The mutation method is actually fine . You can use Continuous Gray Code Optimization. It works very well for federated learning using short sparse mutation lists and gives better generalization than BP.
@cblackall213 жыл бұрын
[19:10] "Putting neural nets in your [software] system, it's a bit like getting gangrene; it will gradually eat the whole system.""
@hmthanhgm Жыл бұрын
Geoff Hinton is legendary
@rb8049 Жыл бұрын
So much intuition.
@hoaxuan70743 жыл бұрын
Eh, ReLU is a switch. Fast Transform fixed-filter-bank neural networks AI462.
@shlim763 жыл бұрын
I feel like watching a lecture on cable TV, where ads come in every 5 minutes.