Siwei Lyu
University at Buffalo, NY·Computer Science
Deepfakes are fake images, audio and videos created by artificial intelligence (AI) to impersonate real people. They are a growing form of online disinformation and can be used to cause harm.
- Fields
Computer Science
Engineering
- Topics
deepfakes
AI-synthesized media
media forensics
- Grants
- 1 known grant, last ended Sep 2026
What they work on
- Strengthen deepfake detectors against anti-forensic attacks by processing each input with a randomly chosen model from an infinite ensemble of classifiers, making detectors harder to fool.
- Attribute a deepfake to the specific generation model that created it, to support tracing its origin and author.
- Add detectable traces to synthesized deepfakes by contaminating the data used to train deepfake generators, an active defense applied before fakes are made.
- Obstruct deepfake generation through data poisoning, which sabotages training of generative models so they become less efficient and produce low-quality fakes.
More about the research
Current detection methods perform well on benchmark datasets but have important limits. This project aims to go beyond detection. It pursues 'active' approaches, which mark deepfakes without interfering with how they are made. It also pursues 'proactive' approaches, which disrupt deepfake creation before it happens.
Why it matters
These methods could help protect individuals from being impersonated by deepfakes and limit AI-generated disinformation online. Tools that resist evasion, trace a fake to its source, or block convincing fakes from being made could help platforms, investigators and the public fight manipulated media.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Matthew Wright
Rochester Institute of Technology, NY·Computer Science
Study how users evaluate suspect digital content through qualitative interviews and user studies, and use the findings to integrate multiple AI-content forensic tools into one navigable platform.
Raymond Yeh
Purdue University, IN·Computer Science·Early-career
Develop semi-parametric vision architectures that consult training examples during prediction, so each output can be traced to specific data, combining deep network accuracy with k-nearest-neighbor transparency.
Jessica Fridrich
Binghamton University, NY·Computer Science
Use artificial but realistic image datasets to compute theoretical limits of steganography and assess how close modern data-driven hiding algorithms and steganalysis detectors come to optimal.
The school
University at Buffalo: 32 labs on PI Pages, $542M a year on research, 316 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.