CODEWITHSUNDEEP
neural-networkconv-neural-network
marz
marz

Reputation: 973

Is theres a relationship between the number of filters in a convolution and the number of epochs during training?

What happens in a CNN with a single Conv Layer with 10 filters if I train the network with 50 epochs?

Isn´t a filter optimized 50 times like having 50 filters? Or 10 filters are optimized 50 times?

If its the second case, which of the 10 filters are used to make the prediction?

Upvotes: 0

Views: 46

Answers (1)

Jacob Soderlund
Jacob Soderlund

Reputation: 329

Each training epoch optimises all 50 of the filters slightly. Each epoch alone makes only small changes to the filter weights, so we need to repeat the training process many times before we have a model that does a good job.

As for which filters are used to make the prediction, all of them! The final fully connected layer takes inputs from all (10 * width * height * channel) conv layer outputs and uses them to make its prediction.

You might want to review how CNNs are structured - this article does a good job of explaining them: https://ujjwalkarn.me/2016/08/11/intuitive-explanation-convnets/

Upvotes: 1

Related Questions