Implement higher derivatives of the NN

achiefa / NTK-interface

GNU General Public License v3.0

0 stars 0 forks source link

Implement higher derivatives of the NN #13

Open achiefa opened 1 month ago

achiefa commented 1 month ago

This PR introduces the machinery needed to compute higher derivatives of the NN up to the third order. These are required in the calculation of the observables related to the Neural Tangent Hierarchy (NTH). The PR also introduce a jupyter notebook to analyse the observable and plot them as functions of the dimension of the network.

achiefa commented 2 weeks ago

At this point, there are few things to do:

[ ] First of all, move Observable.h in ObservableInterface.hpp and ObservableInterface.cpp and make them private.
[ ] Decide whether the implementation of actual observable must be private or public.
[ ] Optimise the second-order numerical derivative