Transfer trained model use cases

[mohammadbashiri]

What are common use cases of this new trained model?

[mohammadbashiri]

I think in general usually some part of the model is transferred. Here are some use-cases I could think of:

1. stage 0: train model -> stage 1: use trained core in previous stage
2. stage 0: train model -> stage 1: continue training (some part of) the model with a different trainer
3. stage 0: train model -> stage 1: continue training (some part of) the model with a different dataset
4. what else?

[KonstantinWilleke]

I think that (1) can be extended to readout and parts of the readout, too. So positions/features only, for example.

[chbehrens]

Our usecase is stage 0: train full on a large dataset, stage 1: train only a new readout on a new smaller dataset during closed loop experiments. Another possible usecase would be to train one model on a certain dataset, then take the core, add another layer or two and train use it to train on data recorded further upstream.