- Testing a bAbI dataset trained model using input()
- sent eth to vendor, but it was recieved as internal transaction and order didnt go through
- transferFrom call failure (gas estimation failed)
- How to add my token to external exchange
- Enterprise Version of Ethereum / Private Blockchains / Consortium Blockchains - How Malicious Actors are deterred?
- Using 3rd party libraries in production
- Making microPayment possible
- How can I stop the flare out of the bottom layers?
- Am I responsible for my spouse's careless behavior?
- Запятые при сочетании вводных - с союзами и частицами
- What is the name of this vest that some Eurodance artists wear such as Haddaway in the “What Is Love” music video?
- How can I turn home gas (LPG) from liquid to gas?
- Why does defvar scoping work differently without an initvalue?
- Reasons to voluntarily change into a fantasy race?
- reasons to have a multi-species crew for a zoo
- Shaivism or Shivaism?
- What is the meaning of Life if all is pre-written or controlled by the will of God?
- What was the reason behind the clash between Vashishtha and Vishwamitra?
- What is $source and $sources in BaseRelationField?
- You don't have to stand over there!
Right way to Fine Tune - Train a fully connected layer as a separate step
I'm using Fine Tuning with caffenet and it works really well but then I read this in Keras blog entry on Fine Tuning (They use a trained VGG16 model):
"in order to perform fine-tuning, all layers should start with properly trained weights:
for instance you should not slap a randomly initialized fully-connected network on top of a pre-trained convolutional base.
This is because the large gradient updates triggered by the randomly initialized weights would wreck the learned weights in the convolutional base.
In our case this is why we first train the top-level classifier, and only then start fine-tuning convolutional weights alongside it."
So as a separate step in Fine tuning they save the output of the last layer before the fully connected layer (the "bottleneck features") and then they train a "small fully-connected model" on those features and only then they put the newly trained fully connected layer on top of the whole net and train the "last convolutional block".