6367Re: DDFA Source
- Jun 13, 2014Hi Ken,
One point that was unclear to me from the paper, was how you propose to implement features discovered vertically, rather than simply one layer thick horizontally. While layering one layer at a time, using outputs from the previous layer as the new inputs is certainly viable, it appears to preclude re-insertion of prior learned features, as it were, to use "larger-building-blocks" to build new higher level building blocks. This would seem to require an update to the network infrastructure. I am curious as to your thoughts and visions going forward.
- << Previous post in topic Next post in topic >>