google-research
Mind the (optimality) Gap: A Gap-Aware Learning Rate Scheduler for Adversarial Nets
This repository contains an implementation and a demo of the learning rate scheduler proposed in Mind the (optimality) Gap: A Gap-Aware Learning Rate Scheduler for Adversarial Nets (AISTATS, 2023). The scheduler stabilizes the training of adversarial nets (e.g., GANs) by dynamically modifying the learning rate. See the paper for more details.
Usage
See the colab (demo.ipynb) in this repo.