├── flip_gradient.py
└── README.md


/flip_gradient.py:
--------------------------------------------------------------------------------
1 | import tensorflow as tf
2 | 
3 | def flip_gradient(x, l=1.0):
4 | 	positive_path = tf.stop_gradient(x * tf.cast(1 + l, tf.float32))
5 | 	negative_path = -x * tf.cast(l, tf.float32)
6 | 	return positive_path + negative_path


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
1 | # GradientReversal
2 | 
3 | Implements the Gradient Reversal layer from Unsupervised Domain Adaptation by Backpropagation (https://arxiv.org/abs/1409.7495) and Domain-Adversarial Training of Neural Networks (https://arxiv.org/abs/1505.07818) in Tensorflow.
4 | 
5 | The forward pass is the identify function, but the backward pass multiplies the gradients by -lambda.
6 | 


--------------------------------------------------------------------------------