Caffe: how to use dropout layer?

Created on 7 Sep 2014  ·  3Comments  ·  Source: BVLC/caffe

to my understanding, dropout is applied to full-connected layer,
how can i use dropout layer in caffe? is there any demo about this?
shall i use it on top of a ip layer?or anything else?

question

Most helpful comment

The dropout layer reduces overfitting preventing complex co-adaptations on the training data. Here I provided an example that takes the output of an InnerProduct layer (ip11), after an ReLU layer as an activation function.

layer {
name: "drop1"
type: "Dropout"
bottom: "ip11"
top: "ip11"
dropout_param {
dropout_ratio: 0.5
}
}

In this layer each hidden unit is randomly omitted from the network with a probability of 0.5 (the dropout ratio)

All 3 comments

You can simple read the definition files included in examples/imagenet. There are two dropout layers in them.

Right. Examples are your friends. Please ask on caffe-users. Issues are for development discussion. Thanks!

The dropout layer reduces overfitting preventing complex co-adaptations on the training data. Here I provided an example that takes the output of an InnerProduct layer (ip11), after an ReLU layer as an activation function.

layer {
name: "drop1"
type: "Dropout"
bottom: "ip11"
top: "ip11"
dropout_param {
dropout_ratio: 0.5
}
}

In this layer each hidden unit is randomly omitted from the network with a probability of 0.5 (the dropout ratio)

Was this page helpful?
0 / 5 - 0 ratings

Related issues

prathmeshrmadhu picture prathmeshrmadhu  ·  3Comments

greatgao picture greatgao  ·  3Comments

malreddysid picture malreddysid  ·  3Comments

iamhankai picture iamhankai  ·  3Comments

FreakTheMighty picture FreakTheMighty  ·  3Comments