def fairness(self, vecs):
r"""Build fairness metrics component.
"""
logits = tf.concat(vecs, axis=1)
for i in range(self.num_dis_layers):
with tf.variable_scope('fair_fc{}'.format(i)):
if i == 0:
logits = FullyConnected(
'fc',
logits,
self.num_dis_hidden,
nl=tf.identity,
kernel_initializer=tf.truncated_normal_initializer(
stddev=0.1))
else:
logits = FullyConnected('fc',
logits,
self.num_dis_hidden,
nl=tf.identity)
logits = tf.concat(
[logits, self.batch_diversity(logits)], axis=1)
logits = BatchNorm('bn', logits, center=True, scale=False)
logits = Dropout(logits)
logits = tf.nn.leaky_relu(logits)
return FullyConnected('fair_fc_top', logits, 1, nl=tf.identity)
def discriminator(self, vecs):
r"""Build discriminator.
We use a :math:`l`-layer fully connected neural network as the discriminator.
We concatenate :math:`v_{1:n_c}`, :math:`u_{1:n_c}` and :math:`d_{1:n_d}` together as the
input. We compute the internal layers as
.. math::
\begin{aligned}
f^{(D)}_{1} &= \textrm{LeakyReLU}(\textrm{BN}(W^{(D)}_{1}(v_{1:n_c} \oplus u_{1:n_c}
\oplus d_{1:n_d})
f^{(D)}_{1} &= \textrm{LeakyReLU}(\textrm{BN}(W^{(D)}_{i}(f^{(D)}_{i−1} \oplus
\textrm{diversity}(f^{(D)}_{i−1})))), i = 2:l
\end{aligned}
where :math:`\oplus` is the concatenation operation. :math:`\textrm{diversity}(·)` is the
mini-batch discrimination vector [42]. Each dimension of the diversity vector is the total
distance between one sample and all other samples in the mini-batch using some learned
distance metric. :math:`\textrm{BN}(·)` is batch normalization, and
:math:`\textrm{LeakyReLU}(·)` is the leaky reflect linear activation function. We further
compute the output of discriminator as :math:`W^{(D)}(f^{(D)}_{l} \oplus \textrm{diversity}
(f^{(D)}_{l}))` which is a scalar.
Args:
vecs(list[tensorflow.Tensor]): List of tensors matching the spec of :meth:`inputs`
Returns:
tensorpack.FullyConected: a (b, 1) logits
"""
logits = tf.concat(vecs, axis=1)
with tf.variable_scope('discrim'):
for i in range(self.num_dis_layers):
with tf.variable_scope('dis_fc{}'.format(i)):
if i == 0:
logits = FullyConnected(
'fc',
logits,
self.num_dis_hidden,
nl=tf.identity,
kernel_initializer=tf.truncated_normal_initializer(
stddev=0.1))
else:
logits = FullyConnected('fc',
logits,
self.num_dis_hidden,
nl=tf.identity)
logits = tf.concat(
[logits, self.batch_diversity(logits)], axis=1)
logits = LayerNorm('ln', logits)
logits = Dropout(logits)
logits = tf.nn.leaky_relu(logits)
return FullyConnected('dis_fc_top', logits, 1, nl=tf.identity)