Search papers, labs, and topics across Lattice.
This paper introduces a population-based evolutionary training strategy for semi-supervised generative adversarial networks (SSL-GANs), addressing the instability often encountered in their training. By framing the discriminator's learning as a multi-objective optimization problem, the method retains a diverse population of discriminators ranked by Pareto dominance, allowing for a nuanced exploration of trade-offs between classification accuracy and real/fake discrimination. Experimental results on MNIST demonstrate that this approach enhances training robustness and achieves superior classification accuracy compared to existing state-of-the-art methods.
Maintaining a diverse population of discriminators leads to more robust training and significantly higher classification accuracy in semi-supervised GANs.
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.
