The world is being quietly rearranged by people who write very long documents.


The title they went with MuDD: A Multimodal Deception Detection Dataset and GSR-Guided Progressive Distillation for Non-Contact Deception Detection Noisy translates that to

New dataset tackles lie detection without touching the body


Researchers created a large dataset pairing video, audio, and skin response measurements from 130 people to train AI systems that detect deception from facial expressions and speech alone, without needing electrodes. This matters because skin response is the most reliable biological signal for lies, but it requires wiring people up with sensors — the new approach tries to learn patterns from skin data and transfer that knowledge to systems that only need a camera and microphone.
Deception detection has been bottlenecked by the fact that visual and auditory cues are unreliable across different people, while the physiological signals that work best require contact sensors — this dataset and method attempt to use reliable biological patterns to improve non-contact detection, which could enable deployment in settings where attaching sensors is impractical or unwanted.

If you insist
Read the original →