The Defense Advanced Research Projects Agency has announced two new initiatives to help the government and industry boost their defenses against manipulated and synthesized media, known as deepfakes.
DARPA said Thursday it will launch an open community research effort to create machine learning models designed to detect synthetic images generated by artificial intelligence.
The AI Forensics Open Research Challenge Evaluation will feature a series of mini challenges to develop models that can identify the difference between authentic and fully synthetic AI-generated images.
Another effort is an analytic catalog that offers government and industry researchers access to open-source resources developed under DARPA’s Semantic Forensics, or SemaFor, program.
Now in its final phase, SemaFor develops new approaches to using semantic technologies to detect, attribute and characterize fraudulent media.
DARPA selected SRI International and PAR Technology‘s government subsidiary to support the development of semantic approaches to media assets potentially manipulated for malicious purposes.