4AIML2041: Artificial Neural Networks

A. Course Framework
Credits:3-0-1-4 Syllabus Version: 1
Contact Hours / Week: 5 Total Contact Hours: 75 Level: 100
Prerequisite: Basic knowledge of linear algebra, calculus, probability theory, and
(If applicable) programming (e.g., Python) is recommended.
Course Learning Objectives:

This course enables the student to:

CLO1: Demonstrate a comprehensive understanding of the mathematical foundations underlying artificial neural
networks.
CLO2: Design, implement, and train feedforward neural networks for tasks such as classification and regression.
CLO3: Explore advanced neural network architectures including convolutional neural networks (CNNs) and recurrent
neural networks (RNNs) for specialized tasks like image recognition and sequence modeling.
CLO4: Apply optimization techniques and regularization methods to improve the training and generalization
performance of neural networks.
CLO5: Utilize popular deep learning frameworks to develop and deploy neural network models efficiently.

Course Outcomes:
On successful completion of the course, students will be able to:
CO1: Understand the fundamental principles of artificial neural networks (ANN) and their applications in various
domains
CO2: Analyze and design basic and advanced neural network architectures for solving complex
problems.
CO3: Implement and train neural networks using appropriate algorithms and tools.
CO4: Evaluate and compare different neural network models in terms of performance and scalability.
CO5: Apply neural network techniques to real-world problems and demonstrate proficiency in solving
them.
PO: PO1/PO2/PO3/PO5 PSO:
PSO1
B. Syllabus

Module 1: Introduction to Artificial Neural Networks L [9] h

Introduction to neural networks: biological inspiration and historical perspective, Perceptron learning algorithm,
Activation functions and their properties, Single-layer perceptron and its limitations, Multilayer perceptron (MLP)
architecture and training algorithms
Module 2: Advanced Neural Network Architectures L [9] h

Convolutional Neural Networks (CNNs): architecture, convolutional layers, pooling layers. Recurrent Neural Networks
(RNNs): structure, backpropagation through time (BPTT), long short-term memory (LSTM), and gated recurrent units
(GRUs). Autoencoders and their applications in unsupervised learning and dimensionality reduction. Generative
Adversarial Networks (GANs) and their role in generating synthetic data.
Module 3:Optimization and Regularization Techniques L [9] h

Gradient descent optimization and its variants: stochastic gradient descent (SGD), mini-batch gradient descent, Adam
optimizer. Weight initialization methods and their impact on training.
Dropout regularization for reducing overfitting. Batch normalization and its effect on training stability.
Module 4: Deep Learning Frameworks L [9] h

Introduction to popular deep learning frameworks: TensorFlow, Keras, PyTorch. Building neural network models using
high-level APIs. Training and evaluation of models with real-world datasets. Deployment of trained models for
inference
on different platforms.
Module 5: Applications of Artificial Neural Networks L [9] h

Image classification and object detection using CNNs. Sequence modeling tasks such as language translation and
sentiment analysis with RNNs. Reinforcement learning and its integration with neural networks for game playing and
robotics. Case studies showcasing the application of neural networks in fields like healthcare, finance, and natural
language processing.

SYLLABUS–LABORATORY P [30] h
1. Implementing a perceptron learning algorithm for binary classification.
2. Training a multilayer perceptron (MLP) for handwritten digit recognition using the MNIST dataset.
3. Building a convolutional neural network (CNN) for image classification using the CIFAR-10 dataset.
4. Designing a recurrent neural network (RNN) for sentiment analysis on movie reviews.
5. Implementing an autoencoder for dimensionality reduction on a high-dimensional dataset.
6. Training a generative adversarial network (GAN) to generate realistic images of handwritten digits.
7. Fine-tuning a pre-trained deep learning model for transfer learning on a specific task.
8. Deploying a trained neural network model using a web-based interface or a mobile application.
C. References
TEXT BOOKS:

1. "Deep Learning" by Ian Goodfellow, Yoshua Bengio, and Aaron Courville.

2. "Neural Networks and Deep Learning: A Textbook" by Charu C. Aggarwal.

REFERENCES:

1. "Pattern Recognition and Machine Learning" by Christopher M. Bishop.

2. "Introduction to the Theory of Neural Computation" by John H. Bruno and Tomaso Poggio.

D. Mode of Assessment

IAT / CCE / SEE

E. Scheme of Evaluation

1. Continuous Internal Evaluation (CIE-Theory) : 50 Marks [scaled down to 25]

Components Average of 2 IATs CCE Total Marks

Max. Marks 20 30 50

2. Continuous Internal Evaluation (CIE-Practical’s) : 100 Marks [scaled down to 25]

Record
Components Experiments Viva-Voce Total
Submission
Max. Marks 50 30 20 100
0. Semester
End
Examination (SEE) Scheme (Theory): 100 Marks [scaled down to 50] Question paper
pattern:
1. The question paper shall have FOUR main questions corresponding to the FOUR
modules. Internal choices shall be given only in the main questions.
2. Each main question will have TWO full questions carrying TWENTY FIVE marks each.
3. A full question may have a maximum of FOUR sub questions, covering the topics
under the module.
4. The students will have to answer all FOUR main questions, selecting ONE full question
out of the two full questions from each module.
5. A minimum of 12.5/25 is required in each CIE-Theory and CIE-Laboratory in order to be
eligible to appear for the SEE.
 Break-up of Marks for Final Evaluation

CIE–Theory CIE–Laboratory SEE Total Marks

25 25 50 100

F. CO-PO-PSO Mapping

CO-PO-PSO Mapping
CO PO PSO
PO1 PO2 PO3 PO4 PO5 PO6 PO7 PO8 PO9 PO10 PO11 PO12 PSO PSO PSO
1 2 3
CO1 * * * *
CO2 * * * * *
CO3 * * * * * * *
CO4 * * * * * *
CO5 * * * *