I need to improve my model of Convolutional Neural Network (CNN). The goal is to recognize facial expression. I've been using some strategies like dropout for regularization and Adam optimazer, but i can get a good accuracy. Therefore I'm trying to implement PCA.

I use to filters for convolution, max pooling after each one and relu for activation. The highest accuracy i could get was 0.6.

I am working with Tensorflow. How can I do that?

Or there is another way to implemente pre-training to a CNN.

Thanks in Advance

  • $\begingroup$ Hi. Regarding PCA you can implement it in TF using tf.linalg.eigh function as eigen decomposition of $X^TX$ or maybe use tf.linalg.svd instead. However, your question is quite broad as we don't know what kind of network, data you are using or even what is the task. It could be that the problem is somewhere else and you don't require PCA to solve it. $\endgroup$ – jojek Jul 15 '20 at 15:28
  • $\begingroup$ I am using a Convolutional Neural Network for facial expression recognition. I used to filter and max pooling after each convolution. But i can't get a good accuracy therefor i want to implement unsupervised pretraining @jojek♦ $\endgroup$ – German_7 Jul 16 '20 at 1:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.