INDUSTRIAL INTERNSHIP AT FRENZO
by
NANTHINI N
727622BAD051
Report submitted in partial fulfillment of the
requirements for the
Degree of Bachelor of Technology
in
ARTIFICIAL INTELLIGENCE AND DATA SCIENCE
Dr. Mahalingam College of Engineering and Technology
Pollachi-642 003
An Autonomous Institution
Affiliated to Anna University, Chennai – 600 025
JUNE 2024
i
CERTIFICATE:
ii
Dr. Mahalingam College of Engineering and Technology
Pollachi – 642003
BONAFIDE CERTIFICATE
Certified that this Industrial Internship Training Report titled
“INDUSTRIAL INTERNSHIP AT FRENZO” is the bonafide work of Miss.
NANTHINI N who carried out the training work under my supervision. Certified
further, that to the best of my knowledge the work reported here in does not form
part of any other project report or dissertation on the basis of which a degree or
award was conferred on an earlier occasion on this or any other candidate.
Mr. M. Vijayakumar Dr. J. Ramprasath
Internship Coordinator Head of the Department
Assistant Professor Associate Professor
Artificial Intelligence and Data Science Artificial Intelligence and Data Science
Dr. Mahalingam College of Engineering Dr. Mahalingam College of Engineering
and Technology Pollachi -642003. and Technology Pollachi -642003.
Submitted for the Internship Viva-Voce examination held on .
Examiner 1 Examiner 2
iii
Dr. Mahalingam College of Engineering and Technology
Pollachi – 642003
DECLARATION
I affirm that the Industrial Internship Training report titled INDUSTRIAL
INTERNSHIP TRAINING AT FRENZO. Being submitted in partial
fulfilment of the requirements for the award of the Degree of BACHELOR
OF TECHNOLOGY IN ARTIFICIAL INTELLIGENS AND DATA
SCIENCE is the original work carried out by me. It has not formed the part
of any other project work submitted for award of any degree or diploma,
either in this or any other institution.
NANTHINI N
(727622BAD051)
I certify that the declaration made above by the candidate is true.
Mr. M. VijayaKumar
Assistant Professor
Internship Coordinator
Department of Artificial Intelligence & Data Science
iv
ACKNOWLEDGEMENT
I would like to express my deep and heartfelt gratitude to my institution and
department for providing me with the opportunity to undertake this internship,
which has allowed me to fulfil my long-cherished dream of gaining practical
experience as an Artificial Intelligence and Data Science Engineer.
I express my sincere thanks to our honorable Secretary Dr. C. Ramaswamy for
providing us with required amenities.
I would like to express my heartfelt appreciation to Dr. P. Govindasamy, the
Principal of our college, for his constant motivation and encouragement
throughout my internship journey.
I am grateful to Dr. J. Ramprasath, Head of the Department in charge, Artificial
Intelligence and Data Science, for her invaluable guidance and direction, which
played a pivotal role in shaping my internship experience.
My heartfelt thanks to my internship guide, Mr. M. Vijayakumar, Assistant
Professor, for her unwavering support and guidance during the course of my
internship. His expertise and mentorship, along with the support of Mr.
P. Gugan Kumar has been instrumental in my professional growth.
v
ABSTRACT
This report presents the internship experience at Frenzo , a renowned Tech
Academy in Coimbatore. This report on the one-month internship provides a
thorough summary of the work done between June 24, 2024, and July 06, 2024. A
variety of machine learning tasks and initiatives were worked on during this time, including
implementing code and solving different problems. Convolutional Neural Network (CNN)
techniques have to be used for the machine learning projects that were chosen at the start of
the internship. Among the difficulties were fixing GPU utilization problems on websites
like Kaggle and changing code to meet project specifications. The emphasis then turned to
choosing image classification projects and the related preprocessing and data collection
activities. We also solved complex 5-class classification tasks by effectively using CNN
models and tackling binary versus multiclass classification conundrums. This extensive
report functions as a thorough documentation of all the actions that took place during the
internship, including project selection, data collecting, model training, fine-tuning, and
problems that were solved by using useful code and internet resources.
NANTHINI N
(727622BAD051)
vi
CHAPTE TITLE PAGE
R NO
NO
List of Abbreviations 1
List of Figures 2
1 Introduction 3
1.1 company Background 4
1.2 Training Objective 4
2 Technical Section 5
2 .1 Machine Learning 5
2. 2 Types of Machine Learning 5
2 .3 Project Selection 6
2 .4 Image Classification and Data Collection 7
2 .5 Data Preprocessing and Model Training 10
2.6 Local System Training and Code Execution 12
2.7 Pretrained models and Model Fine-tuning 14
2 .8 Pretrained Model Challenges and Experimentation 16
2 .9 Documentation and Model Fine- Tuning 18
3 Conclusion and Recommendation 21
4 References 22
vii
LIST OF ABBREVIATIONS
AI Artificial Intelligence
CNN Convolution Neural Network
Colab Google Colaboratory
DL Deep Learning
GPU Graphical processing unit
LSTM Long Short-Term Memory
ML Machine Learning
NLP Natural Language Processing
ReLU Rectified Linear unit
1
LIST OF FIGURES
FIGUR TITLE PAGE NO
E NO
2.1 Machine Learning 5
2.2 Types of Machine Learning 5
2.3 Project Selection 6
2.4 Kaggle 7
2.4.1 CNN Implementation 7
2.4.2 CNN Model Confusion Matrix 8
2.5.1 Model Training 8
2.5.2 Evaluating Trained Model 9
2.6 Evaluating Trained Model 10
2.7 Evaluating Pre-Trained Model 12
2.8.1 Evaluating Pre-Trained Model (VGG16 lr 0.01) 14
2.8.2 Evaluating Pre-Trained Model (MobileNet lr 0.01) 17
2.9 Evaluating Densenet-121 Model 19
2
CHAPTER 1
INTRODUCTION
Frenzo is an initiative towards learning new technology. It is a platform that helps students to turn
their ideas into innovation. Using this platform students can explore various new things in the field
of artificial intelligence, machine learning, data science, and a lot more. Through this platform
students not just learn new things, but they will also be allowed to practically experiment with the
new things that they have learned. In today’s world lot of young minds have great ideas, but due to
a lack of guidance and support, they are not able to implement that idea into a working model. We
help those young talents, provide them support to turn their ideas into innovation. We just do not
stop there we extend our support until they attain a successful position in their career. Our ultimate
aim is to make learning simpler and more interesting for students.
1.1. ABOUT COMPANY
Name: Frenzo
Industry: Deep Learning and Machine Learning
Headquarters: Coimbatore
1.2. COMPANY BACKGROUND:
Innovation in University education is long overdue. It was started by Gugan Kumar CEO &
Founder and with Co-Founder Arun Kumar K. In the 21st Century, Universities can no longer
afford to continue to work with content and forms of education, that are over a century old. Further
the message from Founder as follows, “FRENZO - Tech-Academy seeks to be innovative in
several ways. We view teachers as friends and mentors. We expect students to look up content
online and assimilate it to the extent possible, at home. The classroom is the place for questions,
discussion and dialogue and not for one- way lectures.
Education must also be imparted through experiential learning and exposure of students to ground
realities in the domains they have chosen. I am very keen that the Frenzo- Tech-Academy adopts a
platform that uses information technology to scale education so that equity and access is
significantly enhanced.”
3
1.3. TRAINING OBJECTIVE:
Inspiring free-thinking and individualistic mindsets among students teach them to be game-
changers in their chosen professions and promote the virtues of the scientific method, research, and
scholarly inquiry so students can bring important critical thinking skills to their pursuits outside of
college. As an independent study/online program, Visions in Education utilizes a standards-based
education that addresses individual differences and learning.
1.4. STUDENT WORK ASSIGNMENT:
Over the course of the One-month Machine learning and Deep Learning internship, the worked on a
variety of projects and overcame several obstacles. Selecting a project, obtaining code, and setting
up a daily reporting mechanism on GitHub marked the beginning of the internship. The project of
choice used a convolutional neural network (CNN) to recognize hand data. Early on, there were
difficulties, especially in getting GPU capability enabled for Kaggle model training. Online
resources provided the solution, which made sure the code executed properly. The intern worked on
the Happy or Sad image classification project over the next few days, learning about the
complexities of binary and multiclass classification as well as working with outdated code.
Preprocessing, training models, and data gathering become into important tasks. Pretrained models
like VGG16 were used throughout the internship, with a focus on transfer learning. On a fresh
dataset, produced an amazing accuracy of almost 92%. Advancement proceeded despite obstacles
pertaining to extended model training durations and optimizing hyperparameters.
4
CHAPTER 2
TECHNICAL SECTION
2.1 Machine Learning:
Machine learning is a subfield of artificial intelligence (AI) focused on the development of
algorithms that enable computers to learn from and make predictions or decisions based on data.
Instead of being explicitly programmed to perform a task, machine learning algorithms use
statistical techniques to identify patterns in data and improve their performance over time as they
are exposed to more data.
Figure 2.1 Machine Learning
2.2 Types of Machine Learning:
Figure 2.2 Types of Machine learning
5
2.3 Project Selection:
Project Selection: The first task was to choose a project and determine its title. This decision sets
the direction for the entire internship.
Algorithms Research: The embarked on a quest to find and understand algorithms related
to Convolutional Neural Networks (CNN). This research lays the groundwork for the
upcoming work.
Code Acquisition: To avoid reinventing the wheel, the searched for existing code related
to the chosen project from online sources. This not only saves time but also provides
valuable insights into implementation.
Code Execution in XAMPP: XAMPP, with its GPU support, was chosen as the platform for
running the code. The successfully executed the code to get a feel for the project and assess
its accuracy.
GitHub Repository: Creating a GitHub repository is a crucial step for version control and
documentation. It was set up to post daily reports, ensuring all work is systematically
recorded.
Challenges & Solutions:
GPU Enablement: Initially, there were challenges in enabling GPU support on Kaggle. A
solution was found online, which allowed for GPU utilization.
Data Handling: Running code in Colab had some disadvantages, such as handling data
uploads and processing. However, the noted that Kaggle simplified this process.
Figure 2.3 Project Selection
6
2.4 Image Classification and Data Collection:
Project Selection: This project involved choosing a specific image classification task. Data
Collection: Data was sourced from various online sources. The quality and quantity of data
are critical for model performance.
Data Organization: Data was structured into Train, Validation, and Test sets. Proper data
organization is vital for model training and evaluation.
Google Colab Usage: Then utilized Google Colab with GPU support for code execution, a
standard practice for training deep learning models.
Challenges & Solutions:
Code Modification: Challenges arose when importing the keras.preprocessing.image library.
Then found a solution by changing the import statement to keras.utils as image.
Graph Understanding: Initially, had difficulty understanding the purpose of the plotted
graphs. However, with time and multiple information sources, the graphs' significance
became clear.
Kaggle is the world’s largest data science community with powerful tools and resources to
help you achieve your data science goals.
Figure 2.4 Kaggle
7
Code after implementing solution.
Figure 2.4.1 CNN Implementation
8
Figure 2.4.2 Graph plotting accuracy and loss
Confusion Matrix for the model
Figure 2.4.3 CNN Model – Confusion Matrix
2.5 Data Preprocessing and Model Training:
9
Data Preparation: Data was uploaded and preprocessing techniques applied. The task was
more complex due to the five-class classification problem.
Validation Data Usage: Validation 2 data was utilized for testing the model's performance.
Confusion Matrix: Then plotted a confusion matrix to assess model performance and
identify areas that need improvement.
Overfitting Mitigation: To address early overfitting issues, dropout was introduced as a
regularization technique. An external source was referenced to learn about dropout
Challenges & Solutions:
Early Overfitting: Initially, the model achieved an accuracy of 1, indicating overfitting.
The solution involved introducing dropout to prevent overfitting.
10
Algorithm overfitting
Figure 2.5.1 Model Training
After adding dropout
11
Confusion matrix
Figure 2.5.2 Evaluating Trained Model
2.6 Local System Training & Code Execution:
Code Execution: Ran the code for 20 epochs and plotted training and accuracy matrices.
Local System Usage: Code execution was performed on a local system using
PyCharm. Output Variations: Then observed variations in output between Colab and
the local system, primarily attributed to network performance.
Package Installation: Learning to install packages differently in Colab and PyCharm
was an important part of the project. Tools Used:
PyCharm is an integrated development environment (IDE) used for programming in
Python. It provides code analysis, a graphical debugger, an integrated unit tester,
integration with version control systems, and supports web development with Django.
PyCharm is developed by the Czech company JetBrains.
It is cross-platform, working on Microsoft Windows, macOS and Linux. PyCharm has a
Professional
Edition, released under a proprietary license and a Community Edition released under the
Apache License. PyCharm Community Edition is less extensive than the Professional
Edition.
Challenges & Solutions:
Output Variations: Variations in output between Colab and the local system were noted.
These were attributed to differences in network performance.
12
13
Confusion matrix with class names
Graph plotting accuracy and loss
Figure 2.6 Evaluating Trained Model
14
2.7 Pretrained models and Model Fine-tuning:
New Dataset: A new dataset was downloaded and explored. Understanding the
dataset is crucial for the success of the project.
VGG16 Pretrained Model: Then utilized a VGG16 pretrained model with a learning
rate of 0.01 for training.
Parameter Adjustments: Parameters such as learning rate and number of epochs
were adjusted to achieve a balanced trade-off between training time and accuracy.
Challenges & Solutions:
Training Time: Training time extended due to the large dataset, and GPU limitations
were noted. The solution involved using a higher learning rate.
Trained model model (VGG 16 with 0.01 learning rate)
Figure 2.7 Evaluating Pre-Trained Model
15
Confusion matrix with class names
Graph plotting accuracy and loss
Figure 2.7 Evaluating Pre-Trained Model
16
2.8 Pretrained Model Challenges and Experimentation:
Experimentation: We continued experimenting with pretrained models,
including MobileNet
Accuracy Challenges: Lower accuracy results were achieved, and faced challenges
related to long training times and memory consumption.
Continued experimentation with models like MobileNet, ResNet101, and others.
Continuation of challenges and lower accuracy results.
Challenges & Solutions:
Long training times and memory consumption challenges persisted, and solutions
were sought to improve model performance.
VGG16 Trained model model (0.01 lr)
Figure 2.8.1 Evaluating Pre-Trained Model (VGG16 lr 0.01)
17
Confusion matrix with class names
Figure 2.8.1 Evaluating Pre-Trained Model (VGG16 lr 0.01)
Graph plotting accuracy and loss for VGG 16 (0.1 lr)
Figure 2.8.1 Evaluating Pre-Trained Model (VGG16 lr 0.01)
18
2.9 Documentation and Model Fine-Tuning:
Worked on the Happy or Sad image classification project, using both basic CNN and
VGG19 models. • Achieved a test accuracy of 0.81 with the basic CNN model.
Challenges & Solutions:
The basic CNN yielded lower accuracy and required fine-tuning. Early stopping was
introduced to halt training when accuracy didn't improve. Also gained knowledge
about image recognition and annotation tools during this project.
Densenet-121 Algorithm
Running Epochs
Figure 2.9 Evaluating Densenet-121
19
Graph for validation and testing accuracy
Confusion Matrix
Figure 2.9 Evaluating Densenet-121
20
Classification Report
Figure 2.9 Evaluating Densenet-121
21
CHAPTER 3
CONCLUSION AND RECOMMENDATION
In my quest to become a machine learning and deep learning expert, Frenzo Tech Academy
has been a life-changing event. My development has been greatly aided by the academy's
concentrated approach to imparting practical instruction and real-world skills.
The knowledgeable instructors at Frenzo Tech Academy have been a great resource for advice
and mentoring. Their knowledge has not only improved my practical knowledge but also
given me a thorough understanding of the complexities of deep learning and machine
learning. Effective and efficient problem-solving in the actual world requires
this understanding.
Even though my internship at Frenzo Tech Academy was brief, it was a concentrated learning
opportunity that was insightful and illuminating. I'm excited about the prospect of a longer
internship in the future. I'm sure that a longer-term involvement would provide me even more
insight into the field, and I think it would be quite beneficial to the advancement of my
knowledge and abilities.
Frenzo Tech Academy has been a significant stepping stone in my journey to master machine
learning and deep learning, and I am excited to continue learning and growing in this dynamic
field.
CHAPTER 4
22
REFERENCES
[1] Dataset https://www.kaggle.com/
[2] Online tool https://colab.research.google.com/
[3] Pretrained Models: https://www.tensorflow.org/tutorials/images/transfer_learning
23
