Electronic Thesis/Dissertation
 

Deepfake Detection Models: Techniques for Improving Performance and Visualization

Open Access

Deepfake content is being weaponized through influence operations, as evidencedduring the 2016 election, where nation-state actors threatened to disrupt US politics. The popularity of computer-generated content is at an all-time as generative AI tools like ChatGPT and Dall-E have brought the power and promise of machine learning to the masses, generating excitement and fear. While some industries have rushed headfirst into using the technology, others are being far more cautious - taking a measured approach to understand the risks better. One of the risks is using fake content to alter meaningful establishments like the US voting system. Russia's use of fakes on social media exposed this cyber-attack technique's power. This research seeks to build and train models that can match the speed of deepfake technology advancement while using less data and time to train models with no loss of performance accuracy. Validating the results in this research involved conducting several experiments confirming that introducing image transformation as a prerequisite step to model ingestion can reduce training and validation overhead by 55% while using 90% less data than traditional convolution neural networks, all while having the same or better accuracy. In addition, deep learning model explanation tools confirm that the edges of the image and the eyes, nose, and forehead have a meaningful role in image classification. Custom models can significantly make social media users more aware of deepfake content.

Author Language Keyword Date created Type of Work License
  • All rights reserved
Rights statement GW Unit Degree Advisor Committee Member(s) Persistent URL

Notice to Authors

If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.

Thumbnail Title Date Uploaded Visibility Actions
Preview of Moore_gwu_0075A_16545.pdf Moore_gwu_0075A_16545.pdf 2023-11-14 Open Access