Deepfakes, audiofakes, and the future of media
summary
Reality Defender launches as a production‑focused platform for the 2020 election that combines forensic models, segmentation to localize manipulations, and behavioral checks like head‑movement and FACS signatures to spot mismatches. It applies transfer learning — forensic transfer — to adapt detectors with fewer than a dozen examples, and uses convolutional multitask nets to detect audiofakes while cutting required training data to about 6% (a couple minutes). Releases include the 2017 FaceForensics dataset and an audio corpus of half a million examples.
AI-written summary · about
chapters (22)
- 0:00Introduction
- 0:40Deepfakes
- 2:50Credibility crisis
- 3:43Fake media
- 4:30Cambrian explosion
- 5:18Social media
- 5:58Y reality
- 6:10misinformation singularity
- 6:42what are we doing
- 6:54the 2020 election
- 7:17the goals of Reality Defender
- 9:59examples of work
- 12:16forensic transfer
- 14:19head movement
- 15:28facial expressions
- 16:24audiofake detection
- 17:26new data set
- 18:19education policy
- 18:55education
- 21:11policy
- 23:56impact on education
- 25:13conclusion
related talks
28mDefending our reality in the era of deepfakes
27mTruth or Fiction The Art of Unveiling DeepFakes in Text, Audio, Image
30mReality Under Attack: Just How Serious Is the Deep Fake Threat?
30mAI-Generated Synthetic Media, aka Deepfakes and its impact
47mFighting AI with AI
01h 04mPushing Deepfakes to The Limit - Fake Video Calls With AI
41mThe Deepfake Supply Chain: Weaponizing Synthetic Media into Modern Cybercrime Operation
51mPlaying Offense and Defense with Deepfakes