Re:Evolution Strategies as a Scalable Alternative to Reinfor...
by argonaut on Sunday March 26, 2017 @12:06AM
(#13959082)
The Deep PCA paper shows no such thing. They've cherry-picked bad CNN baselines. The red flag for me is that the "state of the art" methods they compare to achieve > 1% test error on vanilla MNIST, which is totally wrong since CNNs routinely achieve <0.5 % test error (even by the standards of 2014, when the paper was published).
[ Parent
| on HN ]
There are no comments on this comment yet.