Fully Connected Layers Deep Neural Network for extracted features of pre-trained models
2 次查看(过去 30 天)
显示 更早的评论
After Extracting feature vectors from pre-trained models, I would like to use those produced features to train a deep neural network with a number of fully connected layers. This example demonstrates doing just that using a multi-class SVM:
However, this method does not work with huge amount of data as mini-batching is not supported as far as I know. That is why I want to use the TrainNetwork function the same way we deal with image classification problems. Unfortunately it is indicated in the documentation that it only supports image and sequence classification and regression problems.
On the other hand, the Train function is assumed to deal with numeric data arrays, which seems to be useful for training and classifying the produced feature vectors. However, it still does not support mini-batching to deal with huge number of feature vectors. Moreover, it crashes very soon even when using a subset of the feature vectors if I set the number of hidden units above 20. Beside that, it seems to divide the training data into training and validation as it does validation checks during training. And I really want that to happen.
I know it is very easy to do that in platforms such as TensorFlow, by building the layers and performing the mini-batching manually. However, I would like to do this using the simple Matlab TrainNetwork way of defining the layers and options and then visualizing the training and perform the predictions.
Thank you for your help and support.
Yours, Maad
0 个评论
回答(0 个)
另请参阅
类别
在 Help Center 和 File Exchange 中查找有关 Image Data Workflows 的更多信息
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!