jwallwork23 / nn_adapt

Accelerating goal-oriented error estimation using neural networks
1 stars 2 forks source link

Allow more general networks #3

Open jwallwork23 opened 2 years ago

jwallwork23 commented 2 years ago

At the minute, we only support fully connected feed-forward networks with a single hidden layer. It would be good to build on this.

See the TODO in this commit.

jwallwork23 commented 2 years ago

For example, perhaps we could use convolutional neural networks or graph neural networks.

jwallwork23 commented 2 years ago

Note #14 included some small steps in this direction.