Expand criterion pool to support various types of losses

Currently the `build_criterion` function in `train.py` only supports CrossEntropy loss, this needs to be expanded to support additional loss options such as MSE, Binary Entropy etc. As for a temporary solution, frontend will only display CrossEntropy loss for the default option until this feature upgrade is done.