TUTORIAL – HALF DAY SESSION
GRAPH NEURAL NETWORKS: AN INTRODUCTION TO THEORY, APPLICATIONS, AND IMPLEMENTATION

Overview : Graphs are an essential structure for representing complex relationships in various domains such as social networks, molecular graphs, recommendation systems, knowledge graphs, and more. Traditional machine learning algorithms struggle to handle graph data due to its irregular structure and non-grid format. Graph Neural Networks (GNNs) have emerged as powerful deep learning models for processing graph-structured data. They overcome these limitations by learning from the structure of graphs and the node/edge features, allowing them to outperform traditional methods in many graph-related tasks, including:

• Node classification: Predicting the label of a node based on its features and the graph structure.
• Link prediction: Predicting missing or potential connections between nodes.
• Graph classification: Classifying entire graphs

With the increasing availability of graph-structured data across multiple fields, understanding GNNs has become crucial for pushing forward innovations in areas such as drug discovery, fraud detection, recommendation systems, and more. This tutorial will cover both the theoretical foundation of GNNs and provide a hands-on experience where participants can implement and experiment with GNNs using popular Python libraries.

Learning Objectives : The tutorial will help participants

• Understand the fundamental concepts behind Graph Neural Networks, including GNN model architecture.
• Explore the various applications of GNNs across different domains.
• Learn how to implement and train GNNs on a real-world dataset.

Duration : Three hours

Topics to be Covered (120 minutes):

• Limitations of Traditional Neural Networks with Graph Data
• What are Graph Neural Networks?
• GNN Architecture: Message Passing and Node Aggregation
• Popular GNN architectures – GCN, GraphSage, GAT
• Applications of GNNs

Hands-On Implementation (60 minutes):

• Data preprocessing: Converting graphs into suitable formats (e.g., node features, edge lists).
• Implementing a Graph Convolutional Network (GCN) to solve a node classification task (e.g., predicting the category of users in a social network).
• Evaluating the model performance and understanding common metrics used in graph-based tasks

Target Audience : This tutorial is aimed at researchers, data scientists, engineers, and students with a basic understanding of deep learning concepts. As graph-structured data becomes more prevalent across various industries, this tutorial will equip attendees with the knowledge and tools to apply Graph Neural Networks (GNN) to solve real-world problems involving graph-structured data.

Pre-requisites : Participants should have a basic knowledge of deep learning (e.g., neural networks, training algorithms). Prior experience with Python programming is recommended.

Instructor :

Dr. Preety Singh is an Associate Professor in the Computer Science and Engineering Department at The LNM Institute of Information Technology, Jaipur. She did her B.E. from MNREC (currently, NIT Allahabad) and holds an M.Tech. degree from IIT Delhi. She received her PhD degree from MNIT Jaipur in 2013. She has many research papers in peer-reviewed books, journal and international conferences and has been reviewer of premier image processing conferences. Her current research interests are multimedia processing, deep learning, pattern recognition, adversarial machine learning and social media analysis