Generating Images of University of Virginia Architecture with GANs ; Regulating Innovation: How Advocates Fight for Algorithmic Regulation
Harris, Arthur, School of Engineering and Applied Science, University of Virginia
Nguyen, Rich, EN-Comp Science Dept, University of Virginia
Norton, Peter, EN-Engineering and Society, University of Virginia
Digital systems imperfectly sense and represent external phenomena, constraining their capacity to depict or interpret them accurately or equitably. Generative modeling can now create realistic images. To evaluate generative modeling of complex images, images of the University of Virginia’s Rotunda were generated through a Generative Adversarial Network (GAN). The model required several thousand iterations to roughly approximate the Rotunda’s shape. Such results indicate that with sufficient training examples, time to train, and hardware, generative modeling with GANs might handle highly complex images and concepts. Because algorithms are as biased as the real-world data they are trained on, their predictions are biased. Advocacies such as the ACLU cite such biases to demand that tech firms like Facebook be subject to algorithmic regulation. Self-regulation by tech companies has not satisfied their critics.
BS (Bachelor of Science)
Generative Modeling, Algorithmic Bias, GANs, Architecture, Algorithmic Legislation
School of Engineering and Applied Science
Bachelor of Science in Computer Science
Technical Advisor: Rich Nguyen
STS Advisor: Peter Norton
English
All rights reserved (no additional license for public reuse)
2021/05/10