lOMoARcPSD|21343390

AD8701 DL Unit5 Notes

deep learning (Anna University)

Scan to open on Studocu

Studocu is not sponsored or endorsed by any college or university

Downloaded by Shri Harini ([email protected])
 lOMoARcPSD|21343390

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Please read this disclaimer before proceeding:

This document is confidential and intended solely for the educational purpose of
RMK Group of Educational Institutions. If you have received this document through
email in error, please notify the system manager. This document contains proprietary
information and is intended only to the respective group / learning community as
intended. If you are not the addressee you should not disseminate, distribute or
copy through e-mail. Please notify the sender immediately by e-mail if you have
received this document by mistake and delete this document from your system. If
you are not the intended recipient you are notified that disclosing, copying,
distributing or taking any action in reliance on the contents of this information is
strictly prohibited.

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

AD8701

DEEP LEARNING

Department : AI & DS
Batch/Year : 2020- 2024 / IV

Created by : Ms.A.AKILA

Date : 29.09.2023

Signature :

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

1. TABLE OF CONTENTS

S NO CONTENTS SLIDE NO

1 Contents 5

2 Course Objectives 7
9
3 Pre Requisites (Course Names with Code)
4 Syllabus (With Subject Code, Name, LTPC details) 11

5 Course Outcomes 12

6 CO- PO/PSO Mapping 15

7 Lecture Plan 17

8 Activity Based Learning 19-20

9 Lecture Notes 20-44

10 Assignments 46
11 Part A (Q & A) 48
12 Part B Qs 54

13 Supportive Online Certification Courses 56

Real time Applications in day to day life and to
14 57
Industry

15 Contents Beyond the Syllabus 59

16 Assessment Schedule 61

17 Prescribed Text Books & Reference Books 63

18 Mini Project Suggestion 66

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Course Objectives

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

2. COURSE OBJECTIVES

 To understand the basics of deep neural networks

 To understand CNN of architectures of deep neural networks

 To understand the concepts of Artificial Neural Networks

 To learn the basics of Data science in Deep learning

 To learn about applications of deep learning in AI and Data Science

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

PRE REQUISITES

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

3. PRE REQUISITES

PRE-REQUISITE CHART

AD8701 – DEEP LEARNING

MA8391 Probability
and Statistics

AD8402 Artificial
Intelligence I

AD8552 Machine
Learning

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Syllabus

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

4. SYLLABUS

UNIT I DEEP NETWORKS BASICS 9

Linear Algebra: Scalars -- Vectors -- Matrices and tensors; Probability Distributions --
Gradientbased Optimization – Machine Learning Basics: Capacity -- Overfitting and
underfitting -- Hyperparameters and validation sets -- Estimators -- Bias and variance --
Stochastic gradient descent -- Challenges motivating deep learning; Deep Networks: Deep
feed forward networks; Regularization -- Optimization.

UNIT II CONVOLUTIONAL NEURAL NETWORKS 9

Convolution Operation -- Sparse Interactions -- Parameter Sharing -- Equivariance -- Pooling
-- Convolution Variants: Strided -- Tiled -- Transposed and dilated convolutions; CNN
Learning: Nonlinearity Functions -- Loss Functions -- Regularization -- Optimizers -- Gradient
Computation.

UNIT III DEEP LEARNING ALGORITHMS FOR AI 9

Artificail Neural Netowrks – Linear Associative Networks – Perceptrons -The
Backpropagation Algorithm - Hopfield Nets - Boltzmann Machines - Deep RBMs - Variational
Autoencoders – Deep Backprop Networks- Autoencoders

UNIT IV DATA SCIENCE AND DEEP LEARNING 9

Data science fundamentals and responsibilities of a data scientist - life cycle of data science
– Data science tools - Data modeling, and featurization - How to work with data variables
and data science tools - How to visualize the data - How to work with machine learning
algorithms and Artificial Neural Networks

UNIT V APPLICATIONS OF DEEP LEARNING 9

Detection in chest X-ray images -object detection and classification -RGB and depth image
fusion - NLP tasks - dimensionality estimation - time series forecasting -building electric
power grid for controllable energy resources - guiding charities in maximizing donations and
robotic control in industrial environments.

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Course Outcomes

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

4. COURSE OUTCOME

Cognitive/
Affective
Course
Course Outcome Statement Level of the
Code Course
Outcome
CO1 Explain the basics in deep neural networks Understand
K2

Apply Convolution Neural Network for image Apply

CO2
processing K3
Explain the basics of Artificial Intelligence Understand
CO3 using deep learning K2
CO4 Apply deep learning algorithms for data Apply
science
K3

CO5 Apply deep learning algorithms for variety Apply

applications
K3

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

CO – PO/PSO Mapping

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

5. CO-PO/PSO MAPPING

Correlation Matrix of the Course Outcomes to

Programme Outcomes and Programme Specific

Outcomes Including Course Enrichment Activities

Programme Outcomes (POs), Programme Specific Outcomes (PSOs)

P P P P P P P P P P P P PS PS PS
Course O O O O O O O O O O O O O O O
Outcomes 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3
(Cos)
K K K3 A
K4 K5 /K A2 A3 A3 A3 A3 A2 K3 K3 K3
3 5 3
5

CO1 K
2 2 1 2 3 3 3

CO2 K
3 2 1 2 3 3 3

CO3 K
2 3 2 1 1 3 3 3 3 3

K
5
CO4
3 3 3 2 2 3 3 3 3 3

CO5 K
3 3 2 1 1 3 3 2 2 2

12

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Lecture Plan
Unit V

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

LECTURE PLAN – UNIT V

UNIT IV – DATA SCIENCE AND DEEP LEARNING

Sl PROPOSED
. ACTUAL HIGH
LECTURE MODE
N Total No LECTURE PERTA- EST
OF
o of INING COGN
TOPIC DELIVE
Periods CO(s) ITIVE
Date RY
Date LEVEL

Detection in chest X-
1 ray images -object CO5 K3
1 PPT
detection and
classification 20.10.2023
CO5 K3
RGB and depth image 1 PPT
2 fusion - NLP tasks 20.10.2023
dimensionality CO5 K3
1
estimation - time series PPT
3 forecasting 21.10.2023
CO5 K3
dimensionality
estimation - time series PPT
4 forecasting 1 02.11.2023
building electric power CO5 K3
grid for controllable PPT
1
5 energy resources 03.11.2023
building electric power CO5 K3
6 grid for controllable PPT
1
energy resources 03.11.2023
guiding charities in CO5 K3
maximizing donations 1

7 and robotic control in PPT

industrial
environments. 04.11.2023
guiding charities in CO5 K3
maximizing donations 1

8 and robotic control in PPT

industrial
environments. 06.11.2023
guiding charities in CO5 K3
9 maximizing donations 1
and robotic control in PPT
industrial 14
environments. 09.11.2023

17

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

ACTIVITY BASED LEARNING - UNIT I

COMPLETE THE PUZZLE GIVEN BELOW

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Hint to solve the Puzzle

ACROSS
1. step in the data science life cycle where hypotheses are formed and data is explored.
3. step in data modeling where relevant features are selected or created.
4. used for data visualization and exploration.
5. for defining the project scope, objectives, and key performance indicators (KPIs).
6. person responsible for extracting insights from data and making data-driven decisions.
7. phase where insights are communicated to stakeholders and used for decision-making.
8. energy-based modeling approach used in Boltzmann Machines.

DOWN

2. for designing experiments, collecting and analyzing data, and drawing conclusions.
3. for converting categorical data into a numerical format for modeling.
4. used for data storage, retrieval, and preprocessing.
6. final step in the data science life cycle where the model is deployed and monitored.
7. phase where data is collected, prepared, and made ready for analysis.

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Lecture Notes
UNIT V

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

UNIT V APPLICATIONS OF DEEP LEARNING

Detection in chest X-ray images -object detection and

classification -RGB and depth image fusion - NLP tasks -
dimensionality estimation - time series forecasting -building
electric power grid for controllable energy resources - guiding
charities in maximizing donations and robotic control in industrial
environments.

1. Detection in chest X-ray images

Detection in Chest X-ray Images

Chest X-ray image analysis holds a significant position within the field of
medical imaging. These images serve as invaluable diagnostic tools for
identifying a wide range of respiratory and cardiac conditions.

Importance of Chest X-ray Image Analysis

One of the primary reasons for the importance of chest X-ray image analysis
is its potential to facilitate early detection of abnormalities. This early
detection can lead to timely treatment interventions, ultimately improving
patient outcomes. Additionally, automated analysis of chest X-rays allows
radiologists to efficiently review a large volume of patient data, increasing
the efficiency of healthcare delivery.

Challenges in Detecting Abnormalities

Several challenges are associated with the task of detecting abnormalities

in chest X-ray images. Firstly, the anatomy depicted in these images is
complex, encompassing bones, soft tissues, and air-filled spaces, which can
make interpretation challenging. Secondly, the appearance of abnormalities
can exhibit considerable variability, making it difficult for traditional methods
to achieve high accuracy. Finally, a significant class imbalance often exists
in the data, with abnormal cases being much rarer than normal cases,
leading to imbalanced datasets that can affect model performance.

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Chest X-Ray Images

• Chest X-ray images play a crucial role in medical diagnostics, particularly for
detecting respiratory abnormalities.

• Convolutional Neural Networks (CNNs) are commonly employed to analyze

these images, enabling automated identification of conditions such as
pneumonia, tumors, and other respiratory issues.

STEPS INVOLVED IN THE DETECTION OF CHEST X-RAY IMAGES:

1. Data Collection and Preprocessing:

 Start by gathering a large dataset of chest X-ray images. The

dataset should be diverse and cover a range of conditions,
including normal and abnormal cases.

 Preprocess the images to ensure consistency and improve the

model's performance. This may include resizing, normalization,
and data augmentation.

2. Labeling:

 Annotate the images with labels indicating the presence or

absence of abnormalities. For chest X-rays, common abnormalities
include pneumonia, tumors, or other respiratory issues.

3. Model Architecture
 Convolutional Neural Networks (CNNs) are a popular choice for
analyzing chest X-ray images due to their effectiveness in image-
related tasks.
 It consists of multiple convolutional layers with rectified linear unit
(ReLU) activation functions, followed by max-pooling layers to
reduce the spatial dimensions of the feature maps.
 The output of the convolutional layers is then flattened and fed
into a fully connected layer with a ReLU activation function.

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

4. Convolutional Layers
• The core building blocks of a CNN are convolutional layers.
• These layers use filters (kernels) to scan the input image for specific
patterns or features.
• In the context of chest X-rays, these patterns could be indicative of
abnormalities like nodules, opacities, or other relevant structures.

5. Pooling Layers
 Pooling layers follow convolutional layers and are used to reduce the
spatial dimensions of the input volume.
 Max pooling, for example, retains the most important information
from a group of pixels, helping to decrease the computational load
and focus on the most relevant features

6. Activation Functions
 Non-linear activation functions like ReLU (Rectified Linear Unit) are
commonly used to introduce non-linearity into the model.
 This allows the CNN to learn complex patterns and relationships
within the data.

7. Flattening and Fully Connected Layers:

 After several convolutional and pooling layers, the extracted
features are flattened and passed through fully connected layers.
 These layers connect every neuron in one layer to every neuron
in the next layer, allowing the model to make predictions based
on the learned features.

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

8. Output Layer
 The output layer typically consists of one or more neurons,
depending on the specific classification task.

 For chest X-ray analysis, it might be a binary classification

(normal or abnormal) or multi-class classification (categorizing
different abnormalities).

9. Evaluation:
• Evaluate the trained model on the test set to assess its generalization
performance.
• Metrics such as accuracy, precision, recall, and F1 score are commonly
used to quantify performance.
10.Deployment:
• Once satisfied with the model's performance, deploy it for real-world
use.
• This could involve integrating it into a healthcare system to assist
radiologists in detecting abnormalities in chest X-rays.

Below Block Diagram Shows the Diagrammatic representation of Lung Cancer

Prediction

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Model Architecture

 It is a Sequential model which will contain the following parts:

 Three Convolutional Layers followed by MaxPooling Layers.
 The Flatten layer to flatten the output of the convolutional layer.
 Then we will have two fully connected layers followed by the output of
the flattened layer.
 We have included some BatchNormalization layers to enable stable
and fast training and a Dropout layer before the final layer to avoid any
possibility of overfitting.
 The final layer is the output layer which outputs soft probabilities for the
three classes.

2. Object detection and classification

Object Detection: Locate the presence of objects with a bounding box and
types or classes of the located objects in an image.

 Input: An image with one or more objects, such as a photograph.

 Output: One or more bounding boxes (e.g. defined by a point, width,
and height), and a class label for each bounding box.

Object detection: Algorithms produce a list of object categories present in the

image along with an axis-aligned bounding box indicating the position and scale
of every instance of each object category.

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

R-CNN

C. NN model is comprised of three modules; they are:

 Module 1: Region Proposal. Generate and extract category independent

region proposals, e.g. candidate bounding boxes.

 Module 2: Feature Extractor. Extract feature from each candidate region,

e.g. using a deep convolutional neural network.
 Module 3: Classifier. Classify features as one of the known class, e.g. linear
SVM classifier model.

The architecture of the model is summarized in the image below

 The feature extractor used by the model was the AlexNet deep CNN.
 Training is a multi-stage pipeline. Involves the preparation and operation
of three separate models.

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

 Training is expensive in space and time. Training a deep CNN on so many

region proposals per image is very slow.

 Object detection is slow. Make predictions using a deep CNN on so

many region proposals is very slow.
 The architecture of the model takes the photograph a set of region proposals
as input that are passed through a deep convolutional neural network. A pre-
trained CNN, such as a VGG-16, is used for feature extraction. The end of the
deep CNN is a custom layer called a Region of Interest Pooling Layer, or RoI
Pooling, that extracts features specific for a given input candidate region.

 The output of the CNN is then interpreted by a fully connected layer then the
model bifurcates into two outputs, one for the class prediction via a softmax
layer, and another with a linear output for the bounding box. This process is
then repeated multiple times for each region of interest in a given image.

Object Detection

Object detection is a critical component of chest X-ray image analysis,

enabling the localization of abnormalities within the images. Deep learning
models, such as YOLO (You Only Look Once) and Faster R-CNN, have been
successfully employed for this purpose. YOLO is known for its real-time
performance, while Faster R-CNN offers high accuracy.

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

3. RGB and depth image fusion

Introduction

RGB and depth image fusion is a technique that combines information from
conventional RGB images and depth maps. This fusion process enhances the
perception and understanding of the environment, enabling various applications in
robotics, augmented reality, and computer vision.

Importance of RGB and Depth Fusion

The fusion of RGB and depth information allows machines to perceive the world
more like humans. It enhances the ability to understand object shapes, distances,
and their interactions with the environment. This has significant implications for
fields such as robotics and augmented reality.

Applications

Robotics

RGB and depth fusion is vital for robotic perception. Robots equipped with depth
sensors can navigate and interact with their surroundings more effectively. They
can identify obstacles, manipulate objects, and plan paths with greater precision.

Augmented Reality

In augmented reality applications, combining RGB and depth data allows for realistic
object placement and interaction with the real world. This technology is used in
games, simulations, and various interactive experiences.

Fusion Techniques

Depth Sensor Technologies

Depth sensors like Microsoft Kinect and LiDAR systems capture depth information using
structured light or time-of-flight principles. These sensors provide accurate depth maps
that complement RGB images.

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Deep Learning Approaches for Fusion

Deep learning models, particularly convolutional neural networks (CNNs), are

employed to fuse RGB and depth data. These models learn to extract meaningful
features from both modalities and combine them for improved understanding.

Use Cases

Gesture Recognition

RGB and depth fusion is pivotal in gesture recognition systems. By analyzing both RGB
and depth data, computers can accurately detect and interpret hand gestures. This is
used in gaming consoles, human-computer interaction, and sign language recognition.

Object Recognition and Tracking

In computer vision tasks, combining RGB and depth information enhances object
recognition and tracking. It enables algorithms to differentiate between objects with
similar appearances but different 3D shapes, improving the accuracy of tracking and
identification.

In conclusion, RGB and depth image fusion is a powerful technique with diverse
applications in robotics, augmented reality, and computer vision. Advancements in
depth sensing technology and deep learning approaches are poised to drive further
innovation in this field, enabling more accurate and robust fusion systems for a wide
range of applications.

4. NLP tasks

Overview of NLP

Natural Language Processing (NLP) is a field at the intersection of computer science,

artificial intelligence, and linguistics. Its primary objective is to enable computers to
understand, interpret, and generate human language. NLP holds immense importance
in various applications across industries due to its ability to bridge the gap between
human communication and machines.

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

NLP implementation steps

Typically, NLP implementation begins by gathering and preparing unstructured text or

speech data from sources like cloud data warehouses, surveys, emails, or internal
business process applications.

● Pre-processing

The NLP software uses pre-processing techniques such as tokenization, stemming,

lemmatization, and stop word removal to prepare the data for various applications.

➔ Tokenization breaks a sentence into individual units of words or phrases.

➔ Stemming and lemmatization simplify words into their root form. For example, these
processes turn "starting" into "start."
➔ Stop word removal ensures that words that do not add significant meaning to a
sentence, such as "for" and "with," are removed.

● Training

Researchers use the pre-processed data and machine learning to train NLP models to
perform specific applications based on the provided textual information. Training NLP
algorithms requires feeding the software with large data samples to increase the
algorithms' accuracy.

● Deployment and inference

Machine learning experts then deploy the model or integrate it into an existing
production environment. The NLP model receives input and predicts an output for the
specific use case the model's designed for. You can run the NLP application on live
data and obtain the required output.

What are the approaches to natural language processing?

Supervised NLP : Supervised NLP methods train the software with a set of labeled
or known input and output. The program first processes large volumes of known data
and learns how to produce the correct output from any unknown input. For example,
companies train NLP tools to categorize documents according to specific labels.

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Unsupervised NLP : Unsupervised NLP uses a statistical language model to predict

the pattern that occurs when it is fed a non-labeled input. For example, the
autocomplete feature in text messaging suggests relevant words that make sense for

the sentence by monitoring the user's response.

Natural language understanding : Natural language understanding (NLU) is a

subset of NLP that focuses on analyzing the meaning behind sentences. NLU allows
the software to find similar meanings in different sentences or to process words that
have different meanings.

Natural language generation : Natural language generation (NLG) focuses on

producing conversational text like humans do based on specific keywords or topics.
For example, an intelligent chatbot with NLG capabilities can converse with customers
in similar ways tocustomer support personnel.

NLP benefits

● The ability to analyze both structured and unstructured data, such as speech, text
messages, and social media posts.
● Improving customer satisfaction and experience by identifying insights using sentiment
analysis.
● Reducing costs by employing NLP-enabled AI to perform specific tasks, such as
chatting with customers via chatbots or analyzing large amounts of text data.
● Better understanding a target market or brand by conducting NLP analysis on relevant
data like social media posts, focus group surveys, and reviews.

What about ChatGPT?

ChatGPT is a chatbot powered by AI and natural language processing that produces

unusually human-like responses. Recently, it has dominated headlines due to its ability
to produce responses that far outperform what was previously commercially possible.

Why is NLP important?

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Natural language processing (NLP) is critical to fully and efficiently analyze text and
speech data. It can work through the differences in dialects, slang, and grammatical
irregularities typical in day-to-day conversations.

Companies use it for several automated tasks, such as to:

• Process, analyze, and archive large documents
• Analyze customer feedback or call center recordings
• Run chatbots for automated customer service
• Answer who-what-when-where questions
• Classify and extract text

Importance and Applications of NLP

The importance of NLP is evident in its wide-ranging applications. NLP enables

chatbots to provide customer support, helps in sentiment analysis of social media data,
aids in language translation, and powers voice assistants like Siri and Alexa. Moreover,
NLP is pivotal in content recommendation systems, information retrieval, and even
medical text analysis, where it can assist in diagnosing and extracting insights from
clinical reports.

Sentiment Analysis

Sentiment analysis, a subfield of NLP, focuses on determining the sentiment or

emotional tone expressed in textual data. Deep learning models are increasingly being
used for sentiment analysis, as they can capture complex patterns in text data. These
models are trained on labeled datasets containing text samples with associated
sentiment labels, such as positive, negative, or neutral.

Deep Learning Models for Sentiment Analysis

Deep learning models like Convolutional Neural Networks (CNNs) and Recurrent
Neural Networks (RNNs) have demonstrated excellent performance in sentiment
analysis tasks. CNNs can capture local patterns in text, while RNNs can model the
sequential nature of language. This allows them to discern nuances in sentiment,
making them valuable tools for sentiment analysis applications.

Sentiment Lexicons

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Sentiment lexicons are lists of words or phrases categorized based on their

sentiment polarity (e.g., positive, negative, neutral). These lexicons serve as resources
for sentiment analysis tasks, aiding in the identification of sentiment-bearing words
within text data. Lexicons are often used in combination with deep learning models to
enhance sentiment analysis accuracy.

Text Generation

Text generation involves the creation of coherent and contextually relevant text by
machines. Generative models, such as GPT-3 (Generative Pre-trained Transformer 3),
have gained prominence for their ability to generate human-like text. They are
pretrained on vast amounts of text data and can generate text for applications like
chatbots, content creation, and automated storytelling.

Text Generation Applications

Generative models like GPT-3 have been employed in various applications. Chatbots
powered by GPT-3 can engage in more natural and context-aware conversations. In
content creation, these models can assist in generating articles, reports, and creative
writing. They have also been used in automated content summarization and translation
tasks.

Machine Translation

Machine translation is the task of automatically translating text from one language
to another. Neural machine translation models, which are deep learning-based, have
significantly improved translation quality. They work by learning to map input
sentences to their corresponding translations in multiple languages.

Challenges in Multilingual Translation

Multilingual translation poses challenges related to language-specific nuances,

idioms, and variations in sentence structure. Maintaining the fluency and accuracy of
translations across multiple languages remains a complex problem. Additionally,
resource availability for low-resource languages can be limited, hindering translation
quality..

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

5. Dimensionality Estimation

Introduction

Dimensionality estimation is a critical step in data analysis, particularly in handling

high-dimensional datasets. It involves reducing the number of features or variables
while preserving essential information. This process is crucial for various applications,
including pattern recognition and machine learning.

Significance of Dimensionality Estimation in Data Analysis

The significance of dimensionality estimation lies in its ability to simplify complex

datasets. High-dimensional data often suffer from the curse of dimensionality, making
it challenging to analyze and model effectively. By reducing dimensionality, we can
improve model performance, decrease computational requirements, and enhance
interpretability.

Principal Component Analysis (PCA)

PCA is a widely used technique for dimensionality reduction. It works by

transforming the original data into a new set of orthogonal variables, known as
principal components. These components capture the most significant variance in the
data. PCA has been extensively used for feature extraction and visualization in various
domains.

Basics of PCA

PCA begins with mean-centering the data to have a zero mean. It then calculates
the covariance matrix of the features. The eigenvalues and eigenvectors of this matrix
determine the principal components. The components are ranked in descending order
of explained variance.

Using Deep Learning for Dimensionality Reduction

In recent years, deep learning approaches, such as autoencoders, have gained

popularity for dimensionality reduction. Autoencoders are neural networks designed to
map high-dimensional data to lower-dimensional representations. They can capture

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

intricate patterns and dependencies in the data, making them suitable for complex
datasets.

Autoencoders

Autoencoders consist of an encoder and a decoder network. The encoder

compresses the input data into a lower-dimensional representation (encoding), while
the decoder reconstructs the original data from the encoding. The training objective is
to minimize the reconstruction error.

Architecture and Training of Autoencoders

The architecture of an autoencoder depends on the specific problem and data. Deep
autoencoders have multiple hidden layers, allowing them to capture hierarchical
features. Training involves backpropagation and gradient descent techniques to
optimize the encoding and decoding functions.

Applications in Dimensionality Reduction

Autoencoders find applications in various domains, including image compression,

data denoising, and feature learning. They are particularly effective when dealing with
unstructured data, such as images and text.

Anomaly Detection

Detecting Outliers Using Dimensionality Reduction

Dimensionality reduction techniques like PCA and autoencoders can be leveraged

for anomaly detection. Anomalies, or outliers, are data points that significantly deviate
from the expected patterns. Identifying these outliers is crucial in various domains,
including fraud detection and quality control.

Use Cases in Fraud Detection and Quality Control

In fraud detection, dimensionality reduction can help in detecting unusual spending

patterns or account behaviors. By reducing the data to its most informative
components, anomalies become more apparent, leading to improved detection
accuracy.

Quality control processes benefit from anomaly detection as well. In manufacturing,

for instance, dimensionality reduction can highlight deviations in product

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

measurements, allowing for early identification of defects and reducing production

errors.

In summary, dimensionality estimation techniques like PCA and autoencoders play

a vital role in data analysis and anomaly detection. They enable the transformation of
high-dimensional data into lower-dimensional representations, simplifying complex
datasets and facilitating the detection of outliers in various applications.

6. Time Series Forecasting

Time series forecasting is a critical area of data analysis with applications in various
domains. It involves predicting future values based on historical data patterns.

Time Series Data

Time series data is characterized by its temporal nature, where observations are
recorded at successive time points. This type of data is commonly encountered in fields
such as finance (e.g., stock price prediction) and meteorology (e.g., weather
forecasting).

Recurrent Neural Networks (RNNs)

Recurrent Neural Networks (RNNs) are a class of deep learning models specifically
designed for sequential data analysis. In the context of time series forecasting, RNNs
are well-suited to capture temporal dependencies and patterns.

Introduction to RNNs for Time Series Analysis: RNNs are capable of maintaining a
hidden state that captures information from previous time steps, making them suitable
for modeling sequences.

LSTM and GRU Architectures: Long Short-Term Memory (LSTM) and Gated
Recurrent Unit (GRU) are specialized RNN variants that address the vanishing gradient
problem, allowing them to capture long-term dependencies in time series data
effectively.

Forecasting Models

Deep learning-based models have gained popularity for time series forecasting due
to their ability to handle complex patterns. These models leverage neural networks for
predictive tasks.

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Deep Learning-Based Time Series Forecasting Models: Models like the recurrent neural
network (RNN) and its variants, as well as convolutional neural networks (CNNs), are
used to extract features and make predictions.

Hyperparameter Tuning: Fine-tuning model hyperparameters is crucial for achieving

optimal forecasting performance. This includes setting learning rates, batch sizes, and
network architecture parameters.

Evaluation Metrics

Evaluating the accuracy of time series forecasting models is essential to assess their
performance and make improvements.

Metrics for Measuring Forecasting Accuracy: Common metrics include Mean Absolute
Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE). These
quantify the difference between predicted and actual values.

Cross-Validation Techniques: To assess model generalization, cross-validation

methods like k-fold cross-validation are employed. This involves splitting the dataset
into training and validation sets multiple times to evaluate performance.

Real-World Applications

Time series forecasting has practical applications across various industries, enhancing
decision-making and planning.

Examples of Successful Time Series Forecasting in Industry: In finance, accurate

stock price predictions are crucial for investment decisions. In meteorology, weather
forecasting models help prepare for severe weather events. Additionally, demand
forecasting in supply chain management aids in optimizing inventory and logistics.

In summary, time series forecasting is a powerful tool for making predictions based
on historical sequential data. Deep learning models, especially RNNs, have proven
effective in capturing temporal patterns and improving forecasting accuracy in diverse
domains. Evaluation metrics and cross-validation techniques play a vital role in
assessing and refining these models, ensuring their real-world applicability and
reliability.

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

7. Building Electric Power Grid for Controllable Energy Resources

The Need for Smart Grids

Smart grids have emerged as a critical innovation in the field of electric power
distribution. Traditional power grids face significant challenges that hinder their
efficiency and reliability. These challenges include outdated infrastructure, limited
visibility into grid operations, and difficulties in accommodating the growing demand
for electricity. Consequently, there is a pressing need to transition towards smart grids.

Challenges in Traditional Power Grids

Traditional power grids, often characterized by their centralized control systems,

struggle to address modern energy demands. They are vulnerable to failures, lack real-
time monitoring capabilities, and face challenges in integrating renewable energy
sources. These limitations result in inefficiencies, higher costs, and reduced resilience,
particularly in the face of disruptions like extreme weather events.

Benefits of Smart Grids

Smart grids offer a range of benefits that address these shortcomings. They enable
enhanced monitoring and control of the entire electricity network, from power
generation to consumption. This leads to improved reliability, reduced power losses
during transmission, and quicker response to faults. Additionally, smart grids facilitate
the integration of renewable energy sources, enabling a cleaner and more sustainable
energy future.

Deep Learning for Grid Management

Deep learning plays a pivotal role in optimizing the management of smart grids. Its
applications extend to various aspects of grid operations, offering data-driven solutions
to complex challenges.

Demand Prediction Using Neural Networks

Deep learning, particularly neural networks, is employed for accurate demand

prediction. These models analyze historical consumption data, weather patterns, and

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

other relevant factors to forecast electricity demand. Accurate predictions empower

grid operators to optimize energy generation and distribution in real-time, preventing
overloads and minimizing waste.

Grid Optimization with Deep Reinforcement Learning

Deep reinforcement learning (DRL) is harnessed to optimize grid operations.

Through continuous learning from real-time data, DRL algorithms can make decisions
to balance supply and demand efficiently. They adapt to changing conditions,
optimizing the dispatch of energy resources and reducing operational costs.

Integration of Renewable Energy

Smart grids face the challenge of managing fluctuating energy sources, such as
wind and solar power, which are inherently variable. Deep learning algorithms aid in
addressing this challenge.

Managing Fluctuating Energy Sources

Deep learning models analyze data from renewable energy sources and predict their
output. These predictions help grid operators anticipate fluctuations and adjust energy
generation accordingly, ensuring grid stability and reliability even in the presence of
intermittent renewable sources.

Case Studies

To illustrate the practical application of smart grids and deep learning in the energy
sector, several real-world case studies serve as examples.

Examples of Smart Grid Implementations

Numerous regions and utilities worldwide have implemented smart grid solutions.
For instance, cities have adopted advanced metering infrastructure (AMI) to monitor
electricity usage in real-time, and some regions have integrated distributed energy
resources (DERs) seamlessly into their grids. These implementations demonstrate the
potential of smart grids to enhance energy management and sustainability.

In conclusion, the integration of deep learning techniques into smart grids offers
substantial advantages in terms of energy efficiency, reliability, and sustainability. By
addressing the challenges faced by traditional power grids and harnessing the benefits
of smart grids, we can create a more resilient and responsive energy infrastructure for
the future.

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Due to rising electricity demand and finite fossil fuel resources, a transition to
renewable energy is imperative. While renewables can be intermittent, combining
them with other sources can mitigate this issue. We present a hybrid renewable system
integrated with grids for efficient energy prediction using an energy management
system. Our study focuses on optimizing energy production, grid interactions, and
storage, all in tandem.

DEEP LEARNING SOLUTIONS

• PREDICTIVE MODELING

• LOADFORECASTING

• ENERGY MANAGEMENT

• FAULT DETECTION AND MAINTANCE

• GRID CONTROL AND STABILITY

• RENEWABLE ENERGY INTEGRATION

• GRID OPTIMATION

• Deep reinforcement learning (DRL) can be applied to optimize the operation of the
power grid.

• DRL agents can make real-time decisions about the distribution of electricity from
various sources to meet demand while minimizing costs and emissions.

This is especially useful when integrating controllable energy resources like battery
storage and flexible demand

8.Guiding Charities in Maximizing Donations

Charitable Giving and Fundraising

Donations hold immense significance for non-profit organizations, as they provide

the financial support needed to fulfill their missions and make a positive impact on

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

society. However, fundraising for charitable causes presents several challenges. These
challenges include the need to reach potential donors effectively, maintain donor
engagement, and secure sustainable funding sources.

Data-Driven Donor Insights

In the digital age, charities are leveraging data-driven approaches to gain insights
into donor behavior. Predictive modeling techniques analyze historical donation data,
demographic information, and engagement patterns to anticipate how donors might
respond to different appeals and campaigns. This data-driven understanding allows
organizations to tailor their fundraising strategies more effectively.

Targeted Marketing Strategies

Armed with insights from predictive models, charities can implement targeted
marketing strategies. These strategies involve tailoring communication and outreach
efforts to specific donor segments. By delivering personalized messages and content
to donors who are likely to be interested in a particular cause, charities can significantly
enhance their fundraising efficiency.

Recommender Systems

Deep learning, particularly recommender systems, plays a pivotal role in the world
of charitable giving. These systems employ sophisticated algorithms to analyze donor
preferences and recommend causes or campaigns that align with an individual's values
and interests. By leveraging deep learning techniques, charities can provide donors
with highly personalized recommendations, fostering greater engagement and
participation.

Personalization and Engagement

Personalization in charitable giving extends beyond recommendations. It

encompasses crafting individualized donor journeys, recognizing milestones, and
demonstrating the impact of donations. Deep learning helps charities create tailored
experiences that resonate with donors, making them feel more connected to the cause
and the organization's mission.

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Above is the overview of guiding charities in maximizing donations, including the

importance of donations, data-driven donor insights, targeted marketing,
recommender systems, personalization, and ethical considerations.

9.Industrial Automation and Robotics

Industrial automation and robotics have transformed the landscape of

manufacturing and logistics industries. Robots play a pivotal role in these sectors by
carrying out various tasks, such as material handling, assembly, and quality control,
with high precision and efficiency. Their integration has led to increased productivity,
reduced labor costs, and improved product quality. However, harnessing the full
potential of robots requires addressing numerous challenges related to their control
and operation in dynamic industrial environments.

Role of Robots in Manufacturing and Logistics

Robots serve as indispensable assets in modern manufacturing facilities and logistics

operations. They are used for repetitive and labor-intensive tasks, which not only
increases the speed of production but also ensures consistent quality. For instance, in
manufacturing, robots are employed for welding, painting, and even complex assembly
tasks. In logistics, robots are instrumental in tasks like automated material handling,
sorting, and packaging. The role of robots extends beyond routine tasks as they can
adapt to changes in production requirements, making them highly versatile assets.

Challenges in Robotic Control

Despite their benefits, robots face several challenges in industrial environments.

One of the primary challenges is ensuring precise and efficient control. Robots must
navigate complex and dynamic environments, avoiding obstacles and adapting to
unforeseen changes. Additionally, safety is a paramount concern, especially when
robots operate alongside human workers. These challenges require advanced control
mechanisms and intelligent decision-making algorithms to ensure safe and efficient
robot operation.

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Deep Reinforcement Learning (DRL)

Deep Reinforcement Learning (DRL) has emerged as a powerful approach for

addressing the complexities of robotic control. DRL combines deep neural networks
with reinforcement learning principles to enable robots to learn and adapt their
behaviors through trial and error. This approach allows robots to make decisions in
real-time based on sensory input and optimize their actions to achieve specific goals.

DRL for Robot Control

In DRL for robot control, robots are trained using a reward-based system. They
explore different actions and learn which actions lead to favorable outcomes. Over
time, the robot's neural network improves its decision-making abilities, enabling it to
perform tasks with increasing precision and efficiency. DRL has been applied to a wide
range of industrial tasks, from robotic arm control in manufacturing to autonomous
navigation in warehouse environments.

Training Robotic Agents

Training robotic agents using DRL involves creating simulation environments where
robots can learn without the risk of physical damage. The agents receive rewards for
accomplishing tasks and penalties for making mistakes. Through a process of trial and
error, the agents gradually refine their policies and behaviors. Once trained in
simulations, the models can be deployed to physical robots in real-world industrial
settings.

Applications

DRL-powered robotic control has found applications in various industrial contexts.

In robotic assembly lines, DRL helps robots perform intricate assembly tasks with high
precision, reducing the need for human intervention. In autonomous warehouse
operations, robots equipped with DRL capabilities can navigate complex warehouse
layouts, pick and transport items efficiently, and optimize warehouse logistics.

Safety and Human-Robot Collaboration

Ensuring the safety of robotic systems is of utmost importance in industrial

environments. Safety measures include sensor-based collision avoidance, emergency
stop mechanisms, and stringent programming of robot behavior. Moreover,
collaborative robots, or "cobots," are designed to work alongside human workers,

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

offering assistance and augmenting human capabilities while maintaining safety. This
collaborative approach maximizes the benefits of automation while minimizing risks.

The integration of DRL into industrial robotics is poised to revolutionize

manufacturing and logistics by making robots more adaptable, efficient, and safe in
their interactions with the industrial environment and human workers. However, it also
necessitates ongoing research and development to address the evolving challenges in
this rapidly advancing field.

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Assignments

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

ASSIGNMENT :

Data Science Tools

Choose a real-world business scenario (e.g., fraud detection in financial transactions) and

identify three data science tools or libraries that would be most useful for solving this

problem. Provide a brief explanation of each tool's functionality and how it can contribute to

the project's success. Discuss any advantages and limitations of these tools in your chosen

scenario.

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Part A – Q & A
Unit - V

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

PART A

1. What is object detection? CO5,K1

Object detection is a profound computer vision technique that focuses on
identifying and labeling objects within images, videos, and even live footage. Object
detection models are trained with a surplus of annotated visuals in order to carry out
this process with new data. It becomes as simple as feeding input visuals and
receiving a fully marked-up output visual.

2. What is the key component is the object detection? CO5,K1

● Bounding box which identifies the edges of the object tagged with a clear-
cut quadrilateral — typically either a square or rectangle.

● Label of the object, whether it be a person, a car, or a dog to describe the

target object. Bounding boxes can overlap to showcase multiple objects in a
given shot as long as the model has prior knowledge of items it is tagging.

3. What is image classification? CO5,K1

This is the prediction of the class of an item in an image. Image classification
can show that a particular object exists in the image, but it involves one primary

object and does not provide the location of the object within the visual.
4. What is image segmentation? CO5,K1
It is the task of grouping pixels with comparable properties together instead of
bounding boxes to identify objects.
5. What is Object localization? CO5,K1
Object localization seeks to identify the location of one or more objects in an
image, whereas object detection identifies all objects and their borders without much
focus on placement.

6. Types of object detection algorithms and methods used in Deep Learning?

CO5,K1
● R-CNN
● Fast R-CNN
● Faster R-CNN
● YOLO (You Only Look Once)

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

7. What is NLP? CO5,K1

Natural language processing (NLP) deals with building computational algorithms to
automatically analyze and represent human language. NLP is also useful to teach
machines the ability to perform complex natural language related tasks such as
machine translation and dialogue generation.

8. Write the Deep Learning applications i\od NLP? CO5,K1

● Tokenization and Text Classification
● Generating Captions for Images
● Speech Recognition
● Machine Translation
● Question Answering (QA)
● Document Summarization
9. What is time series forecasting? CO5,K1
Time series forecasting is a technique for the prediction of events through a
sequence of time. It predicts future events by analyzing the trends of the past, on
the assumption that future trends will hold similar to historical trends. It is used
across many fields of study in various applications including: Astronomy.

10. Mention the deep learning architectures specialized in time series

forecasting? CO5,K1
● N-BEATS (ElementAI)
● DeepAR (Amazon)
● Spacetimeformer
● Temporal Fusion Transformer or TFT
11. Draw the architecture diagram for N-BEATS? CO5,K1

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

12. What are the advantages of N-BEATS model? CO5,K1

● Expressive and easy to use
● Multiple time-series
● Doubly Residual Stacking
● Interpretability
13. What is the core idea of Spacetimeformer model? CO5,K1
The model would consider both temporal and spatial relationships. This is the
core idea of Spacetimeformer.

14. What is the difference between depth image and RGB image? CO5,K1
An RGB-D image provides a per-pixel depth information aligned with corresponding
image pixels. An image formed through depth information is an image channel in
which each pixel relates to a distance between the image plane and the
corresponding object in the RGB image.

15. What is the depth of RGB image? CO5,K1

The RGB image represents a 24-bit integer value for each pixel with a fixed
resolution of 1/256 of a millimeter. Each bit of blue represents 1/256 millimeter.
Each bit of green represents 1 millimeter. Each bit of red represents 256 millimeters

16. What are RGB images used for? CO5,K1

The main purpose of the RGB color model is for the sensing, representation, and
display of images in electronic systems, such as televisions and computers, though it
has also been used in conventional photography.

17. What is dimensionality estimation? CO5,K1

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Dimensionality estimation in deep learning refers to the process of determining the

effective number of dimensions or features required to represent the data
accurately. It aims to identify the intrinsic complexity or structure of the data that
can be effectively captured by a deep learning model.

18. What are the approaches used for dimensionality estimation in deep
learning? CO5,K1
Principle Component analysis.
Variational autoencoders
Random Projections

Information theory measures

19. List the advantages of DEEP AR model. CO5,K1

Multiple time Series
Rich sets of inputs
Automatic Scaling
Probabilistic forecasting

20. Can deep learning be used in robotics? CO5,K1

The main drive behind the use of deep learning in robotics is that it is more general
than any other learning algorithm. It has been proven deep networks are capable of
thinking and abstraction at a high level

21. How are industrial robots controlled? CO5,K1

The arm has a controller which is the “brain” of the system. The controller holds the
programming code and receives signals from the system (input), processes the
signals, and then sends signals out to the system (output) to control the robot.

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

22. Which algorithm is used in robotics and industrial? CO5,K1

PRM Algorithm

They are used in complex planning systems and also to find low cost paths around
obstacles. PRMs use a random sample of points on their map where a robot device
can possibly move and then the shortest path is calculated.

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Part B – Q & A
Unit - V

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

PART B

1. Explain in detail on how the object is detected and classified using deep learning
concepts. (CO5,K3)
2. Detailed overview on building electric power grid for controllable energy resources
in deep learning concept. (CO5,K3)
3. Explain in detail on the process involved for prediction on historical time
dependent data using neural network. (CO5,K3)

4. Brief note on (CO5,K3)

a) N-Beats
b) Deep AR
c) Space time former
d) Temporal Fusion transformer
5. Explain the role of deep learning in Robotics. (CO5,K3)
6. Build a deep learning model for guiding charities in maximizing donations
(CO5,K3)
7. Build a deep learning model for detecting chest X-ray images using tensor flow.
(CO5,K3)

8. Implement a model for object detection of traffic images using python. (CO5,K3)

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Supportive online
Certification courses
(NPTEL, Swayam,
Coursera, Udemy, etc.,)

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

SUPPORTIVE ONLINE COURSES

https://www.edx.org/learn/linear-algebra
https://online-learning.harvard.edu/course/linear-algebra?delta=0

S Course Course Link

No provider title

1 Deep https://onlinecourses.nptel.ac.in/noc23_cs110/unit?
NPTEL Learning unit=17&assessment=226

2 Deep https://www.coursera.org/specializations/deep-
Learning learning
Coursera Specializa
tion

3 Deep https://www.udemy.com/course/deep-learning-
Learning with-python-and-keras/
with
Coursera Python
and Keras

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Real time Applications in

day to day life and to
Industry

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

REAL TIME APPLICATION - UNIT V

1. Virtual Assistants
Virtual Assistants are cloud-based applications that understand natural language voice commands
and complete tasks for the user. Amazon Alexa, Cortana, Siri, and Google Assistant are typical
examples of virtual assistants. They need internet-connected devices to work with their full
capabilities. Each time a command is fed to the assistant, they tend to provide a better user
experience based on past experiences using Deep Learning algorithms.
2. Chatbots
Chatbots can solve customer problems in seconds. A chatbot is an AI application to chat online
via text or text-to-speech. It is capable of communicating and performing actions similar to a
human. Chatbots are used a lot in customer interaction, marketing on social network sites, and
instant messaging the client. It delivers automated responses to user inputs. It uses machine
learning and deep learning algorithms to generate different types of reactions.
3. Healthcare
Deep Learning has found its application in the Healthcare sector. Computer-aided disease
detection and computer-aided diagnosis have been possible using Deep Learning. It is widely
used for medical research, drug discovery, and diagnosis of life-threatening diseases such as
cancer and diabetic retinopathy through the process of medical imaging.
5. News Aggregation and Fake News Detection
Deep Learning allows you to customize news depending on the readers’ persona. You can
aggregate and filter out news information as per social, geographical, and economic parameters
and the individual preferences of a reader. Neural Networks help develop classifiers that can
detect fake and biased news and remove it from your feed. They also warn you of possible
privacy breaches.
6. Composing Music
A machine can learn the notes, structures, and patterns of music and start producing music
independently. Deep Learning-based generative models such as WaveNet can be used to develop
raw audio. Long Short Term Memory Network helps to generate music automatically. Music21
Python toolkit is used for computer-aided musicology. It allows us to train a system to develop
music by teaching music theory fundamentals, generating music samples, and studying music.
5

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Content Beyond Syllabus

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

CONTENTS BEYOND SYLLABUS :

VARIOUS APPLICATIONS IN DEEP LEARNING

Entertainment (VEVO, Netflix, Film Making, Sports Highlights, etc.)

Wimbledon 2018 used IBM Watson to analyse player emotions and expressions
through hundreds of hours of footage to auto-generate highlights for telecast. This saved them
a ton of effort and cost. Thanks to Deep Learning, they were able to factor in audience response
and match or player popularity to come up with a more accurate model (otherwise it would just
have highlights of the most expressive or aggressive players). Netflix and Amazon are
enhancing their deep learning capabilities to provide a personalized experience to its viewers by
creating their personas factoring in show preferences, time of access, history, etc. to
recommend shows that are of liking to a particular viewer.
VEVO has been using deep learning to create the next generation of data services
for not only personalized experiences for its users and subscribers, but also artists, companies,
record labels, and internal business groups to generate insights based on performance and
popularity. Deep video analysis can save hours of manual effort required for audio/video sync
and its testing, transcriptions, and tagging. Content editing and auto-content creation are now a
reality thanks to Deep Learning and its contribution to face and pattern recognition. Deep
Learning AI is revolutionizing the filmmaking process as cameras learn to study human body
language to imbibe in virtual characters.

Adding Sounds To Silent Movies

An application of both convolutional neural networks and LSTM recurrent neural
networks involves synthesizing sounds to match silent videos. A deep learning model tends
to associate the video frames with a database of pre-recorded sounds to select appropriate
sounds for the scene. This task is done using training 1000 videos – that have drum sticks
sound striking on different surfaces and creating different sounds. These videos are then used
by Deep learning models to predict the best suited sound in the video. And later to predict if the
sound is fake or real, a Turing-test like setup is built to achieve the best results.

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Assessment Schedule
(Proposed Date & Actual
Date)

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

ASSESSMENT SCHEDULE

Tentative schedule for the assessment during

2023-2024 Odd Semester

Name of the
S. No. Start Date Portion
Assessment

1 IAT 1 11.09.2023 UNIT 1 & 2

2 IAT 2 26.10.2023 UNIT 3 & 4

3 MODEL 15.11.2023 ALL 5 UNITS

79 6

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Prescribed Text Books &

Reference

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

PRESCRIBED TEXT BOOKS AND REFERENCE BOOKS

TEXT BOOKS:

1.Ian Goodfellow, Yoshua Bengio, Aaron Courville, ``Deep Learning'', MIT Press, 2016

2.Stone, James. (2019). Artificial Intelligence Engines: A Tutorial Introduction to the

Mathematics of Deep Learning, Sebtel Press, United States, 2019

3.Vance, William , Data Science: A Comprehensive Beginners Guide to Learn the Realms

of Data Science (Hardcover - 2020), Joiningthedotstv Limited

4.Wani, M.A., Raj, B., Luo, F., Dou, D. (Eds.), Deep Learning Applications, Volume 3,Springer

Publications 2022

5.Charu C. Aggarwal, ``Neural Networks and Deep Learning: A Textbook'', Springer

International Punlishing, 2018.

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Mini Project Suggestions

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

18. MINI PROJECT SUGGESTION

Mini Projects

a) Recommendation system
b) Credit Card Fraud Detection
c) Fake News Detection
d) Customer Segmentation
e) Sentiment Analysis
f) Recommender Systems
g) Emotion Recognition
h) Stock Market Prediction
i) Email classification
j) Tweets classification
k) Uber Data Analysis
l) Social Network Analysis

Downloaded by Shri Harini ([email protected])

 lOMoARcPSD|21343390

Thank you

Disclaimer:

This document is confidential and intended solely for the educational purpose of
RMK Group of Educational Institutions. If you have received this document through
email in error, please notify the system manager. This document contains proprietary
information and is intended only to the respective group / learning community as
intended. If you are not the addressee you should not disseminate, distribute or
copy through e-mail. Please notify the sender immediately by e-mail if you have
received this document by mistake and delete this document from your system. If
you are not the intended recipient you are notified that disclosing, copying,
distributing or taking any action in reliance on the contents of this information is
strictly prohibited.

82 6

Downloaded by Shri Harini ([email protected])