Converge of a PINN

25 vues (au cours des 30 derniers jours)
Giulio Vialetto
Giulio Vialetto le 21 Avr 2023
Modifié(e) : Dinesh le 25 Avr 2023
Dear all,
I was trying to define a PINN but the NN training do not converge, please any suggestion?

Réponses (1)

Dinesh
Dinesh le 25 Avr 2023
Modifié(e) : Dinesh le 25 Avr 2023
Hi Giulio,
I have gone through the script you have attached. I do not have any information on the “loss_proiettile” function so I could not run the code. But since the issue is with the convergence of the neural network try tuning some hyperparameters or data
  1. Try changing the initial learning rate to 0.1 instead of 0.01.
  2. Uncomment line number 113, time-based decay learning rates are great for convergence given you start from high learning rate.
  3. Try increasing the data change 'x = 0: 0.1: 40' to ' 0: 0.01 :40 '
  4. Finally, if nothing works try reducing the number of nodes in each layer of the neural network.
if none of the above give a satisfying result then the problem might be with the loss function that is responsible for updating the weights. Refer to research papers on convergence of PINN’s for more information on loss functions.
Hope this helps!

Catégories

En savoir plus sur Sequence and Numeric Feature Data Workflows dans Help Center et File Exchange

Produits


Version

R2019b

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by