Introduction to projects

August 2-20, 2021

Project Structure

The definitive guide to project schedule and structure is in the Daily Guide to Projects, but we include some overview information here.

The NMA projects are conducted over 14 days: all of the course days except the final day. The projects are a core part of the summer school, and will teach different skills from those you learn during the course work.

Your pod will split into two groups at the end of the first day, based on common research interests, which can be

  • neuroscience

  • computer vision

  • natural language processing (advanced)

  • reinforcement learning (advanced)

Note that reinforcement learning is an advanced topic at NMA, taught in week 3 just before the end. Also, NLP uses concepts from week 2, like recurrent neural networks and transformers. We also provide fewer project templates for these topics.

For groups of students from a neuroscience background we recommend to do a neuroscience project because we have many curated datasets and materials available from the comp-neuro course (here). Furthermore, deep learning is much more powerful in an application for which you have domain knowledge (or a big enough dataset, but that’s rare…). In any case, NMA will not teach you how to build huge, energy-hungry models but rather how to combine modelling with deep learning using your insights from domain knowledge.

After groups are formed on Monday, you need to submit this form, and we will assign to each group a project TA based on your chosen topic and the project TA will start meeting with you every 2-3 days to help in any way they can. There are no external mentors for the DL course.

On days 2-5 you will lay the foundation for your projects, by finding an interesting question to ask, and an appropriate dataset for answering the question. We encourage you to be creative with your question, and to use non-standard datasets!

During the second week, you will refine your question iteratively as you try things out. You will experiment with different types of deep neural networks to find one that works for your project. This will culminate in an “abstract” that you write on day 11 (Monday of week 3), which is a full projects day with some structured guidance to the process of writing abstracts.

The rest of week 3 you will continue implementing your project and trying to get an answer to your question, culminating in a short presentation to your pod on the final day 15 (5 minutes, 3-5 slides, informal, take turns to present).

Daily Schedule

Detailed schedule about what to do every day is in the Daily student guide for projects. You should refer back to this document often.

Daily project time is either:

  • Early — 3h before content (time slots 2,4)

  • Late — 3h after content (time slots 1,3,5)

Day 11 (Monday of week 3) is a full day for projects, meaning you don’t have any coursework. This will primarily be used to put together an abstract for your project.

Ten steps how to model materials

These materials (see here) are a step-by-step guide to research which approximately aligns with the first two weeks of the projects schedule. They are similar to the comp-neuro materials.

Project Templates

Project templates (here) are project ideas prepared by the NMA team and by the project TAs. They each include substantial code to get you started in some direction for your project, and they also include slides with background and potential questions to ask. We highly recommend these for beginners. Even if your group is more advanced, you can use the project templates for inspiration and for setting a starting point in your project. There are also some extra ideas from the project TAs that did not become full project templates, that you can check in the “Ideas” page for each topic. There you will also find links to various datasets that may be useful.

Project examples 2020 - comp-neuro

These are projects submitted by NMA-CN students in 2020. They don’t have deep learning, but they should give you an idea of the kind of project you can do in three weeks.

Important deadlines

  • project day 1: by end of your project time, please submit your group name and group members via this Airtable link. We will use this immediately to match your group with a project TA.

  • project day 11: by end of your project time, please submit your abstract via this Airtable link.

  • project day 14: by end of your project time, please submit your slides via this Airtable link.

Presentations — Start preparing project presentation after day 11

  • Groups make an approx five minute project presentation: 1 minute / 1 slide / 1 person

  • All group members should give a part of the presentation!

  • RESULTS ARE OPTIONAL! We would like you to present your experience, and results if you happen to have them

  • PROJECTS ARE DUE at the end of day 14!!

  • Use this presentation style (google slides or powerpoint) or create your own style!

NMA final day — Presentation of projects

  • This happens either in the last hour of normal project time (timeslots 2,4) or during 6:10-7:10 relative to start of tutorials on the last day (timeslots 1,3,5). If it’s not clear, check the calendar.

  • For each group, we will send an email invitation to the following people: TA, project TA and primary student contact.

  • More about scheduling and logistics in the Student Guide to Projects