Hippocampus project: towards collaborative AI-driven research

Are you working on AI and would you like to combine datasets with models and computing power? SURF is transforming the OpenML platform in collaboration with TU Eindhoven, Leiden University and other partners. The Hippocampus project is developing a next-generation open science infrastructure. The aim is to facilitate collaborative AI-driven research with powerful computing resources.

This is one of the 45 projects designed to strengthen the infrastructure for open science in the Netherlands, for which Open Science NL has awarded over 35 million in grants before. The Hippocampus project will run from 2026 to 2030.

The project is named after the hippocampus: the part of the brain responsible for organising and storing memories. Just as this part of the brain structures memories, this platform will systematically store and organise the precise details of AI datasets, computer models and evaluations.

The aim of Hippocampus

The main aim is to transform OpenML, an existing open platform for creating and sharing datasets and AI models, into an advanced collaborative infrastructure for AI-driven research. Here, users can seamlessly execute tasks (e.g. model training, benchmarking) from OpenML on any national, Tier-2, or local computing facilities in an easy user interface. The platform also promotes collaboration, encourages the responsible use of AI and connects researchers to foster open, reliable and AI-driven research.

Key results of Hippocampus

Connecting infrastructures

The platform links major Dutch and international data and computing infrastructures, such as those of SURF, with one another. This makes it easier for researchers, regardless of their discipline or location, to build AI models.

Reusability and reliability

By accurately recording metadata and experiments, AI research becomes reproducible, reusable and verifiable.

Cross-disciplinary

The collaborative infrastructure is designed to help researchers from all scientific fields – from biology to sociology – to use AI effectively and responsibly.

The role of SURF

The link between the Hippocampus project and SURF centres on bringing together data storage, computing power and machine learning workflows into a single, seamless infrastructure. The collaboration works as follows:

1. Direct access to supercomputers and computing power

Challenge: Training AI models requires enormous computing power (such as GPUs). Scientists currently often have to manually transfer code from their test environment to a supercomputer.
Solution: OpenML, the platform behind Hippocampus, will be directly connected to SURF’s national computing infrastructure. Researchers will soon be able to submit computational tasks directly to SURF’s powerful servers via the familiar OpenML interface, without the need for complicated manual intermediate steps. Moreover, we will publish this computational backend API as an open-source standard, so any other software platform can connect to a variety of computational facilities using this standard.

2. Integration via Single Sign-On

Challenge: Researchers often have to create a separate account for each platform, which slows down collaboration across institutions.
Solution: With SURF Research Access Management (SRAM), researchers from all Dutch and affiliated international universities can log in to the Hippocampus platform. They can do so securely using their own institutional account, enabling them to make immediate use of the shared data and models.

3. Standardised AI workflows

Challenge: There are many fragmented AI services.
Solution: SURF is building a central AI hub to standardise AI services. The Hippocampus project ties in with this by ensuring that AI models, datasets and evaluations running via the platform comply with open science guidelines. SURF provides the technical infrastructure, whilst the Hippocampus team optimises the application layer (OpenML) for scientific use.

Learn more?

Do you have any questions?

Please contact Yue Zhao via yue.zhao@surf.nl.