One researcher is using technology to recreate some of the world’s oldest and most beautiful works of art.

Human-computer interaction researcher Victor Dibia is using artificial intelligence to generate African masks based on his custom curated dataset.

The Carnegie Mellon graduate was inspired to explore merging tribal art and AI after attending the Deep Learning Indaba conference in South Africa where Google provided attendees Tensor Processing Units (TPUs)—Google’s custom-developed AI accelerator application.

He trained a generative adversarial network (GAN)—a two-part neural network consisting of generators that produce samples and discriminators that attempt to distinguish between the generated samples and real-world samples—to generate images based on the dataset he built.

Dibia explains in a blog post that he manually created a dataset of over 9000 diverse images depicting African masks in different shapes and textures.

“The goal is not to generate a perfectly realistic mask, but more towards observing any creative or artistic elements encoded in the resulting GAN,” he wrote.

The researcher trained the GAN by using a larger set of non-curated images from a web search with initial results showing the model generating images “distinct from their closest relatives in the dataset.”

“GANs can be useful for artistic exploration,” he wrote of his findings. “In this case, while some of the generated images are not complete masks, they excel at capturing the texture or feel of African art.”

Dibia plans to expand the Africa masks dataset and continue experiments with “conditioned GANs” and its relationship to artistic properties.