Fall 2026: Machine Learning with Graphs – CS 7332 (NU Course Info) & NETS 7332 (NU Course Info)

 

General Information

Time: Mondays & Wednesdays 2:50 – 4:30 PM Eastern

Place: 101 Belvidere Street, 3rd Floor, Rooms 140 A & B. (directions)

Instructor: Tina Eliassi-Rad

Course website on Canvas: https://northeastern.instructure.com/courses/261008

Office hours: Available by appointment. Email teliassirad [at] ccs [dot] northeastern [dot] edu to setup an appointment; begin the subject line with [fa26 nets].

TA: Samantha “Sam” Dies. Available by appointment. Email dies [dot] s [at]
northeastern [dot] edu to setup an appointment; begin the subject line with [fa26 nets].

 

Overview

This 4-credit PhD-level course covers state-of-the-art research on data mining and machine learning with graphs. Topics include, but are not limited to, vertex classification, graph clustering, link prediction and analysis, graph distances, graph embedding and network representation learning, deep learning on graphs, anomaly detection on graphs, graph summarization, network inference, adversarial learning on networks, notions of fairness in social networks, and generative AI methods on graphs.

 

Prerequisites

Students are expected to have taken courses on or have knowledge of the following:

o   Calculus and linear algebra

o   Basic statistics, probability, machine learning, or data mining

o   Algorithms and programming skills (e.g., Python, Julia, C, C++, Java, Ruby, Matlab, or any programming language of their preference)

 

Textbooks

This course does not have a designated textbook. The readings are assigned in the syllabus (see below).

 

Here are some textbooks (all optional) on machine learning and data mining:

 

Resources

 

Grading

o   Class presentations (50%)

o   Each paper will have a lead discussant (LD), an advocate (A), and a critic (C).

Ø  The lead discussant presents the paper (about 20-25 minutes with slides) and moderates the discussion.

Ø  The advocate makes the strongest case for the paper: its contributions, what it gets right, and why it matters.

Ø  The critic makes the strongest case against it:  methodological weaknesses, unsupported claims, and missed comparisons.

Ø  After the lead’s presentation, the advocate and the critic each speak for about 8-10 minutes, using slides that include a brief summary; strengths (at least four specific points for A and at least two for C); weaknesses (at least four specific points for C and at least two for A); clarifying questions (at least two specific points you are unsure about); and concrete suggestions (at least two). The advocate and the critic should conclude by stating what would change their assessment.

Ø  Everyone else should come with at least one question or comment.

Ø  I got this idea from Byron Wallace.

o   Besides the readings, each paper is likely to have additional materials on the Web. Examples include supplemental materials, video, code, data, etc. These are helpful for presentations and class projects.

o   Slides for class presentations are due at 12:00 PM (noon) Eastern on the day assigned to you.

o   Class project (50%)

o   Breakdown: proposal (10%), report (25%), slides & in-class presentation (15%)

o   I will team up students into groups.

o   Each team will choose (by Wednesday, October 14, 2026 at 11:59 PM Eastern) one of the papers in the syllabus to reproduce.

o   In addition to the reproduction, each team will propose extension(s) to the chosen paper and implement those extension(s).

o   Each team will write a report (maximum 6 pages) detailing what was learned. Use the style files at https://media.neurips.cc/Conferences/NeurIPS2025/Styles.zip.

o   Reports & slides on class projects are due on Friday, December 14, 2026 at 12:00 PM (noon) Eastern.

 

Schedule/Syllabus (Subject to Change)

Date

Lecturer

Readings

Wed Sep 9

Tina Eliassi-Rad

Overview

·      The Why, How, and When of Representations for Complex Systems

Mon Sep 14

Tina Eliassi-Rad

Homophily

·      Distribution of Node Characteristics in Complex Networks

·      Combinatorial Characterizations and Impossibilities for Higher-order Homophily

·      Higher-order Homophily on Simplicial Complexes

·      [optional] Information Access Equality on Generative Models of Complex Networks

·      [optional] Effects of Higher-order Interactions and Homophily on Information Access Inequality

Axiomatic Approaches

·      Measuring Tie Strength in Implicit Social Networks 

Wed Sep 16

Tina Eliassi-Rad

Network Comparison and Graph Distances

·      A Guide to Selecting a Network Similarity Method 

·      Network Comparison and the Within-ensemble Graph Distance

·      Non-backtracking Cycles: Length Spectrum Theory and Graph Mining Applications

·      [reference] netrd: A library for Network Reconstruction and Graph Distances

·      [optional] Optimal Transport for Network Comparison: A Review with Machine Learning Applications

Mon Sep 21

Tina Eliassi-Rad

Role Discovery

·      It's Who You Know: Graph Mining Using Recursive Structural Features

·      RolX: Structural Role Extraction & Mining in Large Graphs

·      Guided learning for Role Discovery (GLRD): Framework, Algorithms, and Applications

Wed Sep 23

Tina Eliassi-Rad

Graph Representation Learning: Node Embedding

·      Laplacian Eigenmaps for Dimensionality Reduction and Data Representation

·      node2vec: Scalable Feature Learning for Networks

·      Structural Deep Network Embedding

·      STABLE: Identifying and Mitigating Instability in Embeddings of the Degenerate Core

·      [reference] Machine Learning on Graphs: A Model and Comprehensive Taxonomy

·      [reference] Network Representation Learning: From Preprocessing, Feature Extraction to Node Embedding

·      [optional] DeepWalk: Online Learning of Social Representations

·      [optional] Bypassing Skip-Gram Negative Sampling: Dimension Regularization as a More Efficient Alternative for Graph Embeddings

·      [optional] Next Waves in Veridical Network Embedding

Mon Sep 28

Low-rank Representations of Complex Networks

·      The Impossibility of Low-rank Representations for Triangle-Rich Complex Networks

·      Node Embeddings and Exact Low-rank Representations of Complex Networks

·      [optional] Classic Graph Structural Features Outperform Factorization-Based Graph Embedding Methods on Community Labeling

·      [optional] Link Prediction Using Low-dimensional Node Embeddings: The Measurement Problem

·      [optional] Network Embedding as Matrix Factorization: Unifying DeepWalk, LINE, PTE, and node2vec

Wed Sep 30

Graph Neural Networks I

·      Semi-Supervised Classification with Graph Convolutional Networks

·      Graph Attention Networks

·      Inductive Representation Learning on Large Graphs

·      [reference] Graph Neural Networks: A Review of Methods and Applications

·      [reference] A Comprehensive Survey on Graph Neural Networks

·      [optional] Everything is Connected: Graph Neural Networks

·      [optional] Geometric Deep Learning: the Erlangen Programme of ML (ICLR 2021 keynote)

Mon Oct 5

Graph Neural Networks II

·      Hyperbolic Graph Convolutional Neural Networks

·      Pitfalls of Graph Neural Network Evaluation

·      Design Space for Graph Neural Networks (GitHub page)

·      [optional] The Numerical Stability of Hyperbolic Representation Learning

Wed Oct 7

 

Collective Classification

·      Collective Classification in Network Data

·      Graph Belief Propagation Networks

·      [optional] Cautious Collective Classification

Mon Oct 12

No class (US holiday)

Wed Oct 14

Class project proposals are due at 11:59 PM Eastern.

Label Propagation on Graphs

·      Combining Label Propagation and Simple Models Out-performs Graph Neural Networks

·      Masked Label Prediction: Unified Message Passing Model for Semi-Supervised Classification

·      [optional] Message passing all the way up

Mon Oct 19

GNNs for Recommendation Systems

·      LightGCN: Simplifying and Powering Graph Convolution Network for Recommendation

·      Neural Graph Collaborative Filtering

·      Graph Convolutional Neural Networks for Web-Scale Recommender Systems

·      [reference] Graph Neural Networks in Recommender Systems: A Survey

·      [reference] A Survey of Graph Neural Networks for Recommender Systems: Challenges, Methods, and Directions

Wed Oct 21

Hypergraphs and Higher-order Models with applications to Graph ML for Optimization

·      Assigning Entities to Teams as a Hypergraph Discovery Problem

·      Distributed constrained combinatorial optimization leveraging hypergraph neural networks

·      [reference] A Survey on Hypergraph Mining: Patterns, Tools, and Generators

·      [optional] Random Walks on Hypergraphs with Edge-Dependent Vertex Weights

·      [optional] Hypergraph Neural Networks

·      [optional] Simplicial Attention Networks

Mon Oct 26

Oversmooting and Oversquashing

·      A Survey on Oversmoothing in Graph Neural Networks

·      How does over-squashing affect the power of GNNs?

·      [optional] Neural Sheaf Diffusion: A Topological Perspective on Heterophily and Oversmoothing in GNNs

Wed Oct 28

W-L Graph Kernels and Power of GNNs

·      Weisfeiler-Lehman Graph Kernels

·      How Powerful are Graph Neural Networks?

·      [optional] A Reduction of a Graph to a Canonical Form and an Algebra arising during this Reduction

·      [reference] Theory of Graph Neural Networks: Representation and Learning

Mon Nov 2

Stability and Counting in GNNs

·      Tree Mover’s Distance: Bridging Graph Metrics and Stability of Graph Neural Networks

·      Can Graph Neural Networks Count Substructures?

Wed Nov 4

Invariance and Equivariance

·      E(n) Equivariant Graph Neural Networks

·      Invariant and Equivariant Graph Networks

Mon Nov 9

*-aware GNNs

·      Position-aware Graph Neural Networks

·      Identity-aware Graph Neural Networks

Wed Nov 11

No class (US holiday)

 

Mon Nov 16

This and That

·      PRODIGY: Enabling In-context Learning Over Graphs 

·      Uncertainty Quantification over Graph with Conformalized Graph Neural Networks

·      REGE: A Method for Incorporating Uncertainty in Graph Embeddings

Wed Nov 18

Explainability in GNNs

·      Explainability in Graph Neural Networks: A Taxonomic Survey

·      GNNExplainer: Generating Explanations for Graph Neural Networks

·      GraphFramEx: Towards Systematic Evaluation of Explainability Methods for Graph Neural Networks

·      Trustworthy Graph Neural Networks: Aspects, Methods and Trends

·      [optional] FAIRGEN: Towards Fair Graph Generation

Mon Nov 23

Graph Transformers I

·      Graph Transformer Networks

·      A Generalization of Transformer Networks to Graphs

·      Transformers are Graph Neural Networks

·      [reference] Graph Transformers: A Survey

Wed Nov 25

No class (US holiday)

Mon Nov 30

Graph Transformers II

Wed Dec 2

ML on Heterogeneous Graphs

·      Modeling Relational Data with Graph Convolutional Networks

·      Heterogeneous Graph Transformer

Mon Dec 7

Circuit Tracing in Large Language Models

·      Circuit Tracing: Revealing Computational Graphs in Language Models

·      Tracing Attention Computation Through Feature Interactions

Wed Dec 9

 

TBD

Mon Dec 14

Recent Position Papers

·      Future Directions in the Theory of Graph Machine Learning

·      Position: Graph Learning Will Lose Relevance Due To Poor Benchmarks

·      Why We Must Rethink Empirical Research in Machine Learning

Wed Dec 16

Presentation of class projects

Reports and slides for class projects are due at 12:00 PM (noon) Eastern.

Tue Dec 22

Faculty Grade Deadline

 

Notes, Policies, and Guidelines

o   You are expected to have read the assigned material before each lecture.

o   We will use Northeastern’s Canvas for announcements, assignments, and your contributions.

o   When emailing me, begin the subject line with [fa26 nets].

o   For your class project, you can use whatever programming language that you like.

o   Refresh your knowledge of the university's academic integrity policy and plagiarism. There is zero-tolerance for cheating!

o   AI Usage Policy: You may use AI tools in any way that supports your learning, with two conditions. (1) For each submission, disclose which tools you used, at what stage, and how much of the output you kept. Disclosure will never affect your grade; failing to disclose is an academic integrity violation. (2) You are responsible for everything you submit. Fabricated citations, incorrect arguments, and broken code are your errors and will be graded as such. You should be able to explain and defend any part of your work without the tool.

Ø  Disclosure: I gave Fable 5.1 my original policy. Fable 5.1 critiqued it and proposed a revised policy. I then edited Fable 5.1’s revised policy to produce this policy.