I have a deep neural network, and I am trying to use it to classify signed numbers in 64×64 pcitures of numbers from 0-9. The cost seems to always converge around 2.3, which is 10% accuracy or just random.

Training loss(blue) Validation(orange)

I am trying to solve the following 2-equation system (PDE) for U (output) and V (output) where x(1D space) and t(time)is the input. I have used DeepXDE but the model can’t decrease the loss but rather increases to NAN. The boundary condition for U is 1+0.01sin(2pifreqt) at x = 0. The initial condition (t = 0) for U is 1 and V is 0. Could you please kindly give me any suggestions on where should I look into? Thank you. I have attached the screenshot of the code.

I am trying to solve coupled three PDEs using a DeepXDE PINN model. The Three PDEs are coupled and dependent on each other. I would like to train this model on big time intervals because I would like to make prediction at the end at time = 1E7 (s). Here is my code: