Question about the training process

[umsimonchen]

Thanks for your extraordinary contribution to the forward-forward algorithm and graph-based tasks.

I have a question about the training process. It said that the parameters are updated layer by layer. In other words, only one layer is being trained at each time.

Does it mean that the model is NOT an end-to-end model? If so, is it possible to turn it into an end-to-end one?

I am trying to apply the ForwardGNN to graph recommender systems. Though the performance is comparable, I have no idea how to make it more efficient and compatible since the forward-forward algorithm is similar to other pre-trained models in the recommendation domain.

I'd appreciate it if you could share your precious opinion :D. Best regards.