You have a a research idea where you believe machine learning could help. We offer fast-paced cooperation projects of up to 6 months to get you off the ground. If you are a researcher in Tübingen, read on for details.
► 1️⃣ A research problem in the natural or social sciences, or the humanities
We need a well-defined quantitative metric of performance. What kind of results would make the project a success?
Provide us with relevant references, including baselines that don't use ML techniques.
If you'd like to learn about ML before engaging with us, we recommend this Introduction to Machine Learning series. If you need different materials, write to us for advice!
► 2️⃣ A suitable dataset for machine learning
Data collection is not part of our ≤6-month project horizon.
Datasets don't need to be big per se, but they do need to be representative of the phenomenon.
Knowledge of sampling biases and other dataset limitations is very helpful.
Knowledge of sampling biases and other dataset limitations is very helpful.
► 3️⃣ One or more researchers with domain knowledge and enthusiasm to work collaboratively
To facilitate skills transfer, at least one collaborator from your team should have some programming skills.
► 1️⃣ A quick evaluation of suitability: Can ML really help you? If not, what is missing?
We tell you in days if we have capacities for an evaluation.
After an evaluation (1-2 weeks) we move directly into the execution phase or make concrete recommendations. For example: ♦ if the project needs go beyond the scope of our resources, we connect you to key partners from the ML community in Tübingen and translate your needs to them. You take it from there with collaboration arrangements, joint applications for funding, etc ♦ or we tell you if more or other data is needed.
► 2️⃣ A close, iterative collaboration for 1-5 months, "batteries included"
We select, implement and run state-of-the-art ML algorithms on your dataset.
All computations happen in our own ML Cloud - free of charge.
Work starts shortly after evaluation; we don't prebook capacities.
► 3️⃣ Code, ideas, figures and writing for publications
We stand for open science. Our deliverables, once released, are under a free content license and, by extension, are open access.
You receive a (private) final report of activities including negative results ("what did not work"), guidance for further ML work in your group, and suggestions for further work (that can be used to apply for follow-ups)-