3

I have a dataset I would like to publish - it is likely people will want to use this dataset to train machine learning models. I would like to ensure that any models trained using this dataset are freely available, and perhaps (if possible) further restrict that the code used to train these models is published. Similar to a GPL license or CC-BY-SA. Is there a license I can publish these data under that will do this?

1
  • Open-Source licenses generally only add permissions that recipients wouldn't have by default under copyright law (“all rights reserved”). But these licenses can't really add extra restrictions. Some jurisdictions allow data mining (incl model training) without being bound by licenses or copyright.
    – amon
    Feb 16, 2022 at 22:07

1 Answer 1

4

The intention you describe in the question is very copyleft. Not only the data and derivative works (modified data) should remain under that license, but also the models that are trained using the data, and also the code which is used to for training the models using your data.

Your attempt to force as much as possible (basically anything that touches your data) into the open-source space is risky and might backfire. The headwind comes from license-incompatibility between your data license and on the other side the license of the models, which might prevent that your data is getting used at all.

Furthermore, your license for data might not be suitable for models, which are more like software/code, so you would need to decide which (copyleft?) license you would want to force upon the code of the models. Again this is something users will not welcome and where license incompatibilities will strike.

I am not aware of any license that has the features you are proposing. The licenses commonly used for data are Creative Commons and Open Data Commons.

FOSS licenses for software (such as GPL) are not suitable, as the are drafted to protect code, not data. The CC BY-SA and the Open Data Commons Open Database License (ODbL) are as far as you can get with currently available (commonly used) open source licenses. If you want to get more you might need to contact a lawyer to draft a proprietary license for you.

There is a similar question, which you might want to consider for further reading.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.