Advanced Topics in Machine Learning and Optimization
Objectives
The course aims at introducing students to some selected advanced topics on machine learning and optimization. The topics are chosen so as to cover relevant recent research directions and challenges for the machine learning and artificial intelligence community, as well as promising technological advancements.
Prerequisites
A good knowledge of the basics of machine learning (including deep learning) and artificial intelligence (including logic and probabilistic reasoning) is essential. Students are expected to take this course after completing Machine Learning and Fundamentals of Artificial Intelligence from the first year of the AIS degree (or similar teachings).
Content
A preliminary list of topics that will be covered throughout the course:
- Explainable Artificial Intelligence
- Statistical Relational Artificial Intelligence
- Neuro-Symbolic integration
- Graph Neural Networks
- Structured-output learning
- Interactive Machine Learning
- Interpretable-by-design Neural Networks
Slides
- Graph Neural Networksslides handouts notebook notebook (pdf)
- Structured Output Predictionslides
- Statistical Relational AIslides handouts
- Neuro-Symbolic Integrationslides handouts lab material (tgz)
Exams
The student will be asked to realize a machine learning project concerning one or more of the topics seen during the course and write a short report. Suggested approximate length for the report is 5 pages in double-column format plus references. Potential topics will be provided by the teacher. Students are free to propose alternative topics.
The exam consists of the evaluation of the project (70%) and the discussion of the report and the topics behind it (30%). The student should contact the teacher (or the teaching assistant responsible for the chosen topic) as soon as the report is complete to arrange the discussion. Preliminary versions of the report can also be sent to gather feedback and make adjustments if needed.
Students will be evaluated based solely on their report and interview according to the following criteria:
- Is the report clear and well structured? Are technical terms used appropriately? Are mathematical formulas, if any, correct?
- Is the work properly positioned against the literature?
- Is the model described in sufficient detail? Are all methodological choices motivated?
- Does the evaluation address the research question? Are the results appropriately explained? Are negative results, if any, appropriately contextualized? Does the description enable reproducibility?
Please note that the discussion of the report is conducted asynchronously via email and is separate from the official exam date. The exam date is only used for registration purposes after the exam has been completed. Kindly ensure that you submit your report at least one week before the registration date you plan to use.