training_policies icon indicating copy to clipboard operation
training_policies copied to clipboard

Minimize hyper parameter stealing post submission

Open jonathan-cohen-nvidia opened this issue 5 years ago • 0 comments

We don't know how to actually achieve this, but archiving for the purpose of generating a committee discussion. We would prefer a process where hyper parameter selections are shared prior to submission so that there is no (or little) post-submission re-running with stolen hyper parameters.

Perhaps declaring 1 week prior to submission intended hyper parameter ranges or expected values might work, although that does reveal scale which submitters might want to hold back.

jonathan-cohen-nvidia avatar Jul 23 '19 23:07 jonathan-cohen-nvidia