Rimsha
Reputation: 3
Weights/cost after 'Unsupervised Pre-Training' of DBN with/without Dropout
Do weights/cost of each layer obtained after 'Unsupervised Pre-Training' of a Deep Belief Network differ when dropout is applied and when it isn't ?
I have obtained different sets of weights and different value of cost after pre training a DBN with and without applying Dropout. According to me, Dropping out units should affect only the fine tuning stage and not the Unsupervised pre training where all units are required and none is dropped. Am I getting the concept wrong or the code wrong?
Upvotes: 0
Views: 73
Answers (0)
Related Questions
- Handwritten Signature Verification
- Why do we need to preseve the "expected output" during dropout?
- Keras minibatch gradient descent with Dropout layer
- How to set weights for specific channels in keras with pre-trained weights?
- Why DNN with Dropout always predict one?
- TensorFlow object detection api: classification weights initialization when changing number of classes at training using pre-trained models
- Fine tuning weights in DBN
- How does pre-training improve classification in neural networks?
- Deep Belief Networks vs Convolutional Neural Networks