Some interesting numbers about AlphaGo

送交者: conner 于 2016-03-13, 21:42:52:

The input to the policy network is a 19x19xhttp://48 image stack. The policy network has 13 layers
That to the value network is 19x19x48x2, and it has 15 layers.

所有跟贴:

I suspect the policy network and/or the value network favors the center point - conner (278 bytes) 2016-03-13, 23:23:25 (803752)
- this is very good analysis - pgss (483 bytes) 2016-03-14, 00:04:37 (803758)
these are two deeplearning subsystems, it should also have a knowledge tree - 1FD7 (95 bytes) 2016-03-13, 22:26:17 (803747)
- If true, it is easy to understand its buggy performance in game 4 - 1FD7 (471 bytes) 2016-03-13, 22:58:18 (803750)
it is too far from a mammalian brain. (无内容) - cornbug (0 bytes) 2016-03-13, 21:56:56 (803743)

加跟贴