Implement a Persistent Identifier (PID) Generation Mechanism for Model Cards

[nee1k]

Our current workflow for model card submissions lacks a formal method for generating Persistent Identifiers (PIDs) for AI/ML models.

Proposed Approaches:

1. Hash-Based Generation:

   • Method: Compute a cryptographic hash (e.g., SHA-256) of key metadata fields (such as model name, version, author).
   • Pros:
     • Simple to implement.
     • No external dependencies.
     • Fast and deterministic.
   • Cons:
     • May lead to collisions if metadata isn’t sufficiently unique.
     • If model metadata changes, the PID would change unless carefully managed.

2. Dedicated PID Service:

   • Method: Integrate with a dedicated PID service or registry (e.g., Handle System, DOI) that assigns and manages unique identifiers.
   • Pros:
     • Provides a globally resolvable and persistent identifier.
     • Often comes with additional metadata management and resolution features.
     • Aligns well with FAIR principles by offering external resolvability.
   • Cons:
     • Increases system complexity and introduces external dependencies.
     • May incur costs or require administrative overhead for integration.

[nee1k]

The current approach is to generate the PID as "{author name}{model name}{version}.

'User ID generation' needs to be implemented in the future to avoid duplicate PIDs with same user names.

Registering a user would also help in securely sharing the repository credentials.