Network Analysis and Graph Machine Learning

2-2 ECTS credits
Academic year 2024-2025
DP00AY84

This course will focus on the representation and statistical analysis of large networks, alongside traditional and deep learning techniques applied to graphs. It will explore various use cases of Graph Machine Learning across diverse domains such as Natural Language Processing, Social Network Analysis, Finance, and Computational Biology.

Education information

Implementation date

19.05.2025 - 23.05.2025

Enrollment period

19.03.2025 - 15.05.2025

Education type

Field-specific studies

Alternativity of education

Optional

Location

Linnanmaa

Enrollment and further information

Enrollment open

Enrollment ends 15.05.2025

Contact Lauri Lovén (lauri.loven@oulu.fi) for enrollment issues if there are any.

Education description

Contact teaching hours = 10 hour (Given by Prof. Joydeep Chandra)
Contact exercise hours = 7.5 hour (coding sessions managed jointly by Prof. Joydeep Chandra and Shruti Saxena)
Independent self-studies = 10 hour (Slides and Reference materials to be provided beforehand, which should be looked into before the class)
Lab work and assignments = 26.5 hour

Many real-world systems can be modeled as networks of interconnected entities, where the number of entities can be vast, forming large-scale networks. Examples include knowledge graph entities, keyword co-occurrence graphs in natural language, user interaction graphs in social networks, protein-protein interaction networks, Internet router networks, IoT edge devices, and financial transaction networks. Analyzing these networks is essential for relational learning tasks and creating frameworks that capture the intrinsic structure of data. This course will primarily cover methods for studying the properties of such large networks and applying advanced machine-learning techniques to these networks for various downstream applications.

By the end of the course, participants are expected to achieve the following outcomes:

Represent and analyze graph structures effectively.
Identify significant and recurring patterns in data using network motifs and community structures.
Understand and apply traditional machine learning techniques to graph data.
Utilize graph-structured data to enhance predictive capabilities through graph neural networks.

Course Contents:

The course will cover the following topics:

Fundamentals of graphs and essential background knowledge
Characterization of graph properties, including measures and metrics
Basics of spectral graph analysis and its applications, such as graph partitioning and community detection
Overview of machine learning applications on graphs, including shallow and deep learning techniques for generating node, link, and graph embeddings
Use cases of graph neural networks across various domains

Target group

Human Sciences Doctoral Programme

The Doctoral Degree Program in Economics and Business Administration
The Doctoral Degree Program in Educational Sciences and Psychology
The Doctoral Degree Program in Geography
The Doctoral Degree Program in Human Sciences

Information Technology and Electrical Engineering Doctoral programme

The Doctoral Degree Program in Communications
The Doctoral Degree Program in Electrical Engineering
The Doctoral Degree Program in Information Processing Science
The Doctoral Degree Program in Computer Science and Engineering

Technology and Natural Sciences

The Doctoral Degree Program in Architecture
The Doctoral Degree Program in Geosciences and Mining Engineering and Mineral Processing
The Doctoral Degree Program in Industrial Engneering and Management
The Doctoral Degree Program in Mathematical Sciences
The Doctoral Degree Program in Mechanical Engineering
The Doctoral Degree Program in Physics and Astronomy
The Doctoral Degree Program in Process and Environmental Engineering
The Doctoral Degree Program in Biology
The Doctoral Degree Program in Chemistry
The Doctoral Degree Program in Civil Engineering
The Doctoral Degree Program in Geography

Other target groups

An advanced graduate course primarily intended for doctoral researchers, but master students with the pre-requisite background may also join. Interested senior university researchers (postdocs, lecturers, and professors) may also audit the course. Suitable for audience in ITEE, Faculty of Technology and Oulu Business School.

Programme

The proposed program is a one-week course tentatively scheduled for May 2025, consisting of five teaching sessions. Each session will include a 2-hour lecture delivered exclusively by Prof. Joydeep Chandra, followed by a 1.5-hour hands-on lab session. During the lab sessions, students will solve programming problems that apply the concepts covered in the lectures.

Teaching Session 1 (Day 1):

Introduction and background knowledge of graphs, their applicability to different problems in diverse domains
Network analysis metrics like paths, components, degree distribution, clustering, degree correlations,
Node centrality measures and their applications in different network problems

Teaching Session 2 (Day 2):

Introduction to Spectral analysis of graphs
Graph Laplacian and the use of Laplacian Eigen values
Use of graph spectra for partitioning and community detection;

Teaching Session 3 (Day 3):

Overview of machine learning applications on graphs
Feature extraction of nodes, edges and graphs
Understanding traditional machine learning techniques using these features for downstream tasks

Teaching Session 4 (Day 4):

Understanding concepts of node, edge and graph embeddings
Shallow and deep learning techniques for generating embeddings
Introduction to Graph Neural Networks

Teaching Session 5 (Day 5):

Using attention in Graph Neural Networks
Understanding the concepts of graph transformers
Graph Neural Network applications in diverse domains like biology, natural language processing, social network, cybersecurity, transport

Lab Schedule:

Lab Session 1 (Day 1):

Setting up of the coding environment for graph processing and visualization
Use of networkX library in python for graph creation and processing
Loading graphs and use of networkX library for finding different characteristics of the network
Use of networkX library for graph visualization

Lab Session 2 (Day 2):

Implementing spectral analysis of graphs, by finding the graph Laplacian and its corresponding eigen values and eigen vectors
Implementing graph partitioning
Implementing different community detection techniques
Performance evaluation

Lab Session 3 (Day 3):

Implementing techniques for extraction of structural features of nodes and edges (graphlets, graphlet degree vectors)
Implementing machine learning techniques for different node, edge and graph level classification tasks
Generate the performance scores (Accuracy, Precision, Recall, F1) of each of these machine learning techniques

Lab Session 4 (Day 4):

Implementing shallow learning techniques for generating node embeddings
Implementing deep learning techniques like node2vec, GraphSage, etc., for generating node embeddings
Using node embeddings for downstream classification tasks
Implementing TSNE plots for visualizations based on the generated node embeddings

Lab Session 5 (Day 5):

Learning how to implement an end-to-end machine learning pipeline on a toy problem
Implement data pre-processing techniques
Implementing vanilla Graph Neural Networks, Graph Attention Mechanism, and Graph Transformer for different node classification tasks.
Performance evaluation and visualization