search menu icon-carat-right cmu-wordmark

Machine Learning for Deepfake Detection

This presentation by Shannon Gallagher was delivered virtually at Deepfakes Day 2022 on August 30, 2022.

Software Engineering Institute

Topic or Tag


In this talk, we will go into more detail about some of state-of-the-art deepfake detectors including topics such as data, pre-processing, neural networks, and evaluation. We will also discuss the dangers presented by generative adversarial networks to deepfake detection systems.

In summer 2021, Shannon Gallagher joined CERT as a data scientist. Her interests include modeling, uncertainty quantification, and data visualization. She always enjoys meeting and chatting with potential collaborators. Currently, she is the PI of the "A Prototype Software Framework for Digital Content Forgery Detection" which aims to develop a statistical pipeline to help determine the authenticity of images and videos. Specifically, her team will modify existing algorithms to work at scale (thousands/day) while maintaining accuracy, precision, and recall. Previously, Shannon worked on a team to implement a machine learning algorithm to detect code vulnerabilities on source code intermediate representation.

Shannon joined the SEI after completion of her post doc at the National Institute of Allergy and Infectious Diseases. There she worked on statistical modeling of infectious diseases, competing events analysis for the ACTT-1 COVID-19 trial, and analysis of statistical tests in low event rate settings. Prior to that, Shannon received her PhD in Statistics at Carnegie Mellon University and was advised by Bill Eddy. Her dissertation studied statistical properties of agent-based models. While at CMU, she was a research and teaching assistant and served as President of the Women in Statistics group.