MATLAB Answers

how do I set my fully connected layer to be (3 or 2) classification output?

1 view (last 30 days)
Afiq Amirudin
Afiq Amirudin on 16 Jun 2021
Answered: Vineet Joshi on 19 Jul 2021
Im working with pretrained network.
Currently, I have 3 age group (17-20, 21-40, 41-60) and another one is (female , male). My question is how to change the fully connected layer for this type of classification.
Currently I had 6 classes, 3 age group for each gender. Therefore, my classification output is 6 classes instead. Is this the correct way to classify?

Answers (1)

Vineet Joshi
Vineet Joshi on 19 Jul 2021
Hi
It is not possible to have one layer with arbitary number of neurons (3 or 2) . Keeping this in mind, there are three ways to approach your problem.
  1. Create 6 labels (3 * 2) and train the network for classification on this problem. Since the number of instances are equal, you should get decent result but ensure to not let the model overfit.
  2. You can have common layers upto a point and then split the network into two halves, one with output of 2 and other with output of 3. You can refer to the following example. Assemble Multiple-Output Network for Prediction
  3. You can also have two seperate networks for both the predictions, but this is just the brute force way of doing the above.
Hope this was of some help.
Thanks

Community Treasure Hunt

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

Start Hunting!