Real physical systems as a testbed for AI methodology

If you do basic research in AI & ML, it can be a struggle to find real-world datasets to validate your work. Often, the best we can do is test new methods on synthetic data, making it difficult to asses core assumptions and predict how well an algorithm would work in practice.

The Causal Chambers are here to help. As computer-controlled laboratories built around well-understood physical systems, they provide real-world datasets with a ground truth for fields where such datasets are otherwise scarce or non-existent.

Access the Chambers

The chambers are a tool for educators and scientists who do basic research in AI & ML. Depending on your needs, there are several ways you can access the chambers and their data.

Dataset repository

Fully documented, open-source datasets collected from the chambers. Updated regularly with new experiments & benchmarks.

Please reach out if you need help navigating the repository.

Custom Datasets

Do you have a use case in mind but the appropriate dataset is not yet on the repository?

We can collect custom datasets on request. We are also happy to help you design new benchmarks and collaborate on scientific research.

Own a Chamber

Collect your own datasets with full, uninterrupted access to the chambers. For applications in active learning, RL & control, etc. For conference competitions, teaching and demonstrations.

Manufactured in Switzerland
1-year warranty
Full documentation
Set-up support

Research

Research papers that use chamber data

The Causal Chambers: Real Physical Systems as a Testbed for AI Methodology
Juan L. Gamella, Jonas Peters and Peter Bühlmann
Sortability of Time Series Data
Christopher Lohse and Jonas Wahl
CI4TS Workshop @ UAI 2024 
An Asymptotically Optimal Coordinate Descent Algorithm for Learning Bayesian Networks from Gaussian Models
Tong Xu, Armeen Taeb, Simge Kuccukyavuz, Ali Shojaie

arXiv preprint arXiv:2408.11977

Causality Pursuit from Heterogeneous Environments via Neural Adversarial Invariance Learning
Yihong Gu, Cong Fang, Peter Bühlmann and Jianqing Fan
Addressing Misspecification in Simulation-based Inference through Data-driven Calibration
Antoine Wehenkel, Juan L. Gamella, Ozan Sener, Jens Behrmann, Guillermo Sapiro, Marco Cuturi and Jörn-Henrik Jacobsen

Contact

Do you need help navigating the dataset repository? Do you have an application or case study in mind? Would you or your research group like to own a chamber?

Please reach out via email! We’re happy to help.

 
Causality Pursuit from Heterogeneous Environments via Neural Adversarial Invariance Learning
Yihong Gu, Cong Fang, Peter Bühlmann and Jianqing Fan
Addressing Misspecification in Simulation-based Inference through Data-driven Calibration
Antoine Wehenkel, Juan L. Gamella, Ozan Sener, Jens Behrmann, Guillermo Sapiro, Marco Cuturi and Jörn-Henrik Jacobsen