So I realize this is likely a niche use case, but is it possible to construct a custom loss where inside that loss function, I can access the current value of the simple chain parameters, and even call the simple chain itself?
In principle, I don't see why this shouldn't work(?), but the custom losses seem to be defined only to take the output of the previous layer as arguments.
With the L1/L2Penalty functions, it seems like accessing parameters within the loss can be made to work, but can the simplechain itself be called using these parameters?