1.3 What's the difference between Federated Validation (FV) and Federated Learning (FL)?

Federated Validation (FV) is a system designed to "validate" the trained AI model performance, by using the diverse datasets in different sites.

Federated Learning (FL) is a framework designed to "train" an AI model with more datasets which located in different sites.

Both have the same concept, which is due to the privacy, security and legality concerns, the datasets will be kept in the original places, without disclosing and gathering them together, to complete the AI training or validation tasks.

For AI model that already trained but want to adjust the model weighting by traversing more data, Federated Learning (FL) is designed to conquer this task.

If the AI model is well trained and would like to verify it's accuracy and other performance, Federated Validation (FV) could help in the real world.

Federated Learning (FL) and Federated Validation (FV) could be used in turn.

After a AI model is tuned by FL progress, it could be verified by FV progress afterwards. If the FV result is not perfect as expected, another FL progress could be initialed again.

Taiwan AI Labs Federated Learning and Federated Validation systems share the code-base. Which means with similar operations, you could run FL and FV iteratively. Leveraging the AI model inference tools offered by Taiwan AI Lab, you could create a mechanism while using the AI model to enhance productivity, and pruning it by using FL, validating it by using FV, which keeps your system better and better with more and more perspective data.