Are we allowed to modify loss function completely or can we only tune it’s parameters (beta, wx, wy)?
Thank you for question.
Unfortunately, it is not allowed to completely overhaul the loss function. But the parameters of the loss function, such as beta, wx, wy, provide a way to fine-tune the behavior and characteristics of the loss function within a defined framework.
Are we allowed to use torch.nn.MSELoss or any other losses with or instead of torch.nn.L1Loss ?
Unfortunately, it is not allowed as well. It is supposed to tune loss function by the parameters beta, wx, wy.
So basically, the only thing we have influence on are the hyperparameters, the initial model parameters and the exact shape of Decoder and Encoder, is that right?
Thank you for clarifying and confirming.