UNIT 2: ADVANCED CONCEPTS OF MODELLING IN AI

🟠 A. Short Answer Type Questions

1. Distinguish between training data and testing data.

Training data is used to train the AI model, while testing data is used to evaluate its performance.

2. Define convolution.

Convolution is a simple mathematical operation achieved by multiplying together two arrays of

numbers to produce a third array of numbers.

3. Define reinforcement learning.

The type of AI model that aims to maximise rewards through trial and error is called a reinforcement
learning model.

4. In this learning model, the dataset that is fed to the machine is labelled. Name the model.

Supervised learning

5. What is dimensionality reduction?

Dimensionality reduction refers to reducing the dimensions and still being able to make sense out of
the data. As we reduce the dimension of an entity, some information is lost.

6. What is Machine Learning?

Machine Learning (ML) is a subset of AI where machines learn from data without being explicitly
programmed. It improves automatically through experience and uses algorithms to identify patterns.

7. Define Regression in Supervised Learning.

Regression is a type of supervised learning where the output is a continuous value. It is used to
predict numerical values, such as temperature, prices, or age.

8. What are ANNs?

Artificial Neural Networks (ANNs) are computing systems inspired by the structure of the human
brain. They consist of layers of interconnected nodes (neurons) and are used to recognize patterns
and solve complex problems.

9. How do ANNs handle large datasets?

ANNs handle large datasets by:

 Learning from multiple examples.

 Using parallel processing.

 Adjusting weights automatically through training.

 Being capable of generalizing patterns across big data.

10. What happens in the output layer of a neural network?

The output layer gives the final prediction or decision of the neural network. It transforms signals
from the hidden layers into a readable result, like a class label or a predicted value.

11. Define CNN.

A Convolutional Neural Network (CNN) is a specialized kind of ANN used for image and video
recognition. It uses convolutional layers to extract features and detect patterns like edges, colors, or
textures.

12. Define Association in Unsupervised Learning.

Association is a rule-based method in unsupervised learning used to find relationships or patterns

among items in large datasets. Example: "If a customer buys bread, they also buy butter."

13. What is the significance of hidden layers in neural networks?

Hidden layers allow the neural network to:

 Learn complex patterns in data.

 Perform feature extraction.

 Improve accuracy by enabling deeper understanding through non-linear transformations.

14. Define the learning-based approach

He learning-based approach, the machine learns on its own. The data used in this model is generally
labelled or unlabelled.

15. What is the main goal of modelling in the Al Project Cycle?

The main goal of AI modelling is to select an appropriate AI model that can learn from data and make
predictions.

16 . What advantages do ANNs have over traditional machine learning algorithms, especially in
handling large datasets?

Neural networks, unlike traditional algorithms, continue to improve with more data and perform
exceptionally well with large, complex datasets due to their ability to learn on their own.

17. What is the primary inspiration for the design of neural networks?

Neural networks are designed to mimic the structure and processing of the human brain.

18. Define Machine Learning.

Machine learning enables machines to learn on their own and improve with time and experience. In
machine learning, machines learn from new data fed to them during the testing phase and use this
knowledge to improve their performance in subsequent iterations. Machine learning algorithms
learn from mistakes and also consider exceptions to become more accurate and efficient. For
example, predicting the weather for the next seven days based on the data from the previous year
and week
19 . Define Artificial Intelligence.

AI refers to the field of computer science that creates machines capable of performing tasks that
require human-like intelligence. AI-based systems use algorithms and data to process information
and produce the required output. AI machines are capable of learning on their own without human
intervention. Some applications of AI include facial recognition, autonomous vehicles, and devices
that understand voice commands in natural language.

20. List two important features of AI.

1. AI machines are trained with data and algorithms.

2. AI machines learn from mistakes and experience.

21. Define Deep Learning.

Deep learning enables machines to learn and perform tasks on a large amount of data, or Big data.
Due to the large amount of data, the system learns on its own by using multiple machine learning
algorithms working together to perform a specific task. DL is based on neural networks with multiple
layers that mimic the workings of the human brain. The DL models understand how the human brain
works in different situations and then try to recreate its behaviour. For example, a deep learning
model in a driverless car can identify a person crossing the road.

🟢 B. Long Answer Type Questions

1. Describe the relationship between Artificial Intelligence (AI), Machine Learning (ML), and Deep
Learning (DL).

AI is a broad term that includes both machine learning and deep learning. Machine learning is a sub-
category of AI, and deep learning is a subset of machine learning as it includes multiple machine
learning modules. Deep learning is the most advanced form of AI among these three. AI includes all
the concepts and algorithms that mimic human intelligence. All AML and DL algorithms are AI
algorithms, but not all AI-based algorithms are ML or DL. But all DL algorithms are ML. algorithms

2. Describe the structure of a neural network. How does it help us?

A neural network consists of multiple layers, ie, the input layer, hidden layers, and the output layer.
The input layer's role is to receive data and pass it to the network without processing. Hidden layers,
situated between input and output Layers, perform all the computation. Each node in a hidden layer
has its own machine learning algorithm, which processes data and passes the output to the next
hidden layer. This process continues until the last hidden layer, which forwards the processed data to
the output layer. The output layer, like the input layer, does not process data but presents the final
computed outcome to the user. This layered structure allows neural networks to learn complex
patterns and relationships in data.

3. List the differences between supervised and unsupervised learning models.

Supervised learning works on labelled datasets. It is a task-driven process where models learn to
make predictions or decisions based on input data and corresponding output labels. In a supervised
learning model, we divide the data into a training dataset and a test dataset. The training dataset is
used to train the model, while the test dataset is used to evaluate the model's accuracy in predicting
output.

Unsupervised learning works on unlabelled datasets. It is a data-driven process that allows the
algorithm to explore the data and find its insights by identifying relationships, patterns, and trends.
Its goal is to find similarities and differences between data points.
4. List the importance of neural networks.

Neural networks are important for the following reasons:

i. A neural network can learn 'how' to solve a specific problem efficiently from the training data
provided to it.
ii. Neural networks can process large datasets and perform tasks like speech recognition or image
recognition at an extremely high speed.
iii. They can automatically identify important data features without needing input from a developer.
iv. Neural networks are specialised machine learning algorithms designed to perform specific tasks
quickly and efficiently with high accuracy.
v. When it comes to large datasets, neural networks perform much better than traditional machine
learning algorithms. Unlike traditional machine-learning algorithms that reach a saturation point
and stop improving, large neural networks show better performance with a large amount of data.

5. List the differences between AL, ML, and DL

The following is the basic difference between AL, ML, and DL:

Artificial Intelligence Machine Learning (ML) Deep Learning (DL)

(AI)

AI aims to mimic Machine learning aims to create DL aims to build neural networks to
human intelligence to machines that can learn on their mimic the working of the human
create intelligent own using data and improve over brain and use complex algorithms
machines and time without being programmed and large volumes of data to enable
programs. for it. an AI model to learn.

It is a broad field of It is a subset of AI that involves It is a subset of AI and ML that

computer science that algorithms that learn patterns focuses on training deep neural
simulates intelligent from data. networks.
behaviour.

6. Differentiate between a Rule-Based and a Learning-Based Approach:

Rule-Based Approach Learning-Based Approach

The machine follows rules set by the Machine Learns patterns on its own
developer. from data.

AI is achieved through a rule-based AI is achieved through a learning

technique technique

The labelled data set is used. An unlabelled dataset is used.

It may require less training time. It may require more training time.

7. List the Advantages of Neural Networks:

 Can handle non-linear and complex data.

 Capable of learning and improving over time.

 Effective in pattern recognition (images, voice).

 Works well with large datasets.

  Can be used in varied applications like medical diagnosis, finance, and gaming.

8. Describe the Role of Input, Hidden, and Output Layers in a Neural Network:

 Input Layer: Receives the raw data. Each node represents one feature (e.g., height, color,
etc.).

 Hidden Layers: Perform computations on the inputs. These layers extract patterns and
transform the data.

 Output Layer: Produces the final result, such as a classification label or numerical value.

9. Explain Clustering:

Clustering is an unsupervised learning technique that groups data based on similarities. Each group
is called a cluster—for example, Grouping customers based on purchasing behavior.

10. Describe the Following AI Models and Draw Diagrams:

a. Regression:

 Predicts continuous values.

 Example: Predicting the price of a house based on size.

b. Dimensionality Reduction:

 Reduces the number of input features while keeping important data.

 Helps in removing noise and improving model performance.

11. Neural networks are said to be modelled after the way neurons in the human brain behave. A
similar system is mimicked by the AI machine to perform certain tasks. Explain how neural
networks work in an AI model and mention any three features of neural networks.
Neural networks work like the human brain. Data enters through the input layer, is processed in the
hidden layers, and output is generated. They learn by adjusting weights based on errors using
algorithms like backpropagation.

✅ Three Features of Neural Networks:

1. Self-learning: Improves with more data.

2. Generalization: Applies learning to new situations.

3. Layered structure: Supports deep learning and complex tasks.

12. Discuss the fault tolerance feature of artificial neural networks and its significance.

The data is stored and distributed across the entire neural network, such that even if some nodes of
the network are unavailable or get corrupted, the entire network keeps functioning.

13. Sooji is a beginner in the field of Artificial Intelligence. She got confused in core terms like
Artificial Intelligence (AI), Machine learning (ML), and Deep Learning (DL). Many times, these
terms are used interchangeably, but are they the same? Justify your answer. Help her understand
these terms by drawing a well-labelled diagram to depict the interconnection of these three fields.

No, Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL) are not the same,
although they are closely related and often used interchangeably. AI is the overarching field
encompassing the concept of creating intelligent machines. Machine learning is a subset of AI that
enables systems to learn from data without explicit programming. Deep learning is a specialized
subset of ML that uses artificial neural networks with multiple layers to analyze data.

14. What is modelling?

AI modelling refers to the development of a program or algorithm that can be used to draw
conclusions or generate predictions based on the available data. At this stage, an appropriate data
model is selected and trained with the data that has been acquired and explored in order to achieve
the goal of the AI project. To understand and analyse data as well as find patterns and trends in it,
machines need data in the form of binary numbers, ie, 0s and 1s.
15. Explain the Rule-based Approach.

Rule-Based Approach to AI Models

 Definition: The developer defines relationships and patterns in data using "if-then" rules.

 How it Works: Labelled data is fed to the model along with predefined rules. The machine
follows these rules to produce specific outputs.

 Example (Fruit Recognition):

o To predict if a fruit is a banana or an orange, the machine is trained with labelled

images ("banana" or "orange").

o Rules are effectively embedded through these labels.

o When shown an orange, it compares it to the trained data and identifies it.

 Limitation: The model can't recognize data for which no rules (or labels) were defined (e.g.,
showing an apple when only trained on bananas and oranges).

 Drawback: Learning is static; the machine doesn't adapt to changes in the original training
data after it's been trained.

16. Explain the Learning based Approach, name its subtypes.

Learning-Based Approach to AI Models

 Definition: The machine learns independently by identifying patterns and relationships in

data.
 Data Type: Can use both labelled and unlabelled data.
 Rule Generation: The AI model defines its own rules, rather than the developer.
 Output: The machine communicates observed trends in the training data.
 Example (Tree Image Analysis):
o Given 1000 random images of trees without predefined features.
o A learning-based AI will identify patterns (e.g., height, color, fruit, leaf types).
o It will then group or cluster images based on these discovered patterns.
o May even discover unexpected clustering algorithms.
 Subtypes of the Learning based model are Machine Learning and Deep learning.

17. Explain the learning based models of machine learning.

1. Supervised Learning:

 Data Type: Works with labelled datasets, where each piece of data has a corresponding tag (label)
defining what it represents (e.g., an image of a carrot labeled "carrot").
 Process: It's a task-driven process where the model learns to predict or make decisions based on
input data and known output labels.
 Dataset Division: Data is split into:

 Training dataset: Used to teach the model.

 Test dataset: Used to evaluate the model's accuracy.

 Learning Mechanism: The model learns from known results.

 Key Outcome: Aims for high accuracy in predictions.
 Efficiency: Generally faster training due to pre-existing knowledge of results.

I. Classification:

 Definition: A supervised learning task where the model learns to assign input data into
predefined, discrete categories.
 Training Data: Requires labelled training data, where each data point is assigned a specific
category.
 Process:
o The model is trained on multiple examples of data, each with its corresponding
category label (e.g., images of guitars labelled "guitar" and pianos labelled "piano").
o It learns the patterns and features associated with each category.
o When presented with new, unseen data, the model classifies it into one of the
learned categories.
 Data Type: Works with discrete datasets—data that can be grouped into distinct, separate
categories.
 Examples:
o Classifying emails as "spam" or "not spam."
o Identifying handwritten digits (0-9).
o Categorizing images of animals (e.g., "cat," "dog," "bird").
 Example Algorithm: Email spam detectors.
 II. Regression:

 Definition: A supervised learning model used to predict continuous numerical values.

 Data Type: Works on continuous data, which can take any value within a given range.
 Purpose: Predicts the behavior of one variable based on the value of another.
 Mechanism: Uses a mapping function (f(x)) to help predict future values based on the data
we have.
 Examples:
o Predicting future stock prices based on historical data.
o Predicting tomorrow's temperature based on past weather data (humidity, cloud
cover).
o Predicting an employee's future salary based on past salary data.
 Visual Representation: Often depicted as a solid line on a graph that represents the learned
pattern (the prediction function) among the data points.

2. Unsupervised Learning:

 Data Type: Works with unlabelled datasets, where the data is random, and the trainer may have
no prior information about it.
 Purpose: Identifies hidden relationships, patterns, and trends within the data without predefined
guidance.
 How it Works:
 The model receives raw, unlabelled data (e.g., images of vegetables without names).
 It independently discovers patterns, groups similar data, or extracts insights from the data.
 Learns from "experience" by exploring the data itself.
 Nature: It's a data-driven process allowing the algorithm to explore and find its insights.
 Goal: To find similarities and differences between data points.
 Example (Stray Dog Data):
  Given unlabelled data of stray dogs.
 The algorithm analyzes the data to identify patterns or trends (e.g., shape, size, color,
behavior).
 May discover both known patterns (like color) and unexpected ones (like unusual size
groupings).

I. Clustering:

 Definition: An unsupervised machine learning technique where the model groups an unlabelled
dataset into different "clusters" based on patterns or rules it generates autonomously.
 Data Type: Typically uses unlabelled or random data.
 Process:
 The developer feeds training data directly to the machine.
 The machine identifies patterns or trends to group datasets that share common
characteristics.
 It builds its algorithm by finding similarities and differences within the unlabelled data.
 Dataset Compatibility: Works with discrete datasets—data that can be naturally grouped into distinct
categories.
 Output: The model might generate unusual or unexpected groupings as it discovers its patterns.
 Real-life Example: Search engines utilize clustering to group similar search results together, providing
the closest matches to your query while separating dissimilar information.
 Visual Representation: Data points within a cluster are more similar to each other than to data points
in other clusters.

II. Dimensionality Reduction :

 Human Perception Limit: Humans easily visualize up to three dimensions (length, width, height).
 High-Dimensional Data: Many real-world theories and algorithms (e.g., in Natural Language
Processing, where words can be in "N-dimensions") involve data with more than three dimensions,
which is impossible for humans to visualize.
 Purpose: To simplify complex, high-dimensional data into fewer dimensions, making it easier to
work with and understand.

 How it Works (Analogy):

 Imagine a 3D ball photographed into a 2D image.
 The 2D image loses information (e.g., the back of the ball, its full color consistency).

Further reduction leads to more information loss.

 Trade-off: Dimensionality reduction simplifies data but inherently comes at the cost of losing some
information.
 Outcome: It makes complex data simpler while still allowing for meaningful interpretation.
 Example: Data compression.

III. Association:

 Definition: Discovering relationships or patterns in data where the presence of one item is
linked to the occurrence of another.
 Purpose: Identifies interesting connections between variables in large datasets.
 Primary Application: Frequently used in Market Basket Analysis.
 Market Basket Analysis Example (Rule: If Bread → Butter):
o Analyzes customer purchase patterns.
o Identifies items often bought together (e.g., customers buying bread are highly likely
to also buy butter).
o Applications derived from such rules:
 Product Placement: Placing associated items together on shelves (e.g.,
bread and butter).
 Cross-selling: Suggesting associated items as add-ons (e.g., offering butter
online when bread is purchased).
 Promotions: Offering discounts on associated items (e.g., discount on
butter/jam when bread is bought).
 General Applications:
o Market Basket Analysis (product recommendations).
o Recommender systems.
o Web usage mining.

3. Reinforcement Learning (RL):

 Definition: An AI model that learns to maximize rewards through trial and error by
interacting with an environment.
 Mechanism: It's a feedback-based learning technique:
o Positive feedback (rewards) for good actions.
o Negative feedback (penalties/punishments) for bad actions.
 Key Features:
 o Automatic Learning: Learns without labelled data, solely based on feedback.
o Experience-Driven: Learns from experience and mistakes.
o Environmental Interaction: The agent interacts with and explores its environment
autonomously.
o Goal: Maximize total cumulative positive rewards over time.
o Autonomous & Self-Teaching: A trial-and-error process that learns to achieve
optimal outcomes.
 Goal: To find an optimal action model that maximizes the agent's total cumulative reward.
 Examples:
o Online Recommendation Systems: Agents receive "rewards" (user clicks, purchases)
to refine recommendations.
o Automated robots.
o NLP-based chatbots.
o Self-driving cars.

Supervised Unsupervised learning Reinforcement learning

learning

Supervised learning For unsupervised learning, Reinforcement learning is

requires that the target typically either the target learning what to do and how to
variable is well defined and variable is unknown or has only map situations to actions. The
that a sufficient number of been recorded for too small a learner is not told which actions
its values are given. number of cases. to take.
Supervised Learning deals Unsupervised learning deals Reinforcement Learning deals
with two main tasks: with clustering and associative with exploitation or exploration,
regression and rule mining problems. Markov's decision processes,
classification. Policy Learning, Deep Learning,
and value learning.
The input data in Unsupervised learning uses The data is not predefined in
supervised learning is unlabelled data. reinforcement learning.
labelled data.
Learns by using labelled Trained using unlabelled data Works on interacting with the
data. without any guidance. environment.
Maps the labelled inputs to Understands patterns and Follows the trial-and-error
the known outputs. discovers the output. method.

18. Explain the learning based models of Deep learning.

 Deep learning models are a powerful subset of machine learning, fundamentally built upon the
concept of Artificial Neural Networks (ANNs). The "deep" aspect refers to the use of multiple
processing layers within these networks, enabling them to learn intricate, hierarchical
representations from data.

1. Neural Networks (Artificial Neural Networks - ANNs)

 Definition: Computational networks that are a fundamental component of deep learning

algorithms. They are a subfield of AI and a subset of machine learning.
 Inspiration: Designed to mimic the structure and information processing of the human brain.
 Learning Mechanism:
o Like the human brain, ANNs learn from experience (examples).
o Initially trained on large amounts of input data.
o Recognize patterns within this data using machine learning techniques.
o Once trained, they can make predictions on new datasets.
 Advantages:
o Fast and efficient for problems with very large datasets (e.g., images, videos).
o Autonomous Operation: Once trained, they can independently process new data
and make decisions/predictions without human intervention.

2. Convolutional Neural Network (CNN)

 Definition: An extended version of an Artificial Neural Network (ANN) primarily designed for
image processing.
 Functionality:
o Receives an input image.
o Assigns importance (learnable weights and biases) to various parts of the image.
o Extracts important features.
o This process helps in differentiating one image from another.
 Key Application: Commonly used in image recognition and processing within deep learning.
 Process:
o Analyzes an image.
o Extracts the most relevant features.
o Reduces the image size to make it manageable, crucially while preserving its
important features.

19. What are Neural Networks?

Neural Networks (ANNs) are computational models inspired by the human brain. They learn from examples
and are a core part of deep learning. Initially trained on large datasets, neural networks recognize patterns,
learn using machine learning techniques, and make predictions on new data. They are fast and efficient,
especially for handling large datasets like images and videos. Once trained, they can process new data and
make decisions or predictions without human intervention.

20. Why are Neural Networks Essential?

Neural networks are powerful AI tools that can be effectively used to solve complex real-life
problems. Using algorithms, they can find hidden patterns and relationships in the data, cluster and
classify it, and continuously learn and improve. Neural networks are important for the following
reasons:
 i. A neural network can learn and solve a specific problem very efficiently from the training data
provided to it.

ii. Neural networks can process large datasets and perform tasks, like speech recognition or image
recognition, at an extremely high speed. They can automatically extract important data features
without needing

iii. input from a developer. Neural networks are specialised machine learning algorithms designed to
perform specific tasks quickly and efficiently with high accuracy.

iv. When it comes to large datasets, neural networks perform much better than traditional machine
learning algorithms. Unlike traditional machine-learning algorithms that reach a saturation point
and stop improving, large neural networks show better performance with a large amount of data.

v. When it comes to large datasets, neural networks perform much better than traditional machine
learning algorithms. Unlike traditional machine-learning algorithms that reach a saturation point
and stop improving, large neural networks show better performance with a large amount of data.

21. Give the advantages of Neural Networks.

The following are some of the key advantages of using neural networks:

i. ANNs are fast, efficient, and have powerful parallel processing capacities, enabling them to handle
vast amounts of data and carry out multiple tasks simultaneously.

ii. ANNs can learn and improve from experience, thus enhancing their performance over time.

iii. ANNs can learn and generalise from training data, so there is no need for extensive programming to
instruct the network on how to solve a specific problem.

iv. ANNs are highly fault-tolerant. The data is stored and distributed across the entire neural network
such that even if some nodes of the network are unavailable or get corrupted, the entire network
keeps functioning.

v. ANNs can effectively learn from unorganised, incomplete, complex, and non-linear data and produce
accurate output. This makes them suitable for real-world applications where data may be
incomplete.

vi. ANNs are flexible and can quickly change, adapt, and adjust to new environments and situations.
 22. List Features of Neural Networks.

The following are features of neural networks:

i. Artificial neural networks are extremely powerful computational algorithms or models that
function like the human brain.

ii. Neural network systems are modelled on the structure of and brain and nervous system.

iii. Neural networks can extract patterns from data without any input from the user. Input or
instructions learning algorithm. A neural

iv. Every node in a neural network represents a machine network is effective when solving
problems for which the dataset is very large and complex.

v. It can work with incomplete knowledge and may produce output even with incomplete
information.

23. Explain the working of Neural Networks.

A neural network is based on the model of the human brain and consists of malige layers of
interconnected neurons or blocks called nodes. Each layer performs spеом task and passes
the results to the next layer.

The different layers of the neural networks are listed below:

Input layer: This is the first layer of the neural network. The job of an input layer is to receive
data and pass it on to the rest of the network. No processing occurs at the input layer.

Hidden Layers: These are the next set of layers where all the processing occurs. They are
between the input and output layers. They are called Hidden layers since they are not visible
to the user. There can be multiple hidden layers in a neural network, and their number
depends upon the complexity of the network's function.

Each node in the hidden layers has its own machine learning algorithm that processes the data
received from the input layer. The processed output is then fed to the next hidden layer of the
network until it reaches the last hidden layer. The last hidden layer passes the final processed
data to the output layer.

Output layer: The output layer receives the data from the last hidden layer and is fault-
tolerant to the user. Similar to the input layer, the output layer also does not process the data.
It serves as a user interface, presenting the outcome of the network's computations to the
user.
24. What is CNN

Convolutional Neural Network (CNN) is an extended version of an Artificial Neural Network

that can receive an input image, assign importance to various aspects of the image (in the
form of learnable weights and biases), and extract important features. This helps differentiate
one image from the other.

CNN is a deep learning algorithm commonly used in image recognition and processing. It
analyses an image, extracts the most relevant features, and reduces its size to make it
manageable while still preserving its important features.

25. How does Al make a decision?

Activation functions in neural networks are mathematical functions that help the network learn
complex patterns. They decide whether a neuron should be activated by processing the weighted
sum of inputs plus bias. Inputs pass through layers of neurons, each with weights, and the final
output layer gives a result like a label, probability, or prediction. Activation functions enable neural
networks to recognize patterns, make decisions, and handle complex data effectively, depending on
the network's design and task.