Glossary
Plain-English definitions of the terms used across our licensing guides.
Rights
- Chain of title
- The documented sequence of ownership or license grants that shows you can license a work or dataset.
- Read the guide on chain of title
- Compilation copyright
- Protection for original selection or arrangement of facts; the facts themselves remain free (Feist).
- Read the guide on compilation copyright
- Work made for hire
- A work created by an employee within the scope of employment, or certain commissioned works under a written agreement; owned by the employer.
- Read the guide on work made for hire
Privacy
- De-identification
- Removing or transforming identifiers so data no longer reasonably identifies a person; HIPAA defines two methods for health data.
- Read the guide on de-identification
- Biometric identifier
- Data such as face geometry or voiceprints; regulated by state laws like Illinois BIPA.
- Read the guide on biometric identifier
License terms
- Field of use
- The contractual scope of permitted purposes for licensed data.
- Read the guide on field of use
- Model survival
- Whether a model trained on licensed data may keep being used after the license ends.
- Read the guide on model survival
- Clean room
- An arrangement where a licensee computes on data inside the owner's controlled environment without taking a copy.
Data & models
- Data provenance
- The documented origin, lineage and processing history of data.
- Read the guide on data provenance
- Datasheet
- A structured description of a dataset's motivation, composition, collection and intended uses.
- Read the guide on datasheet
- Fine-tuning
- Additional training of an existing model on a smaller, purpose-built dataset.
- Read the guide on fine-tuning
- Contamination
- When evaluation data has leaked into training data, inflating measured performance.
- Read the guide on contamination