25 ms·
No timing results, no comparisons with Nesterov type methods. To all the commenters that have said "this looks promising": this doesn't look promising at all.
by oiuytrewq 10y ago
No timing results, no comparisons with Nesterov type methods. To all the commenters that have said "this looks promising": this doesn't look promising at all. Why do you think in all the years of people optimizing things with gradient descent no one has tried this? Answer: they have, and it doesn't work.
- thanatropism 10y agoTiming results are a poor substitute to O(N) complexity estimates. I mean, does a method take longer because it's doing lots of virtual memory stuff or because it uses a lot of computron?