ConvOptDL

Training

juliap train.jl --batches-per-episode=<INTEGER> --batch-size=<INTEGER> --num-episodes=<INTEGER> --log-period=<INTEGER> <DATA_FILE.jls>

ex: juliap train.jl --batches-per-episode=400 --batch-size=20 --num-episodes=20 --log-period=20 CIFAR_FS-train.jls

GPU support
implement Block LU factorization in the website below for better performance
- https://locuslab.github.io/qpth/
better QP solver: primal-dual interior point method
- As mentioned in Mattingley & Boyd (2012)