Machine Learning for Graphs/Networks

Topics: Machine Learning for Graphs & Networks, Relational Learning, Deep Learning for Graphs, Graph Neural Networks, Network Analysis and Mining

One of the most common assumptions in many machine learning and data analysis tasks is that the given data points are realizations of independent and identically distributed random variables. This assumption, however, is often violated. Sensors are interlinked with each other in networked cyber physical systems, molecules or proteins interact based on biochemical events, and knowledge bases capture the relationships between different entities and concepts. Technically, in all these domains we have to deal with large-scale complex graphs/network.

In our research, we design machine learning and data analytics approaches which inherently capture the underlying dependence structure provided by the graph. In this regard, we cover a wide range of application use cases including novel models for node classification, graph clustering, anomaly detection, and graph generation and network inference. From a technical perspective, our methods often based on principles such as probabilistic modelling or neural networks.

Deep Learning on Graphs: While neural networks have achieved unprecedented performance on a variety of problems in different fields (e.g. computer vision and speech recognition), in the past, neural network architectures have focused on classical data domains (e.g. images, sequences). Recently the studies have been extended to the graph domain. Due to the unique characteristics of graphs (e.g. neighborhoods of varying size, long-range dependencies between nodes, sparsity), designing effective approaches, however, is highly challenging. In our research, we develop and analyze learning principles that enable to use neural networks in the context of graph data.

Selected Publications: 

Adversarial Attacks & Robustness: Are graph learning approaches robust? How to attack them? How to improve their robustness?

  • Daniel Zügner, Stephan Günnemann
    Adversarial Attacks on Graph Neural Networks via Meta Learning
    International Conference on Learning Representations (ICLR), 2019
  • Daniel Zügner, Amir Akbarnejad, Stephan Günnemann
    Adversarial Attacks on Neural Networks for Graph Data (Best Research Paper Award)
    ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), 2018
  • Aleksandar Bojchevski, Stephan Günnemann
    Adversarial Attacks on Node Embeddings, Arxiv
  • Aleksandar Bojchevski, Stephan Günnemann
    Bayesian Robust Attributed Graph Clustering: Joint Learning of Partial Anomalies and Group Structure
    AAAI Conference on Artificial Intelligence, 2018
  • Aleksandar Bojchevski, Yves Matkovic, Stephan Günnemann
    Robust Spectral Clustering for Noisy Data: Modeling Sparse Corruptions Improves Latent Embeddings
    ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), 2017

Neural Network Architectures for Graphs: How can we design novel neural network architectures that exploit the underlying properties of graphs? How can we capture relational dependencies to improve performance of learning?

  • Johannes Klicpera, Aleksandar Bojchevski, Stephan Günnemann
    Predict then Propagate: Graph Neural Networks meet Personalized PageRank
    International Conference on Learning Representations (ICLR), 2019
  • Oleksandr Shchur, Maximilian Mumme, Aleksandar Bojchevski, Stephan Günnemann
    Pitfalls of Graph Neural Network Evaluation
    Relational Representation Learning Workshop, NIPS 2018
  • Federico Monti, Oleksandr Shchur, Aleksandar Bojchevski, Or Litany, Stephan Günnemann, Michael M. Bronstein
    Dual-Primal Graph Convolutional Networks, Arxiv
  • Aleksandar Bojchevski, Stephan Günnemann
    Deep Gaussian Embedding of Graphs: Unsupervised Inductive Learning via Ranking
    International Conference on Learning Representations (ICLR), 2018 

Graph Clustering / Segmentation / Community Detection: How do clusters/communities behave in real graphs? How to reliably and efficiently detect them?

  • Saskia Metzler, Stephan Günnemann, Pauli Miettinen
    Stability and Dynamics of Communities on Online Question-Answer Sites
    Social Networks, 2019
  • Aleksandar Bojchevski, Stephan Günnemann
    Bayesian Robust Attributed Graph Clustering: Joint Learning of Partial Anomalies and Group Structure
    AAAI Conference on Artificial Intelligence, 2018
  • Aleksandar Bojchevski, Yves Matkovic, Stephan Günnemann
    Robust Spectral Clustering for Noisy Data: Modeling Sparse Corruptions Improves Latent Embeddings
    ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), 2017

Generative Models for Graphs & Graph/Network Inference: How can we automatically generate graphs with realistic properties? Can we infer graphs/networks from other sources of information?

  • Subhabrata Mukherjee and Stephan Günnemann
    GhostLink: Latent Network Inference for Influence-aware Recommendation
    International World Wide Web Conference (WWW / TheWebConf), 2019 
  • Aleksandar Bojchevski, Oleksandr Shchur, Daniel Zügner, Stephan Günnemann
    NetGAN: Generating Graphs via Random Walks
    International Conference on Machine Learning (ICML), 2018
  • Aleksandar Bojchevski, Stephan Günnemann
    Bayesian Robust Attributed Graph Clustering: Joint Learning of Partial Anomalies and Group Structure
    AAAI Conference on Artificial Intelligence, 2018

Scalable Learning / Scalable Graph Analytics: How to ensure fast learning and analytics when operating with graph data?

  • Dhivya Eswaran, Stephan Günnemann, Christos Faloutsos
    The Power of Certainty: A Dirichlet-Multinomial Model for Belief Propagation
    SIAM International Conference on Data Mining (SDM), 2017
  • Dhivya Eswaran, Stephan Günnemann, Christos Faloutsos, Disha Makhija, Mohit Kumar
    ZooBP: Belief Propagation for Heterogeneous Networks
    International Conference on Very Large Data Bases, PVLDB 10(5): 625-636 (2017)
  • Manuel Then, Timo Kersten, Stephan Günnemann, Alfons Kemper, Thomas Neumann
    Automatic Algorithm Transformation for Efficient Multi-Snapshot Analytics on Temporal Graphs
    International Conference on Very Large Data Bases, PVLDB 10(8): 877-888 (2017)