This seminar will look at current developments of machine learning and networking. More specifically we will look at
- how machine-learning techniques can improve network operation
- how networks can improve (distributed) machine-learning scenarios (e.g., edge learning).
A list of papers that serve as starting points is available here.
As a typical seminar, the goal is to make yourself acquainted with the scientific process, especially publication structures and techniques (in particular, technical writing and presentations), as well as with the concrete topic at hand. We will hence mimic both the paper writing process and the paper publication process of a typical conference in this seminar. This comprises:
- Finding relevant literature: Each participant will get a paper as a starting point for a literature search
- Selecting papers: Such a literature search typically produces far too many results to be manageable; we need to condense that done to a manageable reading list of a handful of papers per participant
- Reading papers: Quickly figuring out key information
- Comparing papers: Comparing approaches, concepts, assumptions, limitations; identifying strengths and weaknesses
- Writing a report: Based on the paper analysis, write a report; this also serves to train technical writing skills and is an excellent preparation for writing a Master thesis
- Evaluating reports: we will exchange draft reports among participants, writing reviews of other students' drafts (acting like a programm committee member for a conference)
- Presenting the paper analysis in a talk
- Discussing your own talk and those of others; actively engaging in a scientific discussion
To make sure everybody has the same chances, we will run the seminar as a block event towards the end of the semester. There will be intermediate milestones for all the points above, as well as some meetings at the start of the semester to talk about all these milestones (e.g., discuss technical writing; how to give a research talk; ...).
All material will be provided via Moodle. Please register for this course in Moodle! The Moodle course has all deadlines; the first one is to rank your paper choices by Saturday, April 23rd!
The goal is to have this seminar take place in person, on campus. If the Covid situation should deteriorate, we can discuss options for an online version amont the participants.
You really should have some background in computer networks in general; experience with machine learning is a plus, but not required. If you are unsure, please talk to Holger Karl and we can see whether the class still makes sense for you.
The will be no separate exam as such. The grade will result from the quality of your milestone submissions.
Please check official web page for module relationship.