I am doing a project on EEG signal classification using RNN-LSTM network. I like to know how can we decide the Loss function and Optimizer for EEG classification? Is there any special criteria or anything for this.
There is no a priori rule, but in the following recent paper it seems that they are achieving good performance results by using a combined cnn+ltsm model with a center loss function:
https://ieeexplore.ieee.org/document/8377524/
With respect to the optimizer, I would go for the adam optimizer (a mixed of Adagrid and Rsmprop).