Jul 25, 2017 · 1 min read
I think #11 a.k.a. the generalization gap has been debunked by this paper. When the data shows a high variance in intensities very small batches can sometimes even hurt performance. very good article!
I think #11 a.k.a. the generalization gap has been debunked by this paper. When the data shows a high variance in intensities very small batches can sometimes even hurt performance. very good article!