Neural network learned to draw together with the artist

2022: Presentation of the neural network co-author

Cloudy Yandex Cloud platform, studio Arka and multimedia artist Andrew Berger presented neuronet the author. This was Yandex announced on December 8, 2022. This is one algorithm machine learning that can constantly complement and complete an image with a person. Participants called this method of digital art the "ping-pong principle." From January 28, 2023, everyone in the Yandex Museum will be able to learn more about the project and see how the neural network works.

Illustration created by the co-author neural network

A convolutional neural network VGG-19 was used for the performance. It was deployed in the Yandex DataSphere ML development environment and processed with its help a rough sketch of the artist's illustration. Then Andrew Berger corrected the resulting image. Then the illustration was again "finished" by the neural network - and such a consistent exchange was repeated more than 15 times. In addition to sketches, QR codes, the Yandex Cloud logo, Malevich's Black Square, images of bubble, layered and other natural structures were "mixed" during the performance.

This "ping-pong principle" can be used to generate any images - for exhibitions, industrial and architectural design, advertising. As a result of the collaboration, a blanket with a special digital pattern was created.

Plaid with a digital drawing created in collaboration with a neural network

Many artists explore machine learning territories. There are many scenarios for interaction with neural networks. Basically, the artist loads his request into them and the algorithm gives him something. Some of the most common are the so-called text query generation or source image generation, and depending on what the network is trained on, the artist gets the result. When the first such projects appeared, the results made a splash. We see how the popularity of the entertainment application of these algorithms is growing. However, I was interested in building mutual interaction with artificial intelligence. I wanted not only to train the neural network in my pictures, but also myself, looking at her work, try to understand its logic, build such a dialogue of co-creation, "said multimedia artist Andrew Berger.

Developing a neural network for art science is often a technologically difficult task. To solve it, it may take a lot of computing resources and time for a machine learning specialist. A clear interface, quick settings inside Yandex DataSphere helped specialists create a special neural network in a short time and make an important contribution to the development of digital art, "said Alexey Bashkeev, head of the Yandex Cloud platform.