What if we made a neural network where we take the distribution of activations on an intermediate layer and sample from it, the way we do in the final softmax layer? What would that do that everything?
This isn't Peter West's idea.
If you are inspired by this idea, you can reach out to the authors for collaboration or cite it:
@misc{holtzman-sampling-neural-network-2026,
author = {Holtzman, Ari},
title = {Sampling Neural Network},
year = {2026},
url = {https://hypogenic.ai/ideahub/idea/PkQXdPYqSW5P0rNfuh05}
}Please sign in to comment on this idea.
No comments yet. Be the first to share your thoughts!