If you have ended up on this page, this probably means that we have started (or are about to start) a project together. I have gathered my thoughts and learnings from both working in ML as well as supervising students in the notes below, which are hopefully of help. Especially if you are a student, please take some time to go through them!

Getting Started


To start, there are two documents that contain most of the guidelines and you should read first: 1) The MLO Project Guidelines; they are not written by me, but contain a lot of useful information and the formal organization of our group, such as support, expectations, and deliverables. 2) Tips to Make the Most Out of Your Project With Me.

Don’t be overwhelmed by the documents or the guidelines! In a nutshell: The goal of a project is to do student-driven research where you work on topics that excite you. You’ll be the project leader, while we are your guide and support system. Expect the research journey to be dynamic and often uncertain – that’s normal. The main goal is to learn during a project and take agency, and I highly value healthy relationships so that you can grow as a researcher and have a great experience.

Expectations. The role of me as a supervisor is to guide you, but you are the one who is responsible for the project. In particular, you should take agency: implement and debug code independently, execute experiments, read papers, plan the project and so on. Importantly, let us try to align expectations and goals early on, i.e., tell me what you are trying to get out of the project.

Support. Depending on availability, we usually offer desk space in INJ, where you meet other students or members of the lab. We also setup regular meetings, a Slack channel, and a shared document to keep track of the project. The cluster access is documented here.

Documents. At the beginning of the project, we should set up an overview document of the project where we keep a summary of the research project, meeting notes and references. This is useful for both of us to keep track of the progress and to have a shared understanding of the project. This document will naturally change over time, but is important as it is easy to lose perspective as we get deeper into running experiments.

Please consult the second document above with detailed discussions. You can also come back to it frequently. More resources for doing research (especially empirical ML):

Writing, Figures, Presentations


At some point, we will need to write a report or paper about our project, and students often struggle with this initially. Writing is a skill that can be learned, and the more you write, the better you get. The points below hopefully help to get started.

Figures. Figures are the most important part of a paper. They should be of high quality, informative, and easy to understand. Usually, I strive to have all figures done before even writing the bulk of the paper, as they contain the main storyline. All figures and especially the captions should be self-contained and understandable without reading the text, so the caption should include the main takeaway (e.g., boldfaced) and the context. Also, create PDFs of your figures and include them in the document, so that they are not compressed by the Latex compiler (otherwise they get pixelated when zooming in).

Writing. After having most figures, writing the paper is usually easier. Personally, I almost always think about the story of a paper in the following structure:

  • Importance: What are we looking at and why is it relevant? This sets up the broader context and motivation.
  • Gap: What is missing in the current literature and what are the limitations of existing work?
  • Objective: What specific problem are we trying to solve or question are we trying to answer?
  • Method: How do we approach solving this problem? What are our key technical contributions?
  • Findings: What are the main experimental results and insights we discovered?
  • Implications: What do our findings mean for the field? What are the key takeaways?
  • (Optional) Future work: What interesting directions remain to be explored?

These points should appear in both the abstract (very concisely) and introduction (with more detail and context). The rest of the paper then flows naturally from this structure, with sections organized around presenting and analyzing the key figures and results.

Before starting to write, students new to writing should read this more detailed writing guide (with Latex tips, general structure, nitpicks, etc.): Advice on writing ML papers. Other useful guides:

Presentations. Putting effort into presentations is worth it! It is a great way to communicate your work, to get feedback, or to have people remember your work (and you). The most important takeaway for research presentations in my experience is this: focus most of your effort (not necessarily time of the presentation!) to clearly convey your problem and goal. At the least, everyone in the audience should be able to give a summary of what your project tackled. In contrast, the method or detailed results might be secondary, and an interested listener can look that up later or ask you. In short: get people excited to learn more.

More generally, IMO the best guide to research presentations is by Prof. Püschel at ETH: How to Give Strong Technical Presentations. It is quite detailed, but worth reading.