3 ms·
You still need to train a model on what you're transferring from. It's not like you can manufacture trained models out of thin air.
by 1024core 10y ago
You still need to train a model on what you're transferring from. It's not like you can manufacture trained models out of thin air.
- AlexCoventry 10y agoI think the idea is that, for example, you start with some pre-trained visual object-identification network, remove the final logistic-regression layer, and replace with a layer which you train your problem on.