Improving Compositionality of Neural Networks by Decoding Representations to Inputs

Open in new window