Accuracy Eludes Competitors in Facebook Deepfake Detection Challenge

• 来源期刊: 工程（英文）
• 年，卷（期）: 2020,（12）
• 所属期刊栏目: News & Highlights
• 页码范围: 1339-1340
• 页数: 2页
• 语种: 英文

摘要

The improving power of artificial intelligence(AI)is perhaps most evident in the increasingly realistic manipulation of video and other digital media[1],with the latest generation of Al-altered videos,known as deepfakes[2],prompting a primarily Facebook-sponsored competition to identify them as such.Launched in December 2019,the Deepfake Detection Challenge(DFDC)closed to entries in March 2020[3].The results are now in Refs.[3-5].While somewhat unimpressive,underscoring the difficulty of addressing this growing challenge,they importantly provide a benchmark for automated detection strategies and suggest produc-tive directions for further research.