1
$\begingroup$

I'm trying to build a network. While creating the fully connected part in the last, Which one should we prefer: More layers that regularly reduce with heavy dropouts or fewer layers that reduce drastically with light dropouts?

    x = Dense(1000, activation='relu')(image_features)
    x = Dropout(0.8)(x)  
    x = Dense(500, activation='relu')(x)
    x = Dropout(0.6)(x)  
    x = Dense(200, activation='relu')(x)
    x = Dropout(0.5)(x)  
    x = Dense(100, activation='relu')(x)

or

    x = Dense(500, activation='relu')(image_features)
    x = Dropout(0.4)(x)  
    x = Dense(100, activation='relu')(x)
    x = Dropout(0.4)(x)  
$\endgroup$
1
  • $\begingroup$ It's nearly impossible to predict which one would do better on your problem and data. $\endgroup$
    – gunes
    Oct 2, 2022 at 6:25

1 Answer 1

0
$\begingroup$

Your question does not have a good answer, because there's no hard science behind neural network architectures. The usual answer would be to try both and see which one works better. A more practical one would be to search for examples of people solving similar problems in the past and try similar solutions to theirs first.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.