tf.contrib.layers.batch_norm(
    inputs,
    decay=0.999,
    center=True,
    scale=False,
    epsilon=0.001,
    activation_fn=None,
    param_initializers=None,
    param_regularizers=None,
    updates_collections=tf.GraphKeys.UPDATE_OPS,
    is_training=True,
    reuse=None,
    variables_collections=None,
    outputs_collections=None,
    trainable=True,
    batch_weights=None,
    fused=None,
    data_format=DATA_FORMAT_NHWC,
    zero_debias_moving_mean=False,
    scope=None,
    renorm=False,
    renorm_clipping=None,
    renorm_decay=0.99,
    adjustment=None
)

"Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift"

Sergey Ioffe, Christian Szegedy

Batch Normalization通过减少内部协变量加速神经网络的训练。

可以用作conv2d和fully_connected的标准化函数。

参数：

1 inputs： 输入

2 decay ：衰减系数。合适的衰减系数值接近1.0,特别是含多个9的值：0.999,0.99,0.9。如果训练集表现很好而验证/测试集表现得不好，选择

小的系数（推荐使用0.9）。如果想要提高稳定性，zero_debias_moving_mean设为True

3 center：如果为True，有beta偏移量；如果为False，无beta偏移量

4 scale：如果为True，则乘以gamma。如果为False，gamma则不使用。当下一层是线性的时（例如nn.relu），由于缩放可以由下一层完成，

所以可以禁用该层。

5 epsilon：避免被零除

6 activation_fn：用于激活，默认为线性激活函数

7 param_initializers ： beta, gamma, moving mean and moving variance的优化初始化

8 param_regularizers ： beta and gamma正则化优化

9 updates_collections ：Collections来收集计算的更新操作。updates_ops需要使用train_op来执行。如果为None，则会添加控件依赖项以

确保更新已计算到位。

10 is_training:图层是否处于训练模式。在训练模式下，它将积累转入的统计量moving_mean并 moving_variance使用给定的指数移动平均值 decay。当它不是在训练模式，那么它将使用的数值moving_mean和moving_variance。
11 scope：可选范围variable_scope

注意：训练时，需要更新moving_mean和moving_variance。默认情况下，更新操作被放入tf.GraphKeys.UPDATE_OPS，所以需要添加它们作为依赖项train_op。例如：

  update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)  with tf.control_dependencies(update_ops):    train_op = optimizer.minimize(loss)

可以将updates_collections = None设置为强制更新，但可能会导致速度损失，尤其是在分布式设置中。

返回 该操作的输出API：https://tensorflow.google.cn/api_docs/python/tf/contrib/layers/batch_norm

 

minist例子：

import numpy as np
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data


# define our typical fully-connected + batch normalization + nonlinearity set-up
def dense(x, size, scope):
    return tf.contrib.layers.fully_connected(x, size,
                                             activation_fn=None,
                                             scope=scope)


def dense_batch_relu(x, phase, scope):
    with tf.variable_scope(scope):
        h1 = tf.contrib.layers.fully_connected(x, 100,
                                               activation_fn=None,
                                               scope='dense')
        h2 = tf.contrib.layers.batch_norm(h1,
                                          center=True, scale=True,
                                          is_training=phase,
                                          scope='bn')
        return tf.nn.relu(h2, 'relu')


tf.reset_default_graph()
x = tf.placeholder('float32', (None, 784), name='x')
y = tf.placeholder('float32', (None, 10), name='y')
phase = tf.placeholder(tf.bool, name='phase')

h1 = dense_batch_relu(x, phase,'layer1')
h2 = dense_batch_relu(h1, phase, 'layer2')
logits = dense(h2, 10, 'logits')

with tf.name_scope('accuracy'):
    accuracy = tf.reduce_mean(tf.cast(
            tf.equal(tf.argmax(y, 1), tf.argmax(logits, 1)),
            'float32'))

with tf.name_scope('loss'):
    loss = tf.reduce_mean(
        tf.nn.softmax_cross_entropy_with_logits(logits=logits, labels=y))


def train(mnist):
    update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
    with tf.control_dependencies(update_ops):
        # Ensures that we execute the update_ops before performing the train_step
        train_step = tf.train.GradientDescentOptimizer(0.01).minimize(loss)
    sess = tf.Session()
    sess.run(tf.global_variables_initializer())

    history = []
    iterep = 500
    for i in range(iterep * 30):
        x_train, y_train = mnist.train.next_batch(100)
        sess.run(train_step,
                 feed_dict={'x:0': x_train,
                            'y:0': y_train,
                            'phase:0': 1})
        if (i + 1) % iterep == 0:
            epoch = (i + 1)/iterep
            tr = sess.run([loss, accuracy],
                          feed_dict={'x:0': mnist.train.images,
                                     'y:0': mnist.train.labels,
                                     'phase:0': 1})
            t = sess.run([loss, accuracy],
                         feed_dict={'x:0': mnist.test.images,
                                    'y:0': mnist.test.labels,
                                    'phase:0': 0})
            history += [[epoch] + tr + t]
            print(history[-1])
    return history


def main(argv=None):
    mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)
    train(mnist)


if __name__ == '__main__':
    tf.app.run()