Deep Learning
Program B. Tech (Computer Science Engineering) Semester-V
Course Title Deep Learning Course Code
Category DSE
Scheme and L T P Credits Exam Duration:
Credits 3 2 3L+1P=4 03 Hrs
Class Work Theory Exam Lab Work Lab Exam Total
Marks Distribution
15 60 10 15 75+25 = 100
Course Objectives:
The course enables students to:
To understand the fundamentals of deep learning and neural networks for solving complex
computational problems.
To learn various deep learning architectures including CNNs, RNNs, and autoencoders for
different application domains.
To gain practical experience in training and deploying deep learning models using modern
frameworks.
Course Outcomes:
The students will be able to:
Design, implement, and train deep learning models using TensorFlow, PyTorch, and Keras for
various applications.
Apply transfer learning and fine-tuning techniques to leverage pre-trained models effectively.
Evaluate deep learning models using appropriate metrics and techniques to ensure
generalization and robustness.
Unit 1: Introduction & Fundamental
Introduction to Deep Learning, definition and scope, difference between deep learning and
traditional machine learning, applications of deep learning, deep learning frameworks, basics of
neural networks, artificial neuron model, perceptron learning rule, neuron model and activation
functions (ReLU, Sigmoid, Tanh), network architecture (input, hidden and output layers), forward
propagation and backward propagation.
Scheme & Syllabus for B. Tech – CSE 2023-27 as per NEP 2020.
Unit 2: Training Deep Neural Networks
Loss functions for regression and classification, gradient descent and its variants (batch, stochastic,
mini-batch), advanced optimization algorithms (Adam, RMSprop, AdaGrad), learning rate scheduling,
overfitting in deep neural networks, regularization techniques (L1, L2, dropout, batch normalization),
weight initialization techniques, hyperparameter tuning and model selection methods.
Unit 3: Convolutional Neural Networks (CNNs)
Convolution operation and feature extraction, kernels and feature maps, stride and padding,
pooling layers (max pooling, average pooling), normalization in CNNs, popular CNN architectures
(LeNet, AlexNet, VGG, ResNet), CNN applications, transfer learning and fine-tuning of pre-trained
models.
Unit 4: Recurrent Neural Networks and Advanced Architectures
Recurrent Neural Networks (RNNs) for sequence modeling, backpropagation through time, vanishing
and exploding gradient problems, Long Short-Term Memory (LSTM) networks, Gated Recurrent Units
(GRU), autoencoders and unsupervised learning, attention mechanisms, transformer basics and
sequence-to-sequence models.
Suggested and Reference Books:
Deep Learning" by Ian Goodfellow, Yoshua Bengio, and Aaron Courville
Deep Learning with Python" by François Chollet
Neural Networks and Deep Learning" by Michael Nielsen
"Dive into Deep Learning" by Aston Zhang, Zachary C. Lipton, and others
Attention is All You Need" papers and resources on Transformers
Scheme & Syllabus for B. Tech – CSE 2023-27 as per NEP 2020.
DEEP LEARNING – LAB
Lab Objectives:
• Master the implementation of deep learning models.
• Develop practical skills in building, training, and evaluating neural networks.
• Apply various deep learning architectures to real-world problems.
• Understand hyperparameter tuning and model optimization techniques.
Lab Contents:
1. Neural Network Fundamentals and Activation Functions
2. Building Feedforward Neural Networks with Keras
3. Forward and Backward Propagation Implementation
4. Loss Functions and Optimization Algorithms
5. Regularization Techniques: Dropout, Batch Normalization, L1/L2
6. Convolutional Neural Networks for Image Classification
7. Popular CNN Architectures: VGG, ResNet, Transfer Learning
8. Recurrent Neural Networks and LSTM for Sequence Modeling
9. Autoencoders and Unsupervised Learning
Mini Projects
• MNIST Digit Classification using Feedforward Neural Networks
• CIFAR-10 Image Classification using CNNs
• ImageNet Classification with Transfer Learning (ResNet, VGG)
• Time Series Forecasting using LSTM and GRU
• Sentiment Analysis using Recurrent Neural Networks
• Building a Character-Level Language Model using RNN
Deliverables:
1. Lab Report documenting each task with network architecture diagrams, training curves, and
performance metrics.
2. Project Source Code with a README file describing data preparation, model architecture
choices, and hyperparameter settings.
3. Final project demonstration with visualizations of predictions, loss convergence, and model
interpretability analysis.
Scheme & Syllabus for B. Tech – CSE 2023-27 as per NEP 2020.