Langevin Cooling for Domain Translation

TL;DR

This paper introduces Langevin Cooling (L-Cool), a method that improves domain translation by moving low-density, poorly modeled samples towards high-density regions, enhancing existing unsupervised translation models.

Contribution

The paper proposes Langevin Cooling (L-Cool), a novel technique that refines domain translation by leveraging Langevin dynamics to improve sample quality in low-density areas.

Findings

L-Cool improves image translation quality.

L-Cool enhances language translation performance.

The method effectively moves fringe samples to high-density regions.

Abstract

Domain translation is the task of finding correspondence between two domains. Several Deep Neural Network (DNN) models, e.g., CycleGAN and cross-lingual language models, have shown remarkable successes on this task under the unsupervised setting---the mappings between the domains are learned from two independent sets of training data in both domains (without paired samples). However, those methods typically do not perform well on a significant proportion of test samples. In this paper, we hypothesize that many of such unsuccessful samples lie at the fringe---relatively low-density areas---of data distribution, where the DNN was not trained very well, and propose to perform Langevin dynamics to bring such fringe samples towards high density areas. We demonstrate qualitatively and quantitatively that our strategy, called Langevin Cooling (L-Cool), enhances state-of-the-art methods in image translation and language translation tasks.