gate)
is given as H =
1
√
2
1 1
1 −1
!
.
The Z-gate is defined as Z =
1 0
0 −1
!
. It leaves |0 > invariant, and changes the sign of |1 >. This is
essentially a phase shift. While there are infinitely many single-qubit operations, they can be approximated to
arbitrary precision with only a finite set of these gates.
Quantum circuits are used to implement quantum algorithms, solve specific problems, and simulate
quantum systems. They play a pivotal role in quantum computation and form the building blocks for more
complex quantum algorithms, offering the potential for exponential speedup over classical computing for specific
tasks like factoring large numbers, optimization, and simulating quantum systems. Quantum circuits are a crucial
component of the rapidly evolving field of quantum computing.
4
1.2 Summary
1.2 Summary
The article introduces the fundamental concepts of bits and qubits in the context of computing. It highlights
the significance of the bit as the basic unit of information in traditional computing, capable of representing
binary values. In contrast, the qubit, a key element in quantum computing, possesses the remarkable property
of quantum superposition, allowing it to exist in multiple states simultaneously, greatly expanding its capacity
to process information.
The discussion delves into the mathematical representation of a qubit’s state, emphasizing the complex
numbers that define its probability amplitudes. The physical properties that can serve as qubits, such as
electron spin and photon polarization, are also briefly explained. Furthermore, the concept of the Bloch sphere
is introduced as a geometric representation of the qubit’s pure state space. The article also mentions the
importance of quantum circuits and single qubit operations, which are fundamental to quantum computing,
similar to classical digital circuits in conventional computing.
References
[1] Wittek, P. (2014). Quantum machine learning: what quantum computing means to data mining. Academic
Press.
[2] Bernhardt, C. (2019). Quantum computing for everyone. Mit Press.
About the Author
Dr. Blesson George presently serves as an Assistant Professor of Physics at CMS College
Kottayam, Kerala. His research pursuits encompass the development of machine learning algorithms, along
with the utilization of machine learning techniques across diverse domains.
5
Evaluation Metrics: A Tale of Measuring
Success
by Jinsu Ann Mathew
airis4D, Vol.1, No.11, 2023
www.airis4d.com
Picture this: you’re an explorer in an uncharted territory, armed with a map and a compass, determined to
navigate through a dense and mysterious forest. Your mission is to reach the hidden treasure buried deep within,
but the journey is fraught with twists and turns. To stay on course, you need your trusty tools to guide you.
In the world of data science and machine learning, we find ourselves in a similar scenario. Our terrain
is the vast landscape of data, our destination is valuable insights, and our tools are evaluation metrics. These
metrics are the compass and map that help us make sense of our data, guide our models, and lead us to our
desired outcomes.
Just as the explorer relies on their map to avoid pitfalls and reach the treasure, data scientists rely on
evaluation metrics to navigate the data maze and make informed decisions. In this article, we will cover some
of the major evaluation metrics used in NLP. These metrics include accuracy, precision, recall, F1 score, Area
Under Curve, Mean Reciprocal Rank, Root mean Square,BLEU, ROUGE and perplexity.
2.1 Accuracy
In natural language processing (NLP), accuracy is a commonly used evaluation metric that measures the
overall performance of a classification or prediction model. It is a fundamental metric used to assess the model’s
ability to correctly classify or predict the target variable, such as sentiment analysis, text classification, named
entity recognition, or any other NLP task where the goal is to assign a category or label to a given text or
sequence.
Accuracy is defined as the ratio of correctly predicted instances to the total number of instances in the
dataset .For example, if a model is trained to classify emails as spam or not spam, accuracy would be the ratio
of correctly labeled emails to the total number of emails. The opposite of accuracy is the error rate, which
measures the percentage of incorrectly classified instances
While accuracy is a simple and widely used metric in NLP, it can sometimes be misleading, especially in
cases with imbalanced datasets. Imbalanced datasets have an unequal distribution of classes, where one class
may dominate the dataset. In such cases, a model that always predicts the majority class can achieve a high
accuracy score, but it may not be truly effective for the task. Therefore, while accuracy is valuable for initial
assessments, a comprehensive evaluation of an NLP system often requires a deeper look into other metrics and
contextual considerations.
2.2 Precision, recall, and F1 score
(image courtesy:https://proclusacademy.com/blog/explainer/precision-recall-f1-score-classification-models/)
Figure 1: Confusion matrix
Overall, accuracy is a fundamental evaluation metric in NLP that measures the percentage of correctly
classified instances in a classification task. While it is a simple and widely used metric, it may not always
provide a complete picture of model performance and should be used in conjunction with other metrics.
2.2 Precision, recall, and F1 score
Precision, recall, and F1 score are important evaluation metrics in natural language processing (NLP) that
are used to measure the performance of NLP models. These metrics are based on the concepts of true positives
(TP), false positives (FP), true negatives (TN), and false negatives (FN). Here is a brief overview of each metric:
Precision: Precision is a crucial evaluation metric in classification tasks, quantifies the accuracy of a
model’s positive predictions. It calculates the ratio of true positive predictions (instances correctly identified
as positive) to the total number of positive predictions made by the model. In essence, precision reflects the
percentage of positive predictions that the model got right. High precision means that the model makes fewer
false positive errors, making it particularly useful when minimizing incorrect positive predictions is essential.
It is calculated as TP / (TP + FP).
Recall: Recall, also known as sensitivity or the true positive rate, assesses a model’s capability to identify
all relevant instances within a dataset. It measures the ratio of true positive predictions to the total number of
actual positive instances in the dataset. High recall is valuable when the goal is to minimize false negatives,
ensuring that positive instances are not missed. It is calculated as TP / (TP + FN). F1score: This metric is
the harmonic mean of precision and recall and is used to balance the trade-off between precision and recall.
The F1 score is especially helpful in situations where optimizing for one metric could negatively impact the
other. It becomes particularly relevant when dealing with imbalanced datasets or scenarios where both false
positives and false negatives need to be minimized. By considering both precision and recall, the F1 score
helps in finding an optimal trade-off between these two metrics, ensuring a comprehensive evaluation of the
model’s performance. It is calculated as 2 * (precision * recall) / (precision + recall). The figure (Figure 1)
shows a confusion matrix, which is a table that summarizes the performance of a classification model. Using
this confusion matrix, we can calculate precision, recall, and F1 score for a classification model. For example,
in the figure, precision is calculated as 1 / (1+ 0) = 1, recall is calculated as 1 / (1 + 29) = 0.033, and F1 score is
calculated as 2 * (1 * 0.033) / (1+0.033) = 0.065.
While accuracy is a fundamental evaluation metric in NLP, precision, recall, and F1 score offer a more
nuanced understanding of model performance, especially in cases with imbalanced datasets where one class
7
2.3 Area under the Curve
(image courtesy:https://www.evidentlyai.com/classification-metrics/explain-roc-curve)
Figure 2: ROC curve
significantly outweighs the other. Therefore, a comprehensive evaluation of an NLP system often requires a
deeper look into other metrics and contextual considerations.
2.3 Area under the Curve
The area under the curve (AUC) is a popular evaluation metric in natural language processing (NLP) that
measures the overall performance of a binary classification model.The AUC is calculated as the area under the
receiver operating characteristic (ROC) curve, which is a plot of the true positive rate (TPR) against the false
positive rate (FPR) at different classification thresholds(Figure 3).
The ROC curve is a useful tool for evaluating the performance of binary classification models because it
shows how well the model can distinguish between positive and negative instances at different classification
thresholds. The TPR is the proportion of true positives among all the instances that are actually positive, and
the FPR is the proportion of false positives among all the instances that are actually negative. The ROC curve
is a plot of TPR against FPR at different classification thresholds, and the AUC is the area under this curve.
The AUC ranges from 0 to 1, with a higher value indicating better model performance. An AUC of 0.5
indicates that the model is no better than random guessing, while an AUC of 1.0 indicates perfect classification
performance. The AUC is a useful metric for comparing different binary classification models and for selecting
the best model for a given task.
2.4 Mean Reciprocal Rank:
Mean reciprocal rank (MRR) is a metric used to evaluate the performance of a system that produces a list
of possible responses to a sample of queries, ordered by probability of correctness. The MRR is calculated as
the average of the reciprocal ranks of results for a sample of queries Q, where the reciprocal rank of a query
response is the multiplicative inverse of the rank of the first correct answer: 1 for first place, 1/2 for second place,
1/3 for third place, and so on. The reciprocal value of the mean reciprocal rank corresponds to the harmonic
mean of the ranks.
To better understand MRR, let’s consider an example. Suppose we have a search engine that is designed to
answer questions about famous people. We test the search engine using a set of 10 queries, and for each query,
the search engine returns a ranked list of 5 possible answers. For each query, we also have a set of ground truth
8
2.5 Root Mean Squared Error
answers that are considered correct. Here is an example of the search results for one of the queries: Query: Who
is the founder of Microsoft? Search results: Bill Gates Steve Jobs Mark Zuckerberg Jeff Bezos Elon Musk In
this case, the correct answer is Bill Gates, which is ranked first in the search results. The reciprocal rank for this
query is 1/1 = 1. We repeat this process for all 10 queries and calculate the reciprocal rank for each query. Then,
we take the average of the reciprocal ranks to calculate the MRR for the search engine. Suppose the reciprocal
ranks for the 10 queries are as follows: 1,
1
2
,
1
3
, 0, 1, 0,
1
2
, 0.1.1 The average of the reciprocal ranks is (1/1 + 1/2
+ 1/3 + 0 + 1/1 + 0 + 1/2 + 0 + 1/1 + 1/1) / 10 = 0.65. This means that, on average, the first relevant document
was found at rank 1/0.65 = 1.54. This indicates that the search engine is performing reasonably well, but there is
still room for improvement. Overall, MRR is a useful metric for evaluating the performance of a search engine
that returns a ranked list of answers to queries. It provides a measure of how quickly and accurately the search
engine can provide the most relevant answer to a query.
2.5 Root Mean Squared Error
Root mean square error (RMSE) is a commonly used evaluation metric in natural language processing
(NLP) that measures the quality of predictions. RMSE is used in supervised learning applications, as it uses
and needs true measurements at each predicted data point. RMSE can be expressed as the square root of
the mean of the squared differences between the predicted and true values. To compute RMSE, calculate the
residual (difference between prediction and truth) for each data point, compute the norm of residual for each
data point, compute the mean of residuals and take the square root of that mean. The squaring of differences
ensures that both overestimations and underestimations are treated with equal importance, and it penalizes larger
errors more. In NLP applications, RMSE may be used in scenarios where the goal is to predict a continuous
or numerical outcome, such as sentiment scores, sentiment intensity, or numerical ratings. For example, in
sentiment analysis, if you’re predicting a numerical sentiment score (e.g., 1 to 5) for a piece of text, you can use
RMSE to assess how well your model’s predictions align with the actual scores. Lower RMSE values indicate
better predictive accuracy and a closer match between the model’s predictions and the true sentiment values.
2.6 BLEU Score
BLEU, or Bilingual Evaluation Understudy, is a widely used metric in natural language processing,
primarily for evaluating the quality of machine-generated text, especially translations. It was originally devised
to address the need for automated evaluation of machine translation systems. However, over time, it has found
applications in other NLP tasks, such as text summarization and text generation. BLEU operates on the principle
of comparing the machine-generated text with one or more reference translations provided by humans. This
comparison is performed by assessing the overlap of n-grams, which are sequences of words, between the
generated output and the reference text. ie, the BLEU score is calculated by comparing the n-grams of the
candidate text with the n-grams of the reference text. The score ranges from 0 to 1, with a higher score indicating
better quality of the machine-generated text. Here is an example of how BLEU score can be used in NLP:
Suppose we have a machine translation system that translates English sentences to French. We have a dataset
of 1000 English sentences and their corresponding French translations. We want to evaluate the performance
of the machine translation system using BLEU score. We first split the dataset into a training set and a test set.
We train the machine translation system on the training set and use it to translate the English sentences in the
test set to French. We then calculate the BLEU score of the translations by comparing them to the true French
translations in the test set. Suppose the BLEU score of the machine translation system is 0.8. This means that, on
9
2.7 ROUGE Score
average, the translations of the machine translation system are 80% similar to the true French translations in the
test set. A higher BLEU score indicates better quality of the machine-generated text. While BLEU is a valuable
metric for assessing machine-generated text, it has limitations, such as not capturing fluency, coherence, or the
overall context of the generated text. Consequently, researchers and practitioners often use additional metrics
like ROUGE to provide a more comprehensive evaluation of machine-generated text in specific NLP tasks.
2.7 ROUGE Score
ROUGE, which stands for Recall-Oriented Understudy for Gisting Evaluation, is a set of metrics frequently
employed in the field of natural language processing (NLP) and text summarization to assess the quality of
machine-generated text, particularly summaries or generated content. ROUGE is particularly useful in evaluating
how well the generated text captures the important content from one or more reference texts, making it a key
player in evaluating the performance of text summarization systems.
One of the fundamental purposes of ROUGE is to measure recall, which focuses on the system’s ability to
capture relevant content from the reference texts. Unlike other metrics like BLEU that predominantly emphasize
precision (correctness of specific words or n-grams), ROUGE balances both precision and recall aspects. It
does this through a set of recall-based metrics, each designed to evaluate different facets of content overlap
between the machine-generated text and the reference text. Some commonly used ROUGE metrics include
ROUGE-N, which measures n-gram overlap, ROUGE-L, which considers the longest common subsequence,
and ROUGE-W, a weighted version of the LCS-based measure.
The following five evaluation metrics are available:
ROUGE-N: Overlap of n-grams between the system and reference summaries. ROUGE-1 refers to the
overlap of unigrams (each word) between the system and reference summaries. ROUGE-2 refers to the overlap
of bigrams between the system and reference summaries.
ROUGE-L: Longest Common Subsequence (LCS) based statistics. Longest common subsequence prob-
lem takes into account sentence-level structure similarity naturally and identifies longest co-occurring in se-
quence n-grams automatically.
ROUGE-W: Weighted LCS-based statistics that favors consecutive LCSes.
ROUGE-S:Skip-bigram based co-occurrence statistics. Skip-bigram is any pair of words in their sentence
order.
ROUGE-SU: Skip-bigram plus unigram-based co-occurrence statistics.
ROUGE plays a critical role in evaluating the quality of machine-generated text in NLP applications,
particularly when content overlap with reference texts is a vital factor. It provides a more comprehensive
assessment of content quality compared to metrics that primarily focus on lexical overlap. Researchers and
practitioners often utilize various ROUGE metrics tailored to their specific evaluation needs, offering valuable
insights into the quality and effectiveness of generated text, summaries, or other NLP applications.
2.8 Perplexity
Perplexity is a statistical measure used to evaluate the performance of language models in NLP. It measures
how confidently a language model predicts a text sample and quantifies how ”surprised” the model is when it
sees new data. To better understand perplexity, let’s consider an example. Suppose we have a language model
that is trained on a dataset of 1000 sentences. We want to evaluate the performance of the language model
using perplexity. We first split the dataset into a training set and a test set. We train the language model on the
10
2.9 Conclusion
training set and use it to predict the next word in each sentence in the test set. We then calculate the perplexity
of the language model by comparing the predicted next word with the true next word in each sentence. Suppose
the perplexity of the language model is 50. This means that, on average, the language model is 50 times more
”surprised” by the test set than by the training set. A lower perplexity indicates better model performance, as it
means that the language model is better at predicting the next word in the test set. Perplexity is a useful metric
for evaluating the quality of language models in NLP. It is commonly used to compare different language models
and to tune hyperparameters. Perplexity is fast to calculate because it’s based on the average log-likelihood of
the dataset, which can be approximated using a single pass through the data. However, it’s important to keep in
mind that low perplexity is not always accurate and can be affected by the size and complexity of the dataset.
2.9 Conclusion
In conclusion, evaluation metrics in natural language processing (NLP) play a crucial role in assessing
the effectiveness and quality of NLP models and systems. These metrics provide a standardized and objective
means to measure how well a model performs specific tasks, such as text classification, machine translation,
sentiment analysis, and text summarization. The choice of the appropriate metric depends on the nature of
the task, dataset characteristics, and the specific goals of the application. Metrics like accuracy, precision,
recall, F1 score, AUC, MRR,RMSE, BLEU, ROUGE, and perplexity offer valuable insights into a model’s
performance, allowing researchers and practitioners to fine-tune models and algorithms for better results. As
NLP continues to advance, the development and refinement of these evaluation metrics remain essential for
ensuring the continuous improvement of NLP technologies and their real-world applications, from chatbots and
virtual assistants to language translation and content recommendation systems.
References
Common metrics for evaluating natural language processing (NLP) models
Evaluation Metrics With Python Codes
Accuracy vs. precision vs. recall in machine learning: what’s the difference?
Precision, Recall, and F1 Score: When Accuracy Betrays You
How to explain the ROC curve and ROC AUC score?
Understanding BLEU and ROUGE score for NLP evaluation
Perplexity Intuition (and its derivation)
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.
11
AI Models as Black Boxes: Not for Long
by Linn Abraham
airis4D, Vol.1, No.11, 2023
www.airis4d.com
3.1 Introduction
It is quite common to hear the phrase that “AI/ML models are black boxes”. How true is this and is there
any scope for the state-of-affairs to be improved? Before trying to understand this, a natural question might
arise - Is there a problem if AI models remain black boxes as long as they work? Let’s try to answer this.
(Image Credit: Interpreatable Machine Learning by Christoph Molnar)
Satisfying natural human curiosity. Why do ML models work where traditional methods fail? Why do
classical ML algorithms like random forest algorithms or SVMs show superior performance in certain areas
where deep neural network fail to meet benchmarks? All such questions stem from our natural curiosity to
understand things.
Adding to scientific knowledge. If something works and one is not able to explain why it has worked.
Then you might be adding nothing to the existing scientific knowledge. However if the model works better than
traditional approaches then this shows that the model has learnt something that other approaches have missed.
3.2 Towards Interpretable Machine Learning
Extracting this information out can be very useful to increase our understanding of the scientific problem.
Improving existing models. Being able to understand the inner workings of the ML model allows to
understand it’s weakness and thus build a better model.
Trust, Fairness and Privacy With the ever increasing integration of AI into our everday life, question of
fairness naturally arise. Even in our current scenarios of people making decision, its frustrating to be denied a
loan when you are in dire need of it. But imagine when the machines are doing the job and you desire to know
why your loan request was denied. Since ML models are prone to the bias that exists in the data, can we trust
such a model? How can fairness be programmed into the model? Similarly, privacy becomes a bigger concern
if data that you provide is going to be used in a way that negatively affects you.
3.2 Towards Interpretable Machine Learning
I hope that you are convinced that ML models should be more than just black box algorithms that give
better results on some metrics like classification accuracy or so. It is often the case that a single metric like
classification accuracy is an incomplete description of most real-world tasks. [Doshi-Velez and Kim(2017)].
The opposite of a black-box model might be a model that is said to be interpretable. Interpretable Machine
Learning refers to methods and models that make the behavior and predictions of machine learning systems
understandable to humans. It is currently an area of active research in Machine Learning. But what does it
actually mean to say that a model is interepretable? There is no mathematical definition of interpretability.
However when comparing two models, it could be said that a model is more interpretable than the other if it’s
decisions are easier for a human to understand. What are the ways in which interpretability can be incorporated
into machine learning?
Algorithm Transparency. How does the algorithm learn a model from the data and what kind of
relationships are learned from it? People with experience in Convolutional Neural Networks (CNN) know that
the lowest layers of a CNN learn edge detectors. The least squares method, a kind of linear model, also is a
method where the working of the algorithm is known.
Global Model Interpretability. This involves knowing the entire model at once. To explain the model
output on a global level requires you to explain the trained model, the algorithm and the data involved.
Global Model Interpretability on a Modular Level A Naive Bayes model with many hundreds of features
would be too big for me and you to keep in our working memory. To predict what the output would be given a
data point would be near impossible without actually computation. But what you can do is try to comprehend
the impact of a single weight.
Local Interpretability for a Single Prediction You can zoom in on a single instance and examine what
the model predicts for this input, and explain why.
3.3 A survey of existing interpretability methods
Interpretable Models. Linear regression, logistic regression, decision trees, Support Vector Machines
are some of the easily interpretable machine learning models. There are several methods that are focussed on
such easily interpretable and linear methods.
Model-Agnostic Methods. Such methods have several advantages over methods are specifically designed
based on the model. Better comparisons can be done as the same methods is used to evaluated different models.
Also model specific methods are much more rigid in comparison to model agnostic ones. The alternative to
using model-agnostic methods is to only use interepretable models which may not be ideal in all scenarios.
13
3.3 A survey of existing interpretability methods
Let us take a high level look at model-agnostic interpretability. We capture information about the world by
collecting data. It is then abstracted by learning to predict the data for a specific task using a black-box ML
model. Interpretability is then another layer on top of this that helps human understand the black-box model.
There are several model-agnostic methods that are worth mentioning. LIME or Local Interpretable Model
Agnostic Explanations [Doshi-Velez and Kim(2017)] is one such method. Another one is analyzing Shapley
values. We will be going into detail into both these methods in the upcoming issue.
Example-Based methods. These can be considered model-agnostic as they make any machine learning
model more interpretable. These are different from the other methods in they select data instances and do not
make use of the feature summaries or feature importance. Such methods can be motivated with examples from
daily life. Doctors in their clinical practice often uses patients who have had similar symptomps in order to
make diagnosis. Another example might be a gamer who makes decision based on gameplays with similar
situations he had encountered in the past. The template for such explanations is based on the following. Thing
B is similar to thing A and A caused Y, so I predict that B will cause Y as well. Some machine learning models
are implicitly example based such as decision trees. For a new instance, a knn model locates the k-nearest
neighbors and returns the average of the outcomes of those neighbors as a prediction.
There are several example-based methods that can be used.
1. Counterfactual explanations tell us how an instance has to change to significantly change its prediction.
2. Adversarial examples are counterfactuals used to fool machine learning models. The emphasis is on
flipping the prediction and not explaining it.
3. Prototypes are a selection of representative instances from the data and criticisms are instances that are
not well represented by those prototypes.
4. Influential instances are the training data points that were the most influential for the parameters of a
prediction model or the predictions themselves.
Neural Network Interpretation methods. There are several interpretation methods that are specific to
neural networks. The different categories of techniques that fall under this are:
1. Feature Visualization: Visualizing what features the network has learned.
2. Concepts: Which abstract concepts has the neural network learned?
3. Feature Attribution: These try to explain how each input feature contributed to a particular prediction.
4. Model distillation: This attempts to explain a neural network using a simpler model.
Pixel Attribution can be seen as a special case of feature attribution but for images. It is known under
various names such as sensitivity map, saliency map, pixel attribution map, gradient-based attribution methods,
feature relevance, feature attribution, and feature contribution. Pixel attribution methods can be classified based
on several criteria.
1. There are occlusion or perturbation based methods. Methods such as SHAP and LIME fall under these.
These change parts of the image in order to generate explanations.
2. Gradient-based methods compute the gradient of the prediction with respect to the input features. There
are several gradient based methods that differ in the way the gradient is computed.
What is common to both the methods is that the explanation and the input image are of the same size or shape.
And each pixel is assigned a value that highlights its importance towards the prediction.
Another distinction can be made within pixel attribution methods based on the baseline question.
1. Gradient-only methods tell us whether a change in pixel would cause the model prediction to change.
Examples are Vanilla Gradient and Grad-CAM.
2. Path-attribution methods compare the input image to a reference image or a baseline image. This is
usually taken to be a black image (a ”zero” image). This includes methods such as integrated gradients,
14
3.3 A survey of existing interpretability methods
Figure 1: A graphic showing the big picture of explainable machine learning. (Image Credit: Interpreatable
Machine Learning by Christoph Molnar)
15
3.3 A survey of existing interpretability methods
Figure 2: A classic example of an adversarial attack that that causes a CNN to misclassify a panda as a gibbon.
This is a mistake that a human would never make. (Image Credit: Interpreatable Machine Learning by Christoph
Molnar)
as well as the methods such as LIME and SHAP. Some path attribution methods are ”complete”, meaning
that the sum of the relevance values for all input features is the difference between the prediction of
the image minus the prediction of a reference image. The difference between classification scores of
the actual image and the baseline image are attributed to the pixels. The choice of the reference image
(distribution) has a big effect on the explanation.
Figure 3: Pixel-wise attributions of the Inception V4 network using integrated gradients. Notice that the
grey background of the image has higher attributions than more relevant pixels.(Image Credit: Interpreatable
Machine Learning by Christoph Molnar)
16
REFERENCES
References
[Doshi-Velez and Kim(2017)] Finale Doshi-Velez and Been Kim. Towards A Rigorous Science of Interpretable
Machine Learning, March 2017.
[Molnar(2019)] Christoph Molnar. Interpretable Machine Learning: A Guide for Making Black Box Models
Interpretable. Lulu, Morisville, North Carolina, 2019. ISBN 978-0-244-76852-2.
[Lipton(2017)] Zachary C. Lipton. The Mythos of Model Interpretability, March 2017.
[Miller(2018)] Tim Miller. Explanation in Artificial Intelligence: Insights from the Social Sciences, August
2018.
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.
17
Part II
Astronomy and Astrophysics
Black Hole Stories-4
Particle Paths in Newtons’ s Theory of
Gravitation
by Ajit Kembhavi
airis4D, Vol.1, No.11, 2023
www.airis4d.com
In our earlier black hole stories, we have considered how black holes in binary systems and active galactic
nuclei are detected, how their masses are measured and how lone black holes can be detected using gravitational
lensing. In all these cases, the observations depended on the effects produced by black holes, due to their gravity,
on the stars, gas or light rays. But the effected entities were at a relatively large distance from the black hole, so
the gravitational field was weak; therefore the effects were similar to those produced by the gravitational field
of Newtons’ theory, with only small corrections due to Einstein’s general relativity needed. Over the next few
black hole stories, we will consider the effects produced when particles or light rays (photons) move close to
the black hole, so that the full effects of general relativity are felt, and very interesting new phenomena emerge.
In the present story we will consider how particles move in Newtonian gravitational fields and briefly introduce
general relativity towards the end.
1.1 Planetary Orbits
Consider the motion of a planet, say Earth, around the Sun. To a good approximation, the path, or orbit,
of the Earth can be taken to be a circle, with the Sun at its centre. It was proposed by Johannes Kepler, in the
early years of the 17
th
century, in his First Law of Planetary Motion, that every planet moves in ellipse with the
Sun at a focus of the ellipse. The orbits of several planets are shown in Figure 1, with exaggerated elongations
for clarity.
In case of the Earth the orbit is almost a circle, while some of the other planets, and particularly comets,
can have more eccentric orbits. Kepler did not know the physical reasons behind his laws, and we had to wait for
the understanding until Newton proposed the universal law of gravitation in his book published in 1687 (though
the law was known earlier):
F =
GMm
r
2
Here F is the gravitational force between two particles with mass M and m, r is the distance between them
and G is Newton’s constant of gravitation. The minus sign indicates that gravitation is always an attractive
1.2 Orbits in Classical Mechanics
Figure 1: The elliptical orbits of some planets. Image Credit: CFA/NASA
force. If the Sun and the Earth gravitationally attract each other, then why does not the Earth fall into the Sun?
It would indeed do so if the Earth were at rest, there would be nothing to stop its fall. But in reality the Earth is
moving in space with a high velocity, and the net result is the elliptical orbit, as we will see below.
The force law can be stated in terms of a potential. We say that every particle of mass M gives rise to a
gravitational potential
V (r) = −
GM
r
The force on mass m due to M is then given by m multiplied by the rate at which the potential changes with r,
i.e. the derivative of the potential:
F = −m
dV (r)
dr
= −
GMm
r
2
1.2 Orbits in Classical Mechanics
The motion of the Earth and the Sun can be treated in classical mechanics, which is based on Newton’s
mechanics and its later developments, to a good approximation as a two-body problem, by ignoring the effects
of all the other planets, which are small. While the Sun and Earth are large bodies with radius 700,000 km and
6370 km respectively, these radii are much small than the distance between them, which is about 150 million
km. Therefore one can consider the two to be point particles to a very good approximation. The mass of the
Sun, which is about 2x10
30
kg, is much greater than the mass of the Earth, which is about 6x10
24
kg. It is
therefore possible to a good approximation to assume that the Sun is stationary, and that the Earth is in motion
20
1.3 The Effective Potential
around it, which simplifies the problem considerably. It is of course possible to work out the consequences of
not using the above approximations, some of which we will mention later in this story.
While it is possible to solve the equations of classical mechanics for the Sun-Earth system, it is much more
insightful to approach the problem by taking into account conserved quantities, which are physical quantities
which remain constant during the motion. Two conserved quantities which are important for the two body
problem we are considering are energy and angular momentum. The energy of the system is the sum of its
kinetic energy, which is the energy associated with the motion of the Earth (we are assuming that the Sun is at
rest), and the potential energy due to the gravitational interaction between the Sun and the Earth. The angular
momentum is the product of the mass of the Earth and its velocity and distance from the Sun at a given moment.
Why are these quantities constant?
Noether’s Theorem
The German mathematician Emmy Noether in 1918 established a deep connection between conservation
laws and symmetries for a wide class of physical systems. Her theorem traced each of the conserved quantities,
which are physical quantities associated with a system which remain constant over time, to a symmetry. A
symmetry here is a change or transformation which leaves a system unchanged. In the problem of two interacting
bodies, the gravitational potential is independent of the time, and is spherically symmetric, which means that
it remains the same at all times, and in all directions for a given distance r. From these symmetries, Noether’s
theorem leads to the conserved quantities of energy and angular momentum respectively.
In our simple example, the conservation of energy and angular momentum can be understood from first
principles. Noether’s theorem becomes important in the more difficult and less transparent case of general
relativity. Here it is difficult to intuitively define conserved quantities, and it is necessary to explore applicable
symmetries and then to connect these to associated conserved quantities using Noether’s theorem. In this
manner, for the problem of two particles in general relativity we can define quantities which correspond to
energy and angular momentum. These reduce to their usual forms far from the gravitating particle M where the
gravitational field is weak, but remain useful even close to M.
1.3 The Effective Potential
We are considering a two-body problem for a Sun-Earth like system. As explained above, since the mass
M of the Sun is so much greater than the mass m of the Earth, it is possible to assume that the Sun is stationery
and only the Earth is in motion. The angular momentum of the system, which is then just the Earth’s orbital
angular momentum, is constant. The orbit is then confined to a single plane, so only two coordinates are needed
for the position of m to be known at any time. The two coordinates are taken to be the distance of the Sun from
the Earth r and an angular coordinate φ which is measured in the plane of motion. If the orbit is circular, then r
is a constant and only φ changes with time. If the orbit is an ellipse, say, then r changes with time and so does
φ. The conserved energy E per unit mass can be written as
1
2
(
dr
dt
)
2
+
r
2
2
(
dϕ
dt
)
2
+ V (r) = E
This equation is very simple to understand. On the left hand side the first term is the kinetic energy per unit
mass of radial motion, the second term is the kinetic energy per unit of angular motion, V(r) is the potential
energy per unit mass; their sum is the total constant energy E per unit mass. From the basic definition of angular
momentum L we can write
21
1.3 The Effective Potential
Figure 2: The effective potential for gravitational interaction of two bodies. Details are provided in the text
r ∗ 2(
dϕ
dt
) = L
where L is a constant. The two equations can be combined together to give, after some rearrangement of terms,
1
2
(
dr
dt
)
2
= E −V
e
(r)
Where the effective potential V
e
(r) is given by
V
e
(r) = −
GM
r
+
L
2
2r
2
We know that the rate of change of a potential is the force exerted per unit mass. The effective potential
contains two terms, the first corresponding to the gravitational force and other to the centripetal force due to the
curved motion. From the effective potential it is possible to understand in a simple way much about the possible
orbits of the particle.
When r becomes large, 1/r
2
reduces to zero more rapidly than 1/r. Therefore we can neglect the centripetal
term and the shape of the curve is determined by the gravitational potential. The potential goes to 0 as r becomes
very large. The energy of the particle m is then just the kinetic energy. As r decreases, the gravitational part
becomes more and more negative, as is shown in the figure, to the right of the green point. Eventually the
centripetal part dominates and the potential begins to increase and becomes very large as r becomes smaller. At
22
1.4 Types of Orbits
the green point, there is a minimum in the effective potential.
1.4 Types of Orbits
The total energy E, which is the sum of the kinetic energy and negative potential energy, can be >0, =0 or
<0. If E>0, then at great distances the potential energy is zero and the total positive energy is just the kinetic
energy. The dotted line in Figure 2 represents a particle with E>0. On one side the line extends to infinity,
while for small r, the line meets the effective potential at a point r
min
, where the effective potential is equal to
E and dr/dt=0. The line does not indicate the shape of the orbit, it only indicates how the radial coordinate r
changes for a given positive energy. At r
min
the particle “turns back” in the diagram and r begins to increase
again. What is the shape of such a orbit in space? We will see below that the orbit has the shape of a hyperbola.
Such orbits are followed by some comets: they come in from a great distance from the Sun, reach a closest
distance to the Sun when a cometary tail can be visible, and then swinging by the Sun, go to a great distance
again, never to return.
An orbit which just grazes the red horizontal axis has E= 0 at infinite distance. Such an orbit has the shape
of a parabola, the particle coming in with zero velocity at very large distances, and moving away after reaching
some r
min.
The most interesting are orbits with E<0. Such an orbit can never reach very large distances at which V
e
is
vanishingly small, because the kinetic energy term would have to be negative, which is not possible. So a orbit
with negative energy is forever trapped to remain within a finite distance of the mass that it is orbiting. The
dashed line in Figure 2 represents such a motion. The line now meets the effective potential in two points, r
min
and r
max,
where the former is the closest distance reached to M, and the latter is the furthest distance reached.
The closest distance is known as the perihelion of the orbit, while the furthest distance is known as the aphelion.
Again, the dashed line only indicates the range of r. The shape of the orbit is now an ellipse, like the orbit of the
Earth around the Sun.
The greater the energy of the particle, the higher is its line in Figure 2. Therefore, the greater the energy
E, the smaller will be r
min
and the larger will be r
max.
The ellipse then becomes more elongated. The orbit of
the Halley’s comet, for example, is highly elongated. It comes from a great distance, swings round the Sun and
goes again to its maximum distance, to return after about 80 years.
The lesser the energy of m, the lower is its line in in Figure 2. As the line lowers, r
min
moves outwards,
while r
max
moves inwards. So the perihelion increase and the aphelion decreases. For sufficiently low energy,
the two become equal, so that r is constant. This corresponds to the green point in the figure, where V
e
=E and
dr/dt = 0, so that r is constant. The orbit is circular in shape, with the particle rotating round the centre always
keeping the same distance. The shape of the Earth’s orbit is nearly circular, but nor exactly so. So the line
corresponding to the orbit would be located quite close to the minimum in the potential.
The shape of the effective potential shown in Figure 2 depends on the value of the angular momentum per
unit mass L. As L increases, the position of the minimum in the potential moves upwards and outwards. As L
increases, so do r
min
and r
max
for a given energy. When L decrease, r
min
and r
max
both decrease, and for L=0, the
only term in the effective potential is -1/r, so there is no minimum and the potential plunges to large negative
values as r becomes very small. L=0 means that the particle has no angular motion, and moves only radially.
Such a particle moving inwards plunges to the centre. If the particle moves outwards, then it escapes if E=0
or E>0. If E<0, then the particle moves radially outwards upto a certain distance, after which it falls back to
the centre. This corresponds to a the simple case of a ball thrown from the surface of the Earth. If the ball has
velocity equal to or greater than the escape velocity of 11.4 km/s, it escapes to infinity. If the ball has lesser
23
1.5 Orbital Shapes
Figure 3: Examples of an elliptical orbit (red), a parabolic orbit (green) and a hyperbolic orbit (blue). The
black dot is at a focus of the orbits.
velocity, it falls back to the Earth.
1.5 Orbital Shapes
The equations above tell us how r and φ change with time. While r can increase or decrease, φ always
changes in the same direction. We used the equations and the nature effective potential to develop some insight
into the nature of the orbits. But to get the actual shapes of the orbit, we need to know how r depends on φ. For
that, from the equations for dr/dt and dφ/dt, we can get the equation
dr
dϕ
= sqrt(
2r
2
(E −V
e
(r)))
L
It is possible to solve this differential equation, and as stated above, it turns out that there are three types of
orbits. For E>0 we have hyperbolic orbits, for E=0 parabolic orbits and for E<0 elliptical orbits. The solutions
to the equations provide all parameters for the orbits for given values of E and L. Orbits with the three possible
shapes are shown in Figure 3.
1.6 Approximations Made in the Two-Body Problem
We have assumed in the above treatment that the mass M is much greater than mass m, so M could be
taken to be at rest. If instead the masses are comparable, then we have to account the fact that the two bodies
are in orbit around each other. Even in this case the whole formalism developed above can be preserved by
introducing the concept of reduced mass µ, where
µ =
Mm
M + m
If we assume that a particle with the reduced mass is in orbit around a stationary particle of mass M+m,
24
1.6 Approximations Made in the Two-Body Problem
and r is the relative distance between them, then the whole formalism goes through unchanged. Another
approximation that we made is that the two particle have very small size which can be neglected. If finite sizes
are considered, then the two bodies could have their own spin angular momentum, which can interact with the
orbital angular momentum. The bodies may not be perfectly spherical which can change the shape of the orbits
by changing the gravitational potential. The two bodes can also affect each other through tidal forces. All of
these and other such affects can be taken into account, but those are matters for future stories.
The consideration above are all for a two-body problem. But the Solar system has eight planets, besides
many smaller bodies. The orbit of a given planet can be perturbed by the gravitational effects of the other
planets, particularly the larger and closer ones. These perturbations cause the orbit to depart from a perfect
ellipse. We will have more to say about this in the next story, where we also take into account the consequences
of using general relativity as the theory of gravitation, rather than Newton’s theory.
About the Author
Professor Ajit Kembhavi is an emeritus Professor at Inter University Centre for Astronomy and
Astrophysics and is also the Principal Investigator of the Pune Knowledge Cluster. He was the former director
of Inter University Centre for Astronomy and Astrophysics (IUCAA), Pune, and the International Astronomical
Union vice president. In collaboration with IUCAA, he pioneered astronomy outreach activities from the late
80s to promote astronomy research in Indian universities. The Speak with an Astronomer monthly interactive
program to answer questions based on his article will allow young enthusiasts to gain profound knowledge about
the topic.
25
Shedding Light On The Mysteries Of
Supergiant High Mass X-ray Binaries :
Cosmic Powerhouses Unveiled
by Sindhu G
airis4D, Vol.1, No.11, 2023
www.airis4d.com
2.1 Introduction
X-ray binaries (Fig: 1) are binary star systems in which one of the stars, often a compact object like a
neutron star or a black hole, accretes matter from its companion star. High-mass X-ray binaries constitute a
fundamental category within X-ray binary classifications. A High Mass X-ray Binary (HMXB) is a specific
type of binary star system in which one of the stars is a high-mass star (typically a massive O or B-type star), and
the other is a compact object, such as a neutron star or a black hole. These binary systems are characterised by
their strong X-ray emissions, which result from the interaction between the massive star and the compact object.
Supergiant high mass X-ray binaries are a prominent subtype among the high-mass X-ray binary systems. This
article will provide an explanation of supergiant X-ray binaries.
2.2 Supergiant High Mass X-ray Binaries (sgHMXBs)
A supergiant High Mass X-ray binary (Fig: 2) is a specific type of high mass X-ray binary star system.
These binary star systems are characterised by the presence of a supergiant star as the donor star, which is a
massive and highly luminous star, and a compact object (often a neutron star) as the accreting companion. These
two stars revolve around each other as a result of the gravitational attraction between them. The primary star in
an sgHMXB is a supergiant, which is typically of spectral type O or B, characterised by an intense, slow and
dense, radiatively steady and highly supersonic stellar wind, radially outflowing from the equator. Supergiants
are extremely large and luminous stars with masses much greater than that of the Sun. They are often in an
advanced stage of stellar evolution. Supergiant stars occupy the top region of the Hertzsprung–Russell diagram
(Fig: 3) with absolute visual magnitudes between about -3 and -8. The temperature range of supergiant stars
spans from about 3,400 K to over 20,000 K. The secondary star in the Supergiant high mass X-ray binary system
is a compact object, usually a neutron star. Supergiant high mass X-ray binaries (SgXBs) shine as some of the
most brightest X-ray sources in the sky.
The intense gravitational field of the compact object leads to the accretion of matter from the supergiant.
2.3 Classical Supergiant High-Mass X-ray Binaries
Figure 3: Hertzsprung-Russell Diagram Source: ESA.
In SgHMXBs, the supergiant star loses mass through powerful stellar winds and transfers this material to the
compact object. The transferred material forms an accretion disk around the compact object. The matter
falling onto the compact object generates intense X-ray emission, making sgHMXBs detectable by X-ray
telescopes. The X-ray emission is a distinctive feature of these systems. In a few exceptional cases, accretion
in supergiant high-mass X-ray binaries (sgHMXBs) can occur through Roche-lobe overflow, resulting in higher
X-ray luminosities compared to systems that primarily accrete via stellar winds. Such a mechanism has been
observed in Cen X-3, and it has been more recently proposed for IGR J08408-4503 at periastron. Supergiant
HMXBs have relatively short orbital periods, typically on the order of days to several weeks. This is because
the massive primary star loses mass through strong stellar winds, and this material is captured by the compact
companion.
sgHMXBs often exhibit variability in their X-ray emissions. This variability can be periodic and is usually
associated with the orbital motion of the two stars. The rate of mass transfer and X-ray emission can change as
the stars move relative to each other.
The INTEGRAL satellite possesses superior high-energy sensitivity compared to earlier generations of
hard X-ray observatories. As a result, sgHMXBs are no longer in the minority. Supergiant high mass X-ray
binaries constitute approximately one-third of all the HMXBs identified within the Milky Way.
2.3 Classical Supergiant High-Mass X-ray Binaries
Classical sgHMXBs are those where the X-ray emission from the system is relatively unobscured and easily
detectable. In these systems, the X-ray radiation from the binary components can escape into space without
significant interference from intervening material. These systems are often characterised by prominent and
28
2.4 Obscured Supergiant High-Mass X-ray Binaries
stable X-ray emission. Neutron stars or black holes in classical sgHMXBs accrete matter from their massive
supergiant companion stars, leading to the emission of X-rays. Examples include Vela X-1 and Cygnus X-1.
The interactions between the neutron star and the stellar wind in Vela X-1 have unveiled that the massive star’s
wind experiences significant disruption due to the gravitational and photoionization effects of the neutron star.
2.4 Obscured Supergiant High-Mass X-ray Binaries
Obscured sgHMXBs are those in which the X-ray emission from the system encounters a significant level
of absorption due to the presence of dense, intervening material, such as stellar winds, circumstellar disks, or
interstellar dust. This absorption can make the X-ray emission from the system more challenging to detect
or study. Obscured sgHMXBs are often associated with Be stars that have dense circumstellar disks or other
massive stars with dense stellar winds. INTEGRAL, a high-energy space telescope, has been instrumental in
discovering and studying obscured sgHMXBs, thanks to its ability to detect X-ray radiation at high energies
that can penetrate through the obscuring material.
Heavily obscured supergiant high-mass X-ray binaries (sgHMXBs) exhibit certain attributes similar to
classical sgHMXBs. However, a notable contrast lies in the fact that obscured sgHMXBs experience substantially
higher X-ray absorption, typically around ten times greater than what is observed in classical systems, surpassing
the galactic absorption levels.
2.5 Some Examples Of sgHMXBs
2.5.0.1 Cygnus X-1
Cygnus X-1 is one of the most famous sgHMXBs. It consists of a supergiant star and a black hole. Cygnus
X-1 was the first X-ray binary discovered and remains an important object of study in high-energy astrophysics.
The binary system has a relatively short orbital period, with the supergiant star completing an orbit around the
compact object in approximately 5.6 days.
2.5.0.2 Vela X-1
Vela X-1 is another well-known sgHMXB, and it contains a neutron star as the compact companion. It
is part of the Vela constellation and is a strong X-ray source. The binary system has a relatively short orbital
period of about 8.964 days. This means that the supergiant star completes an orbit around the neutron star in
just under nine days.
2.5.0.3 Cen X-3
Centaurus X-3 is an sgHMXB located in the Centaurus constellation. It consists of a massive supergiant
star and a neutron star. The system exhibits X-ray emission due to accretion. The two stars orbit each other
every 4.8 hours.
2.5.0.4 LMC X-4
This sgHMXB is located in the Large Magellanic Cloud, a satellite galaxy of our Milky Way. LMC X-4
contains a supergiant star and a neutron star, and it’s known for its variable X-ray emission. The two stars orbit
each other every 1.4 days.
29
2.6 Importance Of The Study Of sgHMXBs
2.6 Importance Of The Study Of sgHMXBs
sgHMXBs are of great interest to astronomers and astrophysicists because they provide a unique opportunity
to study the interaction between massive stars and compact objects. They offer insights into the physics of
accretion processes and the behaviour of matter under extreme gravitational conditions. Studying sgHMXBs is
essential for understanding the evolution of massive stars, the formation of compact objects, and the physical
processes occurring in these extreme environments. These systems are valuable laboratories for astrophysical
research and contribute to our knowledge of the high-energy astrophysical phenomena associated with X-ray
binary systems. sgHMXBs are powerful X-ray sources, which can be used to study the interstellar medium and
the evolution of galaxies.
References:
Accretion in supergiant High Mass X-ray Binaries
Towards a unified view of inhomogeneous stellar winds in isolated supergiant stars and supergiant high
mass X-ray binaries
Optical/infrared observations unveiling the formation, nature and evolution of High-Mass X-ray Binaries
The dark side of supergiant High-Mass X-ray Binaries
High-Mass X-ray binary: Classification, Formation,and Evolution
X Ray Binaries Monitoring
A catalogue of high-mass X-ray binaries in the Galaxy: from the INTEGRAL to the Gaia era
Formation of wind-captured disks in supergiant X-ray binaries Consequences for Vela X-1 and Cygnus
X-1
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.
30
Part III
Biosciences
The common Ectoparasite in Ornamental
Fishes
by Geetha Paul
airis4D, Vol.1, No.11, 2023
www.airis4d.com
1.1 Introduction
Interest in fishes can be divided into two separate “streams’ ’- for the table and for ornamentation. The
ancient Egyptians may have been the first aquarists.They kept tanks of cold water for decoration. Certainly,
selective breeding of colour strains started in China and the cultivation of gold fishes was well established in
the Sung dynasty. Later in the sixteenth century, keeping fishes in Glass Bowls spread to Europe.
As devoted enthusiasts of ornamental fish, we are deeply invested in the happiness and health of our aquatic
companions- the freshwater ornamental fishes. This study delves into the world of two common ectoparasites
of the same genus Argulus, Argulus foliaceus and Argulus japonicus, endearingly known as ’fish lice,’ which,
despite their small size, pose significant threats to our fish friends. These parasites, residing in our fish tanks and
aquariums, can cause various health issues, notably skin problems, and serve as vectors for diseases within our
aquatic community. Through our careful observations, we’ve noticed subtle signs of distress in our fish, such
as decreased activity levels and the drooping of their tails. These observations have heightened our awareness
of the urgent need for effective management strategies. By understanding the impact of these ectoparasites, we
can take proactive measures to protect our aquatic companions. Our ultimate goal is to maintain the joy and
vitality that these beautiful creatures bring into our lives, ensuring they thrive in their underwater heaven.
1.2 Symptoms of Infection:
Argulus infestations in ornamental gold fishes is a great problem in this industry and among the fish lovers..
Based on the intensity of infestation, symptoms such as pale coloration and erratic swimming are observed
in infested goldfish. Clinical signs such as bleeding fins, losing scales, fin rot, acute hemorrhagic inflamed
skin wounds, and superficial mucus expulsion etc. are also observed.(Figure 1.2) Argulosis is one of the most
common ectoparasitic infestations found in both marine and freshwater fishes. It was reported to be the cause
of the high rate of mortality in fish farms of the USSR in the 1950s [Bauer1962].
A. foliaceus is oval with a rounded carapace,2 compound eyes, sucking mouth parts with a piercing stylet
and 2 suction cups.5 - 7mm in length and 5 - 6mm in width. Female is larger than the male . A. japonicus is
a small golden brown ectoparasite with a stumpy tail.4 - 9 mm long 3 - 6 mm width, ribbed suckers,flagellated
swimming legs.
1.2 Symptoms of Infection:
Figure 1: Shows- Argulus infested goldfish and Argulus on the operculum of the Goldfish.
Figure 2: The parasite Argulus attached to the tail of a goldfish along with a microscopic view of the parasite
is shown. Blood stains resulting from the wounds produced are also seen on the tail.
Figure 3: Shows- A. foliaceus male and female. A. japonicus male and female
33
1.3 Mode of Propagation:
Figure 4: Shows the mature parasite (Argulus) on the body of the host for copulation
Figure 5: Shows the detaching female after copulation to lay eggs.
Image courtesy: https://winvertebrates.uwsp.edu/gorzek 490 2014.html
Body can be divided into the head, thorax and abdomen. The head is covered by a flattened horseshoe-
shaped carapace, maxillipeds, peroral sting and basal glands. The thorax has four segments. The mouthparts
of Argulus are greatly reduced, and the most distinguished feature is the modification of the second maxillae
into two suction cups by which the parasite holds onto its host, each bearing a pair of swimming legs and the
abdomen as a simple bilobed segment.
1.3 Mode of Propagation:
Argulus has a direct life cycle, meaning it only requires one host (Goldfish and Koi carp) to completely
develop from an egg to a mature, reproducing adult. The parasite swims inside the water, senses the host by
vision, smell,touch and gets attached to the host by its suckers. Usually the copulation is seen on the host’s
body surface but copulation has also been observed on other types of solid surfaces such as leaves and stones.
(Pasternak, et al., 2000). Argulus foliaceus most commonly attaches to the host on the skin epithelium of the
body and fins, but has been observed to also attach to the gills. This louse feeds by piercing the host skin,
injecting a toxin, and drawing blood. They use their mandibles to scrape skin into their mouth. They use their
needle-like mouthparts to inject chemicals into the host’s body. These chemicals may help turn nearby tissues
into liquid so that they can be sucked into the mouth. Infection of A. foliaceus often results in host tissue
damage and sometimes mortality. Argulus foliaceus infects irrespective of the health of the host fish. It causes
pathological changes to the fish, due to direct tissue damage and secondary infections.
After taking a meal, mature females leave their hosts to lay eggs. Eggs are laid on bottom vegetation as
34
1.4 Something Fishy in your Aquarium?
Figure 6: Shows the precautions to reduce infections
Figure 7: Table shows the treatment and dosage of Tri Chloroform, Ivermectin and doramectin
strings or clutches that contain an average of 100-150 eggs, but there can be as few as four or as many as 250
eggs. The female lays her eggs in the winter, but there is a wide variability in hatching time, possibly due to fish
availability. (Harrison, et al., 2006; Pasternak, et al., 2000). They glue their eggs in rows on hard surfaces and
leave them to hatch on their own. The newly hatched larvae do not resemble the adults at all. Their antennae,
mouthparts, and first two pairs of thoracic limbs are bristly and used for swimming. They grow by moulting,
or shedding their external skeletons (exoskeletons). After the second moult, fish lice replace the bristles on the
antennae with strong claws in preparation for their new lives as parasites. The claws are used to grab on to
their first host. As they grow and develop, fish lice will change hosts several times. As the larvae mature, they
develop thoracic limbs, sucking mouthparts, and reproductive organs.
1.4 Something Fishy in your Aquarium?
Does your fish scratch against rocks in the tank? It may have an itchy infestation of fish lice (A. foliaceus).
These tiny crustaceans look like clear discs (0.75 inches, 19.05 millimetres) with eyes attached to their scaly
skin. If you find them, put the infested fish(es) in a separate tank. Check with a pet shop for treatment options for
both fish and tank. Parasites can be pulled off with tweezers and their wounds treated with special antiparasitic
medicines.
1.5 Control measures
In the case of aquarium fishes, parasites can be removed manually from parasite infested fish. Although
this method is quite effective, it is labour intensive and impractical when there are large numbers of fish or
parasites. Furthermore, prolonged handling of fish to remove parasites can cause physical and physiological
35
1.5 Control measures
stress and may be impractical with small fish. The job becomeseven more difficult when the parasites are
notlarge enough to be seen by the naked eye. This method is probably more applicable for fish infected by
crustacean parasites which are relatively large enough to be removed manually. Incomplete extraction of the
attached animal or rupture of host muscle tissues can cause damage, leading to secondary microbial infections.
Most of the applied methods to fight against crustacean ectoparasites are based on the use of chemicals like
Ivermectin, Doramectin,Tri Chloroform bath etc. has been widely used in freshwater aquaculture. Ectoparasites
on freshwater fish may be eliminated by immersion of the infected host into high NaCl concentrations.
References
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10090776/
https://www.msdvetmanual.com/exotic-and-laboratory-animals/aquarium-fish/bacterial-diseases-of-fish
https://animaldiversity.org/accounts/Argulus foliaceus/
https://winvertebrates.uwsp.edu/gorzek 490 2014.html
https://animals.jrank.org/pages/1853/Fish-Lice-Branchiura-BEHAVIOR-REPRODUCTION.htm
https://www.researchgate.net/publication/24444249.The influence of risk factors on the abundance, egg
laying habits and impact of Argulus foliaceusin stillwater trout fisheries.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10090776/
Cross, D., R. Stott. 1974. The effect of Argulus foliaceus L. on the growth and mortality of a grass carp
population. Fisheries Management, 5: 39-42.
Kumar, S., Kumari, P. (2021). Advances in Management Methods for Argulosis in Aquaculture. In: Gupta,
S.K., Giri, S.S. (eds) Biotechnological Advances in Aquaculture Health Management . Springer, Singapore.
https://doi.org/10.1007/978-981-16-5195-3 19
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.
36
Part IV
General
LiDAR - A versatile Imaging Technology
by Ninan Sajeeth Philip
airis4D, Vol.1, No.11, 2023
www.airis4d.com
Figure 1: This is the image of a forest taken by a terrestrial LIDAR. The striking difference is the clarity of the
details that in an optical camera, the imaging technology would smooth out and distort, Image Credit: YouTube.
LiDAR, which stands for Light Detection and Ranging, is a remote sensing technology that uses laser
pulses to measure the distance of objects and create 3D maps of the environment. Compared to other imaging
devices such as cameras and radars, LiDAR offers several advantages that make it better suited for specific
applications. For example, LiDAR generates highly accurate and precise data with laser-level accuracy, providing
cm-level precision (mm precision in some 2D LiDARs). At the same time, the radar’s resolution is significantly
lower, posing challenges in precise tracking and distinguishing individuals or objects in crowded environments.
LiDAR is also more resilient in harsh lighting and weather conditions than cameras.
LiDAR technology has become a critical tool in various industries, including autonomous vehicles, envi-
Figure 2: Sweep is a California-based technology startup that has developed a low-cost LiDAR system that
is even more efficient than its predecessors. It uses intermediate pulse trains of different patterns to reduce
computing power as well as the required power output of the laser source. Image Credit: IEEE Spectrum.
ronmental monitoring, archaeology, and robotics.
Though these make it very promising, there are several challenges ahead. The disadvantages of LiDAR
compared to other imaging devices are relatively few, but they do exist.
1. Cost: One of the most significant disadvantages of LiDAR is its high price tag. However, the cost of
LiDAR is rapidly decreasing, with newer sensors and built-in sensors and microcontrollers making them lighter
in required accessories than their predecessors.
2. Computing power: LiDAR requires a significant amount of computing power compared to cameras
and radar, as they need to compute hundreds of thousands of independent pixels every second and transform
them into actions, which makes it prone to system malfunctions and software glitches.
3. Limited detection of colour and texture: LiDAR depends on pulse trains of laser that is mostly
monochromatic. Hence by design, they cannot detect colour or texture, which can be a disadvantage in some
applications. For example, LiDAR cannot read the words on a signboard or the colour of a stoplight.
4. Limited range: LiDAR has a limited measurement range compared to radar. While LiDAR can detect
objects up to 100 meters away, radar can see objects kilometres away.
However, LiDAR is widely used in the design of autonomous vehicles. It helps the vehicle sense and
understand its surroundings by creating 3D mappings of its environment, including objects like buildings,
roads, and other vehicles. This information is then combined with sensory inputs and maps to ensure safe
navigation. Because of its fast and accurate mapping capability with almost zero focusing issues, (See Figure
1), LiDAR can track obstacles and vehicles to maintain safe distances and identify road signs, traffic signals, and
road markings for real-time hazard analysis.LiDAR is also used to estimate carbon stocks, flood risks, erosion,
and the suitability of forest habitats in environment science.
Compared to other imaging devices such as cameras and radars, LiDAR offers several advantages that
make it better suited for specific applications. Given below are some benefits that LiDARs offer.
1. Accuracy and precision: One of the primary advantages of LiDAR is its accuracy and precision.
LiDAR generates highly accurate and precise data with laser-level accuracy in precise tracking and distinguishing
39
1.1 Do it Yourself LiDAR
individuals or objects in crowded environments. LiDAR is also more accurate than cameras, which can become
unreliable in shadows, bright sunlight, or the oncoming headlights of vehicles.
2. Resilience in harsh conditions: Because they use pulse laser beams, LiDAR is more resilient in
extreme lighting and weather conditions than optical surveillance cameras. LiDAR systems efficiently work
in any lighting condition, giving them an advantage over cameras that need proper lighting to perceive their
surroundings. Radar also works well in poor weather conditions, but the antenna size limits its resolution.
3. 3D perception: LiDAR provides a 3D map of the world, which can create accurate models of objects
or terrain. This 3D perception you observe in Figure 1 is highly accurate compared to cameras because the
laser pulses are sent and captured by LiDARs that overcome the limitations faced by ordinary cameras in the
presence of shadows, bright light, etc. LiDAR thus gives self-driving cars a three-dimensional image of their
surroundings.
4. Detection of featureless objects: LiDARs,with their ability to precicely measure distances are better
suited for detecting featureless objects like blank walls, whereas a vision-based camera might fail. LiDARs can
thus estimate gaps between leaves to judge the health of a tree or the surface elevation and slope of the terrain
precisely. Their ability to penetrate dense foliage and forest canopies gives them an edge over other methods to
estimate the habitats and carbon footprints precisely.
5. All-weather operation: LiDAR systems work in any lighting condition, giving them an advantage over
optical cameras, which depend on incoming light to perceive their surroundings.
1.1 Do it Yourself LiDAR
One of the challenges that hinters LiDAR technology development is its cost. A reasonably good 2D or
3D LiDAR scanner alone would cost a few thousand USD or more, making it out of reach for most enthusiasts.
However, Scanse, a California-based company, has an innovative technology for producing low-cost, efficient
LiDAR called Sweep. Instead of continuous trains of laser pulses, they use groups of micro pulses that are
independently tracked and processed, making the system less expensive and more efficient in identifying multiple
echos from scattered beams. These units can work with embedded systems like Raspberry Pi, offering custom
building for specific applications. Sweep weights 120 grams with a measurement range of 40 meters with high
tolerance to ambient light, gathering 500 points per second with an accuracy of 1 centimetre. These capabilities
make it highly impressive for anyone wanting to make a baby walk with the new technology.
Sweep has an add-on kit that makes it an efficient 3D scanning LiDAR and costs less than $400 per unit.
They are also open source and provide much documentation on every detail of functionality and its application.
The goal is to develop it into a fully functional and user-friendly 3D scanning LiDAR that would be affordable.
Complete technical information is available in [1]
1.2 Conclusion
In conclusion, LiDAR is a remote sensing technology that uses laser pulses to measure the distance of
objects and create 3D maps of the environment. Compared to other imaging devices, LiDAR offers several
advantages that make it better suited for specific applications. LiDAR has become a critical tool in various
industries, including autonomous vehicles, environmental monitoring, archaeology, and robotics. While some
trends and devices may replace LiDAR technology in the future, LiDAR remains a valuable and essential tool
for many applications.
40
1.2 Conclusion
References
[1] https://s3-us-west-1.amazonaws.com/scannable/SWEEP DATA SHEET.pdf
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.
41
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.
