A very hacky (incomplete - e.g. there's no backpropagation) implementation of a depedently-typed neural net.
Written for COMSM0018 Applied Deep Learning at the University of Bristol. See the presentation I gave.
Forked from Soham Chowdhury (mrkgnao)'s Idris implementation of Justin Le (mstksg)'s dependently-typed neural network in Haskell.
Also heavily inspired by Huw Campbell's Grenade library.