Scribd
Upload
14 views

CS1

Uploaded by

Mukesh Saini
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
14 views

CS1

Uploaded by

Mukesh Saini
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 26

Graph Neural Networks

(AIMLCZG514)
BITS Pilani
Pilani Campus
BITS Pilani
Pilani Campus

Module I
Course Introduction

COURSE OBJECTIVES:
CO1: To learn how to extract meaningful information
from graphs using Graph Neural Network.
CO2: To learn how to apply neural network on graph
CO3: To learn how apply deep learning models on
graph data
CO4: To learn how to use Graph Neural Network for
specific applications such as web, social networks,
Natural Language Processing, computer vision, Drug
development, etc.

BITS Pilani, Pilani Campus


Resources

TEXT BOOKS (T1, T2 & T3):


 (T1) Graph Neural Networks: Foundations, Frontiers, and
Applications, Editors Lingfei Wu, Peng Cui, Jian Pei,
Liang Zhao, Springer, 2022.: DOI https://
doi.org/10.1007/978-981-16-6054-2
 (T2) William L. Hamilton. (2020). Graph Representation
Learning. Synthesis Lectures on Artificial Intelligence
and Machine Learning, Vol. 14, No. 3 , Pages 1-159.
 (T3) Ma, Y., & Tang, J. (2021). Deep Learning on Graphs.
Cambridge: Cambridge University Press.
doi:10.1017/9781108924184

BITS Pilani, Pilani Campus


Resources

TEXT BOOKS (T1, T2 & T3):


 Not all topics are covered in one text/reference book.

* Research Papers will be used for most of the topics.


* Most of the contents are theoretical, but their practical
application will be explained.
* Research Paper will also be provided to you, for your
better understanding

BITS Pilani, Pilani Campus


Resources

REFERENCE BOOKS (R):

 (R1) NEURAL NETWORK FUNDAMENTALS WITH


GRAPHS, ALGORITHMS, AND APPLICATIONS
McGraw-Hill, Inc. CONTENTS _ John Fiamazo -
Academia.edu.
 (R2) Social Media Mining - An introduction, Reza
Zafarani, Mohammad Ali Abbasi, Huan Liu ,
Cambridge University Press , 2014 [
http://dmml.asu.edu/smm/SMM.pdf - Open Access].

BITS Pilani, Pilani Campus


EVALUATION SCHEME
EC Component &
Duration Weightage Date, Time
No Nature

1. Quiz1 15 Minutes TBA


15%
1. Quiz2 15 Minutes TBA

Assignment1
1. 15 Days TBA
15%
1 Assignment2 15 Days TBA

Mid-Semester Test Saturday, 20 July, 2024


2. 2 hours 30%
(Closed Book) 1:00 PM to 3:00 PM - Afternoon

Comprehensive
Saturday, 28 September, 2024
3. Exam 3 hours 40% 1:00 PM to 3:30 PM - Afternoon
(Open Book)
BITS Pilani, Pilani Campus
Few Requests

 Email etiquettes.
 No “reply to all”; unless & until unavoidable!
 Continuous reading of papers & books.

BITS Pilani, Pilani Campus


Course Outline

 Basics of graph theory


Preliminaries
 Basics for Big Graph Data
 Representation Learning for graphs
Node & Graph Embedding
 Machine learning on graphs
Label propagation for node classification
GNN framework
Attention network
GCN
Link Prediction

BITS Pilani, Pilani Campus


Course Outline

 Graph RNN & Algebraic Neural Networks


 Deep Generative Models
 Weisfeiler -Lehman Algorithm & Stability Properties of
GNN
 Applications of graph NN in subgraph mining,
 Recommendation systems,
 Knowledge graph embedding and reasoning
 Community structures in networks
 Some more advance topics (if time permits)

BITS Pilani, Pilani Campus


What are Graphs?
Graph: Set of vertices and edges.

BITS Pilani, Pilani Campus


What do we study graphs?

 Graphs provide universal representation of data. In other


words, data in any domain can be structured as graphs.
Eg. social networks, transportation networks, protein-
protein interactions, etc.
 Many real world problems can be formulated as
computational tasks in graphs.
Eg. inferring node attributes, detecting anomalous
nodes (node classification), predicting relationships
between nodes (link prediction), etc.

BITS Pilani, Pilani Campus


Various type of Graph Data

BITS Pilani, Pilani Campus


Origins of Graph Theory

BITS Pilani, Pilani Campus


BITS Pilani
Pilani Campus

Basics of graph theory


Basics of graph theory

 Graph Terminologies.
Nodes
Edges
Degree and Degree distribution
Simple Graph & Multiple Graph
Weighted - unweighted,
Directed & Undirected Graph
Storage of graph: Matrix-list..
Tree, Forest, Spanning tree
Traversal: DFS, BFS
Paths, Connected – disconnected

BITS Pilani, Pilani Campus


Basics of graph theory

 HandShaking Theorem
 Havel Hakimi Theorem For Degree Sequence of a Graph
 Complement of a Graph in Graph Theory Discrete
Mathematics
 Complete Graph
 Euler Graph/Hamiltonian Graphs
 Isomorphism in Graph Theory
 Planer / Non-Planer Graph
 Edge CutSet/Vertex CutSet (maxflow - mincut)
 Bipartite graph

BITS Pilani, Pilani Campus


Basics of graph theory

 NP Complete Problems.
 Graph Matching.
Edit Distance
 Sub Graph Isomorphism.
 Graph Summarization
 Graph Sampling
 Graph Partitioning
Cut Vertex
Cut Edge
Bipartite graph

BITS Pilani, Pilani Campus


Basics of graph theory ….

 Random Walk
 Cliques
 Pattern Matching.
 Disk Resident Graphs.
Why the algorithms using random edges/nodes are
not suitable for disk resident graphs
 Connected Components
Largest Connected Components
 Lowest common ancestor.
 ……

BITS Pilani, Pilani Campus


Types of Graph Algorithms

 The Graph algorithms are broadly classified in two


different cases:
Multiple graphs
Chemical data
Biological data
Single Large Graph
Web Graph
Social media graphs

BITS Pilani, Pilani Campus


Various type of Graph Data

Search “image.Google” for various types of large graphs

BITS Pilani, Pilani Campus


Various type of Graph Data

 Chemical Data: Often represented as graphs in which


the nodes correspond to atoms, and the links correspond
to bonds between the atoms.
In some cases, substructures of the data may also be
used as individual nodes.
Such as Benzene rings
Polymers

The individual graphs are quite small, though there are


significant repetitions among the different nodes. This
leads to isomorphism challenges in applications such
as graph matching.
BITS Pilani, Pilani Campus
Various type of Graph Data

 Biological Data: Similar way as chemical data. However, the


individual graphs are typically much larger. Furthermore, the nodes
are typically carefully designed portions of the biological models
 Such as Proteins
Amino acids
A, T, G,C in DNA
A, U, G, C in RNA
Dominance and recessive property of Genes
Mendel's experiment

A single biological network could easily contain thousands of nodes.


The sizes of the overall database are also large enough for the
underlying graphs to be disk-resident.

BITS Pilani, Pilani Campus


Various type of Graph Data

 Computer Networked and Web Data: The number of nodes in the


underlying graph may be massive as well as edges. (massive
domain issue in networked data.)
it may be hard to hold in the available storage space.
So we need to be designed some techniques to
summarize and work with condensed representations
of the graph data sets.
Google maps….
The edges may also arrive in the form of a data stream.
A new challenge arises, it may not be possible to store
the incoming edges for future analysis.
Therefore, the summarization techniques and the
stream summaries may be leveraged for future
processing of the underlying graphs.

BITS Pilani, Pilani Campus


Various type of Graph Data

 XML data :Complex and semi-structured data is often


represented in the form of XML documents
 XML data is a natural form of labelled graph data.
 Mining and management algorithms for XML data are
also quite useful for graph data,
 In addition, the attribute-value combinations
associated with the nodes makes the problem much
more challenging.
Since it is much more difficult to design mining
and management operations for structured data.

BITS Pilani, Pilani Campus


Real Time Scenarios with Graph
Learning
1) It is assumed that the underlying graphs are massive
and cannot be held in main memory.
2) In many cases, the problem of graph management and
Learning has been studied from the perspective of
structured and XML data.
3) Many of the graph algorithms are sensitive to the
application scenario in which they are encountered.
4) Different application result in different kind of graphs
and corresponding challenges are also quite different.

BITS Pilani, Pilani Campus

You might also like