After Goya (2020)
Project selected for the VisionarIAs international open call
GAN generated prints etchings and video
Aarati has worked with Goya’s collections of etchings prints to identify themes across aspects such as pose, composition, or subject matter within a visual archive and highlight explore these themes through creative machine learning experiments. Firstly, she created a body of generative works based on the archive using a DCGANS model. Overtime the neural networks become better and better at imitating the original images and you can slowly start to see details form that reflect the visual essence of the training dataset.
The generated images become a dream-like blend of the originals, showing interesting details from the training dataset. She was interested in connecting details in the generated images with instances within the training dataset. She finds the in-between stages when the neural networks are still grasping at the forms to be extremely interesting as well. The installation presents a series of prints and a video showing the transition process.
The second component of the project she explored, used a pre-trained machine learning model to tease out themes around pose and composition in the original training dataset. This part of the project is a data visualization showing different continuities within pose and composition, through the use of these machine vision overlays.
Aarati Akkapeddi is a first-generation Indian-American, cross-disciplinary artist, educator, and programmer interested in the poetics and politics of datasets. They work with both personal and institutional archives to explore how identities and histories are shaped by different methods of collecting, preserving, and presenting data. Their work has been supported by institutions such as NYC Media Lab, Beamcenter, ETOPIA Center for Art & Technology, & LES Printshop. She lives and works in Occupied Lenapehoking (New York).
Getting machines to “see” like people do is one of the main goals of artificial intelligence, with applications ranging from controlling robots and driving cars to detecting terrorists. Machine vision works by first collecting live images through cameras, and then interpreting these images with complex calculations in software. Many of the most successful techniques use trained models from deep learning to estimate things such as what content is in an image, what emotion someone is showing, or whether a piece of art is a pastiche of a great master or not