Honors College Theses

Publication Date

11-30-2021

Major

Information Technology (B.S.)

Document Type and Release Option

Thesis (open access)

Faculty Mentor

Hayden Wimmer

Abstract

Deep learning is a type of Artificial Intelligence (AI) that mimics the workings of the human brain in processing data such as speech recognition, visual object recognition, object detection, language translation, and making decisions. A Generative adversarial network (GAN) is a special type of deep learning, designed by Goodfellow et al. (2014), which is what we call convolution neural networks (CNN). How a GAN works is that when given a training set, they can generate new data with the same information as the training set, and this is often what we refer to as deep fakes. CNN takes an input image, assigns learnable weights and biases to various aspects of the object and is able to differentiate one from the other. This is similar to what GAN does, it creates two neural networks called discriminator and generator, and they work together to differentiate the sample input from the generated input (deep fakes). Deep fakes is a machine learning technique where a person in an existing image or video is replaced by someone else’s likeness. Deep fakes have become a problem in society because it allows anyone’s image to be co-opted and calls into question our ability to trust what we see. In this project we develop a GAN to generate deepfakes. Next, we develop a survey to determine if participants are able to identify authentic versus deep fake images. The survey employed a questionnaire asking participants their perception on AI technology based on their overall familiarity of AI, deep fake generation, reliability and trustworthiness of AI, as well as testing to see if subjects can distinguish real versus deep fake images. Results show demographic differences in perceptions of AI and that humans are good at distinguishing real images from deep fakes.

Thesis Summary

Generative adversarial network (GAN) is a deep learning method used to create deepfakes. Deep fakes is a machine learning technique where a person in an existing image or video is replaced by someone else’s likeness. We built a GAN to generate deepfakes of the CelebA dataset and created a survey to test how well humans can detect real versus fake images. The survey also employed a questionnaire asking participants their perception on AI technology based on their overall familiarity of AI, deep fake generation, reliability and trustworthiness of AI, as well as testing to see if subjects can distinguish real versus deep fake images. Results show demographic differences in perceptions of AI and that humans are good at distinguishing real images from deep fakes.

Download

Share

COinS