The scripts can be executed within the pyadjoint docker image (http://www.dolfin-adjoint.org/en/latest/download/index.html)

fenicsproject run quay.io/dolfinadjoint/pyadjoint:master

Some ideas related to the project can be found in the notes.

• What does the learned operator look like/properties?
• Say we learn on one-few resolutions and fixed example (think bcs in Stokes); does that stab help in other resulutions/examples?
• Are we learning some stabilization?
• PDE cstr literature on problems we're looking at - one shot?

• Keras like spec of the NN model with possibility to set(init) weights, store/load model
• Stability testing