E-mail senden E-Mail Adresse kopieren

Aleksandar Bojchevski




Stuhlsatzenhaus 5
66123 Saarbrücken (Germany)

Weitere Informationen

Veröffentlichungen von Aleksandar Bojchevski

Jahr 2022

Konferenz / Medium

Advances in Neural Information Processing Systems 35 (NeurIPS 2022)NeurIPS

Konferenz / Medium

Advances in Neural Information Processing Systems 35 (NeurIPS 2022)NeurIPS

Konferenz / Medium

International Conference on Learning RepresentationsICLR

Lehre von Aleksandar Bojchevski

Winter 2021/22

Trustworthy Graph Neural Networks

Seminar Description:

Graph Neural Networks (GNNs) have emerged as a fundamental building block in many artificial intelligence systems. Even beyond uses where the graph structure is explicit (e.g. social networks), they show impressive performance for general object-oriented perception, representation, and reasoning. In this seminar we will cover GNNs that are not only accurate or efficient, but also robust, privacy-preserving, fair, uncertainty-aware, and explainable. We will explore how GNNs fail w.r.t. these trustworthiness aspects and how to improve them.

Organization (block format):

Each student will receive a few research papers on a single topic which they should carefully read and analyze. Starting from these initial papers they should explore the surrounding literature and summarize their main ideas and findings in a 4-page seminar paper. Students will also participate in a peer-review process where they have to provide constructive feedback on each other's work (1 page review for 3 other papers). Finally, each student will prepare and deliver a presentation about their topic during a block seminar at the end of the semester.


  • Seminar paper (40%)
  • Presentation (30%)
  • Reviews (30%, 10% for each review)


  • Kick-off meeting at the start of the semester (online via Zoom)
  • Deadline for the first draft of the seminar paper
  • Deadline for the final version of the seminar paper
  • Deadline for the reviews
  • Feedback round / practice talk with your supervisor
  • Final presentations at the end of the semester

Exact dates and times will be determined soon.

Preliminary list of topics:

  1. Adversarial attacks
  2. Heuristic defenses
  3. Provable defenses
  4. Privacy attacks
  5. Privacy-preserving models
  6. Bias
  7. Fairness
  8. Instance-level explanations
  9. Model-level explanations
  10. Uncertainty-aware models

More details and the final list of topics will be provided in the kick-off meeting.

You should attend this seminar to:

  • Explore and learn about state-of-the-art research on Graph Neural Networks
  • Improve your scientific writing
  • Improve your presentation skills
  • Participate in a review process akin to international conferences