A better loss function to train with

[aminzabardast]

What is the preferred loss?

[aminzabardast]

the optimization issue is a regression type optimization issue. Proposed loss functions are:

• Mean Squared Error (MSE)
• Mean Absolute Error (MAE)
• Logcosh
• huber

According to an article by Prince Grover, Logcosh has the advantage of behaving like L1error (MAE) in large error values and behaving like L2 error (MSE) around small error values which makes it a very good choice.