AI-Lab is a platform run jointly by several institutes of the University of Lübeck. It is designed to offer employees and students an environment in which research, learning and teaching of AI-centered topics can be performed. It offers a coding environment for preparing training/testing data and building up various AI-Structures as well as dedicated hardware to perform the necessary computations.

Using AI-Lab, you can write Notebooks to create your own AI-creation workflow; prepare and enrich training data, define neural networks or other common AI algorithms using state-of-the-art deep learning libraries and utilize Kernels to compute the resulting AI model on high-end machines.

If you have more complex projects, you can use Pipelines to split your workflow into multiple sections you can reuse and combine.

Notebooks are designed as a coding environment with added human-readability in mind. Notebooks are separated into blocks: Code blocks contain the actual written code, while text blocks are treated as comment and can be enriched through markup language. This creates a document that interweaves the actual code you write with readable and understandable commentary and explanation.

Kernels are the background runtimes that interpret and calculate your notebooks. When you run a code block within your notebook, the code is sent to the Kernel, interpreted and the result is returned.

A Kernel can have just a basic programming language interpreter, or it can come pre-packed with existing libraries for you to use for various purposes.

Currently Available Kernels

• iPython-Kernel: A blank python interpreter.
• PyTorch: Based on the iPython-Kernel, this Kernel adds the PyTorch AI-Library.
• TensorFlow: Similarly to PyTorch, this Kernel adds the TensorFlow AI-Library to the Python interpreter.
• R: An interpreter for the R coding language.

As a newly added feature, AI-Lab supports the use of Pipelines through Elyra and KubeFlow. This allows you to create multiple scripts and notebooks and pipe the results of one into the next for a more organized and compartmentalized workflow, as well as more efficient resource usage.