def initialize_generator(self, discriminator, *inputs):
if len(inputs) != 2:
raise ValueError('this generator is supposed to get two inputs.'
' the message and the key.')
message_input, key_input = inputs
if self.tie_alice_and_bob:
if self.alice_share_bitwise_weights != self.bob_share_bitwise_weights:
raise ValueError('bob and alice are tied however they do not '
'have the same weight sharing settings.')
enc_dec = ElementWise(
self.alice_bitwise_latent_dims,
activation='tanh',
share_element_weights=self.alice_share_bitwise_weights)
alice_encryption = Flatten()(enc_dec([message_input, key_input]))
bob_decryption = Flatten(name='decryption')(enc_dec(
[alice_encryption, key_input]), )
else:
alice_encryption = Flatten(name='decryption')(ElementWise(
self.bob_bitwise_latent_dims,
activation='tanh',
share_element_weights=self.bob_share_bitwise_weights)(
[message_input, key_input]), )
bob_decryption = Flatten(name='decryption')(ElementWise(
self.bob_bitwise_latent_dims,
activation='tanh',
share_element_weights=self.bob_share_bitwise_weights)(
[alice_encryption, key_input]), )
eves_opinion = discriminator([message_input, alice_encryption])
alice = Model(inputs=inputs, outputs=alice_encryption)
alice_bob = Model(inputs=inputs,
outputs=[bob_decryption, eves_opinion])
def discriminator_loss(y_true, y_pred):
return K.abs(0.5 - K.mean(y_pred))
alice_bob.compile(optimizer=Adam(),
loss=[mean_absolute_error, binary_crossentropy])
return alice_bob, alice
def initialize_model(self):
message_input = Input(shape=(self.message_length,), name='message_input')
possible_ciphertext_input = Input(shape=(self.message_length,), name='possible_ciphertext_input')
bitwise_function = Flatten()(
ElementWise([8, 1], activation='tanh')([
message_input,
possible_ciphertext_input
])
)
dense = Dense(
self.message_length,
activation='relu'
)(bitwise_function)
pred = Dense(
1,
activation='sigmoid'
)(dense)
model = Model(inputs=[message_input, possible_ciphertext_input], outputs=pred)
model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['acc'])
self.model = model
return [model]
def initialize_discriminator(self, *inputs):
if len(inputs) != 2:
raise ValueError(
'this discriminator is supposed to get two inputs.'
'the message and the key.')
message_input = inputs[0]
possible_ciphertext_input = inputs[1]
bitwise_function = Flatten()(ElementWise(
self.eve_bitwise_latent_dims,
activation='tanh',
share_element_weights=self.eve_share_bitwise_weights)(
[message_input, possible_ciphertext_input]))
dense = Dense(self.eve_latent_dim[0],
activation='relu')(bitwise_function)
for units in self.eve_latent_dim[1:]:
dense = Dense(units, activation='relu')(dense)
pred = Dense(1, activation='sigmoid')(dense)
model = Model(inputs=inputs, outputs=pred)
model.compile(optimizer=Adam(),
loss='binary_crossentropy',
metrics=['acc'])
return model
def initialize_generator(self, discriminator, *inputs):
if len(inputs) != 2:
raise ValueError('this generator is supposed to get two inputs.'
' the message and the key.')
message_input, key_input = inputs
alice_encryption = Flatten()(
ElementWise(self.alice_bitwise_latent_dims, activation='tanh',
share_element_weights=self.alice_share_bitwise_weights)([
message_input,
key_input
])
)
bob_decryption = Flatten(name='decryption')(
ElementWise(self.bob_bitwise_latent_dims, activation='tanh',
share_element_weights=self.bob_share_bitwise_weights)([
alice_encryption,
key_input
]),
)
eves_opinion = discriminator([message_input, alice_encryption])
alice = Model(inputs=inputs, outputs=alice_encryption)
alice_bob = Model(inputs=inputs, outputs=[bob_decryption, eves_opinion])
def discriminator_loss(y_true, y_pred):
return K.abs(0.5 - binary_accuracy(y_true, y_pred)) ** 2
alice_bob.compile(optimizer=Adam(lr=0.0008),
loss=[mean_absolute_error, discriminator_loss])
return alice_bob, alice
def initialize_model(self):
input_1 = Input(shape=(self.input_length, ), name='input_1')
input_2 = Input(shape=(self.input_length, ), name='input_2')
# reshape = Reshape((-1, 1))
# ORIGINAL BITWISE FUNCTION IMPLEMENTATION
#
# bitwise_function = Flatten()(
# TimeDistributed(
# Dense(1, activation='tanh')
# )(LocallyConnected1D(
# self.latent_dim,
# kernel_size=2,
# strides=2,
# activation='relu'
# )(reshape(
# Intertwine()([
# input_1,
# input_2
# ]))
# )
# )
# )
bitwise_function = Flatten()(ElementWise(
[self.latent_dim, 1],
activation=['relu', 'tanh'],
share_element_weights=True)([input_1, input_2]))
model = Model(inputs=[input_1, input_2], outputs=bitwise_function)
model.compile(optimizer=Adam(), loss=mean_absolute_error)
self.model = model
if self.verbose:
model.summary()
return [model]