Introduction to transformer neural networks

When: 6 March 2025 13:00 - 17:00
Where: CIT Smitsborg 5431.0074 BlueGene, Nettelbosje 1 Groningen

Fast and dirty introduction to transformer neural networks

Although AI seems, and also is, complex, these algorithms do enable a lot of research, also for the non-computer scientist. Large language Models, like chatGPT, can now be used to answer questions about huge amounts of data, where previously manual labour was required, posing the problem in natural words rather than the need to write complicated algorithms that might not even be up to the task.

We’ll discuss how to apply such models, and how to validate your work. We’ll look at pre-trained models at huggingface.co that can be used off-the-shelf and provide examples on how they can be applied for research.

Who Should Attend?

Anyone who wants to try some AI and meet with peers.

Prerequisite knowledge

Python programming skills, understanding of (complex) algorithms not needed. Some mathematics, like basic calculus and linear algebra. If you don’t have these skills and expertises, you’re still welcome, but you’ll be engulfed and should only join knowing some things will be unclear.

Content

✅ Supervised learning and how models are validated
✅ Neural networks and gradient descent
✅ Transformer models and LLMs

Out of scope

In depth programming, debugging and practical skills for running AI experiments and understanding of model architectures. We’ll try to show you quickly and dirty what you can achieve with existing models.

Preparation

A laptop with a browser, Eduroam and a Google account, e.g. your UG-account. Please use Firefox, Chrome or Safari, since Microsoft Edge doesn’t always work. You’ll get a 10’ assignment in advance to check if everything works well and avoid teething issues.

You’ll work in an online python environment, Google Colab, that’s what the Google account is for. Please decide in advance whether you’ll use your private or your UG account, or a burner account if you’re concerned about privacy.

Result

You’ll have done some research with data science and AI and have gotten some basic insights about AI models.

Trainer

This course will be given by Herbert Kruitbosch.

Enrollment and course fee

Attendance is free for UG staff and PhD-students, but registration is required.

More information

• This course will be hosted by the Research Support Hub (expert workshops, support café and guided working).
• For more questions on the workshop you can mail the trainer Hertbert Kruitbosch (h.t.kruitbosch rug.nl).