Hi,

It is acceptable to use transfer learning as long:

1) The participant has full access to the model, e.g., the transfer-learnt model is their own or pubicly available. So no closed, third-party models available only through API.

2) It is not just taking an existing model as-is without any additional effort. A non-zero task-related effort is required e.g. fine-tuning. Even minimal fine-tuning is sufficient.