machine-learning - 对于深度学习，使用激活 relu，输出在训练期间变为 NAN，而使用 tanh 则输出正常

Question

我训练的神经网络是深度强化学习的批评网络。问题是，当其中一层的激活设置为 relu 或 elu 时，经过某个训练步骤后，输出将为 nan，而如果激活为 tanh，则输出为正常。代码如下(基于tensorflow):

with tf.variable_scope('critic'):

        self.batch_size = tf.shape(self.tfs)[0]

        l_out_x = denseWN(x=self.tfs, name='l3', num_units=self.cell_size, nonlinearity=tf.nn.tanh, trainable=True,shape=[det*step*2, self.cell_size])

        l_out_x1 = denseWN(x=l_out_x, name='l3_1', num_units=32, trainable=True,nonlinearity=tf.nn.tanh, shape=[self.cell_size, 32])
        l_out_x2 = denseWN(x=l_out_x1, name='l3_2', num_units=32, trainable=True,nonlinearity=tf.nn.tanh,shape=[32, 32])
        l_out_x3 = denseWN(x=l_out_x2, name='l3_3', num_units=32, trainable=True,shape=[32, 32])

        self.v = denseWN(x=l_out_x3, name='l4', num_units=1,  trainable=True, shape=[32, 1])

这是基本层构建的代码:

def get_var_maybe_avg(var_name, ema,  trainable, shape):
    if var_name=='V':
        initializer = tf.contrib.layers.xavier_initializer()
        v = tf.get_variable(name=var_name, initializer=initializer, trainable=trainable, shape=shape)
    if var_name=='g':
        initializer = tf.constant_initializer(1.0)
        v = tf.get_variable(name=var_name, initializer=initializer, trainable=trainable, shape=[shape[-1]])
    if var_name=='b':
        initializer = tf.constant_initializer(0.1)
        v = tf.get_variable(name=var_name, initializer=initializer, trainable=trainable, shape=[shape[-1]])
    if ema is not None:
        v = ema.average(v)
    return v

def get_vars_maybe_avg(var_names, ema, trainable, shape):
    vars=[]
    for vn in var_names:
        vars.append(get_var_maybe_avg(vn, ema, trainable=trainable, shape=shape))
    return vars

def denseWN(x, name, num_units, trainable, shape, nonlinearity=None, ema=None, **kwargs):
    with tf.variable_scope(name):
        V, g, b = get_vars_maybe_avg(['V', 'g', 'b'], ema, trainable=trainable, shape=shape)
        x = tf.matmul(x, V)
        scaler = g/tf.sqrt(tf.reduce_sum(tf.square(V),[0]))
        x = tf.reshape(scaler,[1,num_units])*x + tf.reshape(b,[1,num_units])
        if nonlinearity is not None:
            x = nonlinearity(x)
        return x

这是训练网络的代码:

self.tfdc_r = tf.placeholder(tf.float32, [None, 1], 'discounted_r')
self.advantage = self.tfdc_r - self.v
l1_regularizer = tf.contrib.layers.l1_regularizer(scale=0.005, scope=None)
self.weights = tf.trainable_variables()
regularization_penalty_critic = tf.contrib.layers.apply_regularization(l1_regularizer, self.weights)
self.closs = tf.reduce_mean(tf.square(self.advantage))
self.optimizer = tf.train.RMSPropOptimizer(0.0001, 0.99, 0.0, 1e-6)
self.grads_and_vars = self.optimizer.compute_gradients(self.closs)
self.grads_and_vars = [[tf.clip_by_norm(grad,5), var] for grad, var in self.grads_and_vars if grad is not None]
self.ctrain_op = self.optimizer.apply_gradients(self.grads_and_vars, global_step=tf.contrib.framework.get_global_step())

Answer 1

看起来你正面临着 ReLu 激活函数梯度爆炸的问题(这就是 NaN 的意思——非常大的激活)。有几种技术可以解决这个问题，例如batch normalization (更改网络架构)或微妙的变量初始化(这是我首先尝试的)。

您正在对不同层中的 V 变量使用 Xavier 初始化，这确实适用于逻辑 sigmoid 激活(请参阅 the paper by Xavier Glorot and Yoshua Bengio )，或者换句话说，tanh。

ReLU 激活函数(及其变体，包括 ELU)的首选初始化策略是 He 初始化。在tensorflow中它是通过tf.variance_scaling_initializer实现的:

initializer = tf.variance_scaling_initializer()
v = tf.get_variable(name=var_name, initializer=initializer, ...)

您可能还想尝试使用较小的 b 和 g 变量值，但仅通过查看模型很难说出确切的值。如果没有帮助，请考虑向模型中添加批归一化层以控制激活分布。