Torch-in-DIGITS currently supports SGD, NESTEROV and ADAGRAD through these switches though I don't recall testing them explicitly. RMSPROP, ADADELTA and ADAM could also be added with similar switches.

Torch-in-DIGITS currently supports SGD, NESTEROV and ADAGRAD through these switches

Oh, don't know how I missed that!

Looks like all of these solver types are supported for Torch (am I looking in the right place)? I've added the six solvers that Caffe now supports for Torch, but I don't know enough about the other solver types to feel comfortable adding them at the moment.

Yes that's correct, Torch support a bunch of optimization methods through the optim package. Let's stick to those six solvers. I'll try your new patch though I'll be at a loss to verify the solver type is producing the desired effect.

I just checked that the optimization method was passed to the Lua wrapper and that this produced slightly different learning curves.

Me too. Let's merge and address any solver-specific improvements in future PRs.