zkp-gravity / 0g-halo2

Proof of inference of Weightless Neural Networks using Halo2
MIT License
6 stars 0 forks source link

DNN to select WNN Permutation #37

Open georgwiese opened 1 year ago

georgwiese commented 1 year ago

TBD on what level, but suggest how can DNN be used to select the best permutation & sampling size to fit the best for a particular task.

Also, answer the question if there is a Perfect training - where are we loosing on accuracy, and what/where more degrees of freedom are needed.

georgwiese commented 1 year ago

Not sure I understand this item!