I'm using an AdamOptimizer, and I compute the gradients, but before applying the descent step, I scale (i.e.: multiply) gradients with constants, to mimic having a different learning rate per parameter instead. Is this equivalent to having a learning rate per parameter?

Staring at the formulas of Adam, I believe they might be the same, although it have the impression my results are not the same, when using multiple optimizers, each with their own learning-rate parameter.


Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.