Graph & Causal Representation Learning logopeterindia.net
Knowledge Hub

Graph Representation Learning & Causal Representation Learning

How machines learn useful vectors from graphs and from data generated by hidden causes — surveys, foundational papers, libraries, benchmarks and courses for both fields.

96 topics•11 categories•Updated October 2026

Overview & Surveys

7 entries

Graph representation learning learns low-dimensional vectors for nodes, edges or whole graphs so that structure and attributes can feed downstream models. Causal representation learning (CRL) asks a different question: can we recover high-level causal variables, and the relations between them, from low-level data such as images? The two meet in graph-structured causal models, causal graph neural networks and invariance-based learning. Start with the surveys, then pick a track below.

01Representation Learning on Graphs: Methods and Applications (Hamilton et al.)Early survey of node embeddings and graph neural networks. 02Graph Representation Learning (Hamilton, book)Free book covering embeddings, GNNs and generative graph models. 03A Comprehensive Survey on Graph Neural NetworksTaxonomy of graph neural network models. 04Graph Neural Networks: A Review of Methods and ApplicationsGeneral design pipeline and applications. 05Toward Causal Representation Learning (Schölkopf et al.)The paper that framed causal representation learning. 06Towards Causal Representation Learning (Schölkopf, Proceedings of the IEEE)Journal version of the above. 07Causality for Machine Learning (Schölkopf)Short introduction to causality in ML.

Node Embeddings & Graph Neural Networks

10 entries

Shallow embeddings and message-passing models.

08DeepWalkOnline learning of social representations with random walks. 09node2vecBiased random walks for scalable feature learning. 10node2vec project pageCode and datasets. 11LINE: Large-scale Information Network EmbeddingFirst- and second-order proximity embeddings. 12Semi-Supervised Classification with Graph Convolutional Networks (GCN)The widely used graph convolutional network. 13Inductive Representation Learning on Large Graphs (GraphSAGE)Sampling and aggregating neighbours. 14Graph Attention Networks (GAT)Attention over neighbours. 15How Powerful are Graph Neural Networks? (GIN)Expressive power and the Weisfeiler-Lehman test. 16Neural Message Passing for Quantum ChemistryThe message passing framework. 17Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and GaugesUnified view of symmetry-based architectures.

Graph Transformers & Self-Supervised Learning

8 entries

Beyond message passing, and learning without labels.

18Do Transformers Really Perform Bad for Graph Representation? (Graphormer)Transformer with structural encodings. 19Recipe for a General, Powerful, Scalable Graph Transformer (GPS)Combining local message passing with global attention. 20Rethinking Graph Transformers with Spectral AttentionLearned positional encodings from the Laplacian. 21Variational Graph Auto-EncodersUnsupervised graph embedding. 22Deep Graph InfomaxContrastive learning by maximising mutual information. 23Strategies for Pre-training Graph Neural NetworksNode- and graph-level pre-training. 24Graph Contrastive Learning with Augmentations (GraphCL)Augmentation-based contrastive learning on graphs. 25Graph Self-Supervised Learning: A SurveyContrastive, generative and predictive methods.

Knowledge Graph & Heterogeneous Graph Embeddings

6 entries

Learning from relations between entities.

26Translating Embeddings for Modeling Multi-relational Data (TransE)The classic translation-based model. 27Modeling Relational Data with Graph Convolutional Networks (R-GCN)GCNs for multi-relational graphs. 28RotatE: Knowledge Graph Embedding by Relational RotationRelations as rotations in complex space. 29Knowledge Graph Embedding: A Survey of Approaches and ApplicationsSurvey of embedding models. 30Heterogeneous Graph Attention NetworkMeta-path based attention. 31PyKEENLibrary for knowledge graph embeddings.

Causal Representation Learning: Core Papers

9 entries

Recovering causal variables from data.

32Weakly-Supervised Disentanglement Without CompromisesDisentanglement with pairs of related observations. 33Challenging Common Assumptions in the Unsupervised Learning of Disentangled RepresentationsWhy unsupervised disentanglement needs inductive biases. 34Towards Nonlinear Disentanglement in Natural Data with Temporal Sparse CodingIdentifiability from temporal structure. 35Nonlinear ICA Using Auxiliary Variables and Generalized Contrastive LearningIdentifiable nonlinear latent models. 36Interventional Causal Representation LearningIdentifiability from interventions. 37Embrace the Gap: VAEs Perform Independent Mechanism AnalysisWhy VAEs can recover independent latent mechanisms. 38CausalVAE: Structured Causal Disentanglement in Variational AutoencoderVAE with a structural causal layer. 39Learning Linear Causal Representations from Interventions under General Nonlinear MixingIdentifiability for linear latent causal models. 40Causal Component AnalysisCombining causal discovery with independent component analysis.

Invariance, Out-of-Distribution Generalisation & Causal Graph Learning

7 entries

Using causal structure for robustness, including on graphs.

41Invariant Risk MinimizationLearning invariant predictors across environments. 42In Search of Lost Domain GeneralizationCareful comparison of domain generalisation methods. 43Causal Inference in Statistics: A Primer (Pearl et al.)Companion site for the introductory text. 44Elements of Causal Inference (open access)Foundations and learning algorithms; open-access PDF from MIT Press. 45Discovering Invariant Rationales for Graph Neural NetworksCausal rationales for OOD graph tasks. 46Handling Distribution Shifts on Graphs: An Invariance PerspectiveRisk extrapolation for graphs. 47Graph Out-of-Distribution Generalization with Controllable Data AugmentationAugmentation for graph OOD.

Causal Discovery & Structure Learning

8 entries

Finding the graph itself.

48DAGs with NO TEARSContinuous optimisation for DAG learning. 49Causal Discovery with Reinforcement LearningRL-based structure search. 50Differentiable Causal Discovery from Interventional DataNeural DAG learning with interventions. 51A Survey on Causal Discovery Methods for I.I.D. and Time Series DataOverview of discovery methods. 52Causal-learn (py-why)Python package for causal discovery. 53gCastle (Huawei Noah's Ark Lab)Causal structure learning toolchain. 54TetradCausal discovery software from Carnegie Mellon. 55causalnexBayesian networks for causal reasoning; check maintenance status before adopting.

Libraries & Frameworks

13 entries

Open-source tools for both fields.

56PyTorch GeometricThe most widely used PyTorch GNN library. 57PyTorch Geometric on GitHubSource code and releases. 58Deep Graph Library (DGL)Framework-agnostic graph neural network library. 59DGL on GitHubSource code. 60SpektralGraph neural networks in Keras and TensorFlow. 61TensorFlow GNNGoogle's TF-GNN library. 62JraphGraph networks in JAX. 63NetworkXPython graph analysis and construction. 64Karate ClubUnsupervised graph embedding methods. 65DoWhy (py-why)Causal inference with explicit graphical models. 66EconML (py-why)Heterogeneous treatment effects with machine learning. 67CausalML (Uber)Uplift modelling and causal inference methods. 68disentanglement_lib (Google Research)Disentanglement metrics and models; research code.

Benchmarks & Datasets

9 entries

Where methods are compared.

69Open Graph Benchmark (OGB)Realistic large-scale graph benchmarks and leaderboards. 70Open Graph Benchmark: Datasets for Machine Learning on GraphsThe OGB paper. 71Benchmarking Graph Neural NetworksStandardised GNN benchmarking framework. 72Long Range Graph BenchmarkTasks requiring long-range dependencies. 73TUDatasetCollection of graph classification datasets. 74Stanford SNAP datasetsLarge network datasets. 75GOOD: A Graph Out-of-Distribution BenchmarkOOD benchmark for graphs. 76Causal ChamberReal physical systems with ground-truth causal structure. 77CausalWorldRobotic manipulation benchmark for causal reasoning.

Courses, Books & Reading

9 entries

Where to learn.

78Stanford CS224W: Machine Learning with GraphsCourse by Jure Leskovec with open lecture materials. 79CS224W lectures on YouTubeRecorded lectures. 80Graph Neural Networks (Petar Veličković, resources)Talks and notes by a GNN researcher. 81A Gentle Introduction to Graph Neural Networks (Distill)Interactive primer. 82Understanding Convolutions on Graphs (Distill)Interactive explanation of graph convolutions. 83Causal Inference for the Brave and TrueFree book with Python examples. 84Causal AI (Robert Ness, Manning)Book on causal modelling with code. 85Brady Neal: Introduction to Causal InferenceFree course and lecture notes. 86CLeaR: Conference on Causal Learning and ReasoningConference on causal learning and reasoning.

Related Pages

10 entries

Neighbouring pages on this site.

87Graph Neural NetworksOverview of GNNs. 88GNN FrameworksFrameworks for GNNs. 89GNN LibrariesLibraries for GNNs. 90Understanding GNNsPrimer on GNNs. 91Causal AICausal AI models and tools. 92Knowledge GraphsKnowledge graph resources. 93Deep LearningDeep learning hub. 94Machine LearningMachine learning hub. 95Generative AIGenerative AI hub. 96Graph Data PlatformsGraph data platforms.

Frequently Asked Questions

What is graph representation learning?

Graph representation learning maps nodes, edges or entire graphs to dense vectors that preserve structure and attributes, using random-walk embeddings, graph neural networks or graph transformers, so the vectors can drive tasks such as node classification, link prediction and graph classification.

What is causal representation learning?

Causal representation learning aims to discover high-level causal variables and their relations from low-level observations such as images or sensor streams. Identifiability results typically need extra assumptions, such as interventions, multiple environments or temporal structure, so it remains mainly a research area.

How are the two fields related?

Both treat structure as information. Graph methods learn from known relational structure, while causal methods try to learn or use the structure that generated the data. They meet in graph-structured causal models, causal graph neural networks for out-of-distribution generalisation, and causal discovery on networks.

Which library should I start with?

For graph learning, PyTorch Geometric or DGL are the common starting points. For causal work, DoWhy and EconML cover causal inference and effect estimation, while causal-learn and gCastle cover causal discovery. Pure causal representation learning mostly has research code attached to individual papers.

How current is this page?

Links were checked in October 2026. Causal representation learning is a fast-moving research area, so most entries are papers rather than production tools. Sites that block automated checks are kept and may need a manual click.