Population-Based Multi-Objective Training of Discriminators for Semi-Supervised GANs
For practitioners using semi-supervised GANs, this work offers a more stable training approach that improves classifier accuracy, though the improvement is demonstrated only on a single dataset (MNIST).
The paper tackles training instability in semi-supervised GANs by formulating discriminator learning as a multi-objective optimization problem with a population of discriminators ranked by Pareto dominance. On MNIST with limited labels, the proposed method improves training robustness and achieves higher classification accuracy compared to SSL-GAN and CE-SSL-GAN baselines.
Semi-supervised generative adversarial networks (SSL-GANs) can exploit large unlabeled datasets while retaining a classifier in the discriminator, but their training is often unstable. This paper proposes a population-based evolutionary training strategy in which discriminator learning is formulated as a multi-objective optimization problem. Instead of aggregating the supervised and unsupervised components of the SSL objective into a single scalar loss, the method maintains a population of discriminators ranked by Pareto dominance, enabling the exploration of different trade-offs between classification accuracy and real/fake discrimination. This formulation aims to improve both roles of SSL-GANs: learning accurate classifiers and training generators capable of producing realistic samples. We analyze several variants, including an elitist strategy and a mono-objective ablation, to assess the role of multi-objective selection. Experiments on MNIST with limited labels show improved training robustness compared to SSL-GAN and CE-SSL-GAN state-of-the-art baselines, while the elitist variant consistently achieves the highest classification accuracy.