Aspect-augmented Adversarial Networks for Domain Adaptation

About this repo

This repo contains the code and resources of the following paper:

• Aspect-augmented Adversarial Networks for Domain Adaptation. Yuan Zhang, Regina Barzilay, and Tommi Jaakkola.

This paper introduces an adversarial network method for transfer learning between two (source and target) classification tasks or aspects over the same domain.

Experiments

• The directory synthetic contains the implementation of training model on synthetic datasets. The synthectic data generator is also included.
• The medical data is not available due to the privacy issue. The synthectic dataset somehow represents the challenge in the medical data.

Structures

• /nn: source code of NN library based on this repo
• /word2vec: word2vec (from here) for training word embeddings
• /synthetic: source code for generating and training models on synthetic data

Dependencies

Theano >= 0.8, Python >= 2.7, Numpy

To-do

• source code for the review dataset