[Re] Controllable Unsupervised Text Attribute Transfer via Editing Entangled Latent Representation

Zhi Wen, Shih-Chieh Fuh, Andrei Romascanu

Abstract:

Adversarial editing is a common technique
used for attribute transfer. In the reviewed pa-
per, the authors applied the technique on en-
tangled latent representations to build a con-
trollable and flexible model for text attribute
transfer. In our ablation study, we studied the
effect of latent space dimension and number of
Transformer layers on the performance of the
original model. We found that the pre-trained
model provided by the authors had a lower per-
formance than the reported performance. In
addition, we have reported several issues re-
garding the model implementation, but we be-
lieve that the overall structure of the model de-
sign remains correct and valid.

Track: Ablation

NeurIPS Paper Id: https://openreview.net/forum?id=SyeoerSlUB