Given the black-box nature of the state-of-the-art AI models and the lack of associated formal guarantees, there is a growing interest in using formal methods for AI-based systems to ensure their reliability and interpretability. This direction is a key component of the so-called “Third wave of AI”. Similarly, there is a growing interest in leveraging data-driven machine learning for knowledge discovery and boosting logical inference. This course will introduce recent developments in both directions and outline several promising future research directions. Overall, the students will be exposed to the following topics:
- Verification of AI systems
- Symbolic explanations of neural networks
- Training and querying with logic
- Robust training methods
- Neural network repair
- Probabilistic circuits
- Neurosymbolic computing
- Machine learning for verification
The course will be taught in person. For students preferring to attend remotely, the video recording of the lectures will be made available online.
The students will work in a group of 2 or individually on:
- a course project,
- conference-style paper presentation on one of the topics taught in the course towards the end of the semester (see dates above). The students are free to select any paper, but it must be approved by us.
Both project and presentation will make up 50 % of the grade each.
|Sep 21||Register your group for the project and presentation|
|Oct 26||Selecting a paper to present|
|Nov 10||Project submission|