Artificial Intelligence By Example

Dropout and Layer 2

Dropouts and the designing of a layer were explained previously.

However, let us focus on some features of the dropout node: dropout keep probability and the outputs in the following diagram.

Dropout node

A dropout keep probability plays an overfitting reduction role. The probability leads to dropping neurons during the training process. The scaling is done automatically. Furthermore, this dropout feature is turned on and off during testing:

with tf.name_scope('dropout'):keep_prob = tf.placeholder(tf.float32)tf.summary.scalar('dropout_keep_probability', keep_prob)dropped = tf.nn.dropout(hidden1, keep_prob)

TensorBoard receives ...