Hasso-Plattner-Institut25 Jahre HPI
Hasso-Plattner-Institut25 Jahre HPI

Graph Neural Networks (Wintersemester 2020/2021)

Dozent: Prof. Dr. Holger Giese (Systemanalyse und Modellierung) , Christian Medeiros Adriano (Systemanalyse und Modellierung) , Sona Ghahremani (Systemanalyse und Modellierung)

Allgemeine Information

  • Semesterwochenstunden: 4
  • ECTS: 6
  • Benotet: Ja
  • Einschreibefrist: 01.10.200-20.11.2020
  • Lehrform: Projekt / Seminar
  • Belegungsart: Wahlpflichtmodul
  • Lehrsprache: Englisch

Studiengänge, Modulgruppen & Module

Data Engineering MA
Digital Health MA
IT-Systems Engineering MA
  • OSIS: Operating Systems & Information Systems Technology
    • HPI-OSIS-K Konzepte und Methoden
  • OSIS: Operating Systems & Information Systems Technology
    • HPI-OSIS-T Techniken und Werkzeuge
  • OSIS: Operating Systems & Information Systems Technology
    • HPI-OSIS-S Spezialisierung
  • SAMT: Software Architecture & Modeling Technology
    • HPI-SAMT-K Konzepte und Methoden
  • SAMT: Software Architecture & Modeling Technology
    • HPI-SAMT-T Techniken und Werkzeuge
  • SAMT: Software Architecture & Modeling Technology
    • HPI-SAMT-S Spezialisierung


Graph Neural Networks are a set of novel techniques aimed at enabling deep neural networks to work on heterogeneous graphs, e.g., social networks, knowledge graphs, recommender systems, gene regulatory networks, traffic networks, etc. These heterogeneous graphs became central components of complex software systems, which make predictions and interventions on the structure and behavior of the system and its context (e.g., users, environment). Take for instance, the mitigation of the spread of an event (accident, rumors, disease) through a network (traffic, social network). The heterogeneous network would model the current state of connections, their strength, and multiple properties in their nodes and edges. This complex data would be used to  predict the propagation paths and indicate the best set of nodes to intervene to mitigate the spread of rumors or disease and its effect on a system as a whole (traffic, population). Interventions can involve changes in the connections (isolate, circumvent, shortcut) and the attributes of the nodes (duplicate, reset, extend).

Note however, that neural networks algorithms were initially developed to solve problems in the domains of computer vision (3D and 2D) and natural language processing (1D). These domains present Euclidean spaces that allow neural networks to leverage on the regularities inherent to coordinate systems, vector space structures, or fixed distributions. Because these regularities are not guaranteed to be present in heterogeneous networks, new extensions were developed to these algorithms.

In this seminar we will study extensions to neural networks and apply them to heterogeneous graphs of real world problems. We will tackle these domain-specific problems in a principled way by solving the more generic graph-related problems of node classification, structure discovery, path identification, subgraph generation. 


[1] Zhou, J. et al. (2018). Graph neural networks: A review of methods and applications. https://arxiv.org/pdf/1812.08434.pdf

[2] Xu, K., et al., (2018). How powerful are graph neural networks?. arXiv preprint arXiv:1810.00826. https://arxiv.org/pdf/1810.00826.pdf

[3] Liu, Z., Zhou, J. (2020) Introduction to Graph Neural Networks (Synthesis Lectures on Artificial Intelligence and Machine Learning) https://www.amazon.de/dp/1681737655/ref=rdr_ext_tmb 

[4] Leng, S., Xu, Z., & Ma, H. (2019). Reconstructing directional causal networks with random forest: Causality meeting machine learning. Chaos: An Interdisciplinary Journal of Nonlinear Science, 29(9), 093130

[5] Ke, N. R., et al., (2019) Learning neural causal models from unknown interventions. arXiv preprint arXiv:1910.01075.

[6] Y. Yu, et al., (2019) Dag-gnn: Dag structure learning with graph neural networks. arXiv preprint arXiv:1904.10098, 2019.

Lern- und Lehrformen

The course is a project seminar, which has an introductory phase comprising initial short lectures. After that, the students will work in groups on jointly identified experiments applying specific solutions to given data sets and finally prepare a presentation and write a report about their findings concerning the experiments.

There will be an introductory phase to present basic concepts for the theme including the necessary foundations for online machine learning.

Announcement regarding the coronavirus regulations: 
Because of the Coronavirus situation and corresponding restrictions outbreak, we will organize all meetings as online meetings by default. This especially applies to the first lecture. If all participants agree to and the current restrictions as well as seminar room availability allow it, further meetings may also take place at HPI.


We will grade the group's experiments (60%), reports (30%), and presentations (10%). Participation in the project seminar during meetings and other groups' presentations in the form of questions and feedback will also be required.


After the introductory phase with few initial short lectures, we will identify the group topics and then there will be regular individual feedback rounds of the groups with their supervisors. In addition, there will also be regular meetings during the semester for the whole project seminar to discuss the progress of all groups and open questions in general.

 Lectures are scheduled for 

  • Tue 11h00 - 12h30
  • Wed 13h30 - 15h00

The first lecture will be held on Tuesday, November 10 at 11h00, via an online meeting.

The interested students need to contact us christian.adriano(at)hpi.de to obtain the online meeting credentials by November 09 for attending the lecture. 


Please note that you do not have to enroll for this course at the Studienreferat. After the introductory meeting, we will report any participants that decide to actually take the course to the Studienreferat. 

If you have any further questions, please also send an email to christian.adriano(at)hpi.de.