Machine Learning for Complex Networks

    Thesis Topics

    Please feel free to contact us if you are interested in writing a bachelor or master thesis with us. Below you can find a (non-exhaustive) list of open thesis topics:

    A Systematic Evaluation of Graph Embeddings for Supervised and Unsupervised Link Prediction and Graph Reconstruction

    • Investigate effect of hyperparameter tuning for specific data sets
    • Check how well different embedding techniques generalize with the same default parameters

     Robustness of Graph Embeddings in Uncertain Network Data

    • Check how graph embeddings are affected by missing or spurious edges
    • Test in different graph learning tasks

     Optical Network Recognition

    • Develop a method to reconstruct networks from images/plots
    • Support for node colors (communities) + vertex labels, edge weights, etc.
    • Investigate methods to support curved edges
    • Evaluation in different plots generated by different layout algorithms

     German AI Map

    • Creation of an interactive zoomable map of researchers in AI and Data Science using DBLP data
    • Support for Citation and Collaboration networks

    pathpy-torch_geometric API

    • Functions to convert to/from torch-geometric data sets
    • Support for temporal and heteregeneous graphs
    • pathpy as bridge to load networks (incl. attributes) from network repositories (e.g. Netzschleuder, KONECT etc...)

    Higher-order-network models and ML applications

    • Generative models for higher order networks
    • Statistical inference of higher order network representations
    • ML applications of higher order models (classification, ranking, clustering, representation learning etc.) 

    Exploring feature space for graph classification and embeddings. 

    •  Include graph features into ML methods for graphs
    •  Methods to be explored include graph neural networks, graph embeddings, random forests and SVMs
    •  Potential features include spectral features of graph operators, graphlets and other higher order features
    •  Test and compare to existing methods on benchmarks