SEMESTER-V

COURSE 14B: Deep Learning Fundamentals

Theory Credits: 3 3 hrs/week

Course Objectives:
1. To understand the theoretical foundations and mathematics underlying deep learning
algorithms.
2. To explore neural network architectures and their optimization techniques.
3. To apply convolutional and recurrent neural networks to real-world data.
4. To understand the role of autoencoders, generative models, and transfer learning.
5. To develop practical deep learning models using open-source frameworks such as
TensorFlow or PyTorch.

Course Outcomes:
After successful completion of this course, students will be able to:
1. Explain the fundamental principles of neural networks and deep learning.
2. Implement and train deep neural networks for various data modalities.
3. Analyze and optimize network performance using gradient-based methods.
4. Apply advanced architectures such as CNNs, RNNs, and GANs to solve complex
problems.
5. Design, evaluate, and deploy end-to-end deep learning systems for practical applications.

Unit I: Introduction to Deep Learning and Neural Networks

Introduction to AI and Machine Learning – Motivation for Deep Learning – Biological Neurons
and Artificial Neurons – Perceptrons and Multilayer Perceptrons – Activation Functions – Loss
Functions – Backpropagation and Gradient Descent
Unit II: Optimization and Training Deep Networks
Challenges in Training Deep Networks – Vanishing and Exploding Gradients – Regularization
(Dropout, Batch Normalization) – Weight Initialization – Optimizers (SGD, Adam, RMSProp) –
Hyperparameter Tuning and Model Evaluation
Unit III: Convolutional Neural Networks (CNNs)
Convolution and Pooling Operations – CNN Architectures (LeNet, AlexNet, VGG, ResNet,
Inception) – Transfer Learning – Visualization and Interpretation – Applications in Image
Recognition and Computer Vision
Unit IV: Sequential Models and Recurrent Neural Networks (RNNs)
Sequence Modeling – RNNs and LSTMs – GRUs – Attention Mechanisms – Transformers
Overview – Applications in Natural Language Processing (Text, Speech)
Unit V: Generative and Advanced Deep Learning Models
Autoencoders – Variational Autoencoders (VAEs) – Generative Adversarial Networks (GANs) –
Reinforcement Learning Basics – Ethical and Societal Implications of Deep Learning

Text Books:

1. Fundamentals of Deep Learning (2nd Edition) by Nikhil Buduma, Nithin Buduma & Joe
Papa, O’Reilly Media
2. Programming Neural Networks with Python” by Joachim Steinwender and Roland
Schwaiger (Rheinwerk Computing).
 SEMESTER-V
COURSE 14B: Deep Learning Fundamentals
Practical Credits: 1 2 hrs/week

No. Practical Title Objective Key Concepts / Tools

Implement a Perceptron from Understand the basic neuron NumPy, dot product, step
1
Scratch model and activation functions function
Train a Simple Neural Learn supervised learning and Feedforward, weight
2
Network for AND/OR Gates error correction updates, loss function
Handwritten Digit Implement your first deep Keras/TensorFlow,
3
Recognition using MNIST neural network softmax, dense layers
Visualize Activation Compare sigmoid, tanh, and Matplotlib visualization,
4
Functions ReLU behaviors activation study
Build a Multi-Layer
Introduce hidden layers and Dense networks, batch
5 Perceptron (MLP) for
backpropagation training
Classification
Implement Gradient Descent Learn the optimization concept Mean squared error,
6
from Scratch manually learning rate tuning
Image Classification with
Learn CNN architecture and Conv2D, MaxPooling,
7 Convolutional Neural
filters Flatten layers
Networks
Train a Neural Network with Prevent overfitting and improve Dropout layers,
8
Dropout Regularization generalization training/testing accuracy
Predict House Prices Using a Apply deep learning to Normalization, linear
9
Regression Neural Network regression problems output layer
Implement a Basic
Understand neural networks in Q-learning, reward
10 Reinforcement Learning
decision-making function, exploration
Agent