Neural Network That Predicts Your Reaction to a Video
Meta has developed a unique neural network called TRIBE, capable of predicting how your brain will respond to a video — even before you hit the play button, and without any brain scanning!
This 1-billion-parameter model analyzes video, audio, and text to precisely determine which areas of the brain will be activated.
TRIBE was trained on 80 hours of TV shows and movies, enabling it to correctly predict more than half of brain activity patterns across 1,000 brain regions. It performs especially well with multisensory information — where vision, sound, and language interact — outperforming traditional models by 30%.
Interestingly, the system showed high accuracy in the frontal lobes, responsible for attention, decision-making, and emotional reactions to content. This opens new possibilities for developing methods aimed at maximizing viewer engagement at the neural level — which could make “doomscrolling” even more addictive.
In short: input — a video; output — information on which brain regions will light up. Multimodality really amplifies the effect!
https://github.com/facebookresearch/algonauts-2025
https://arxiv.org/abs/2507.22229
@science
Meta has developed a unique neural network called TRIBE, capable of predicting how your brain will respond to a video — even before you hit the play button, and without any brain scanning!
This 1-billion-parameter model analyzes video, audio, and text to precisely determine which areas of the brain will be activated.
TRIBE was trained on 80 hours of TV shows and movies, enabling it to correctly predict more than half of brain activity patterns across 1,000 brain regions. It performs especially well with multisensory information — where vision, sound, and language interact — outperforming traditional models by 30%.
Interestingly, the system showed high accuracy in the frontal lobes, responsible for attention, decision-making, and emotional reactions to content. This opens new possibilities for developing methods aimed at maximizing viewer engagement at the neural level — which could make “doomscrolling” even more addictive.
In short: input — a video; output — information on which brain regions will light up. Multimodality really amplifies the effect!
https://github.com/facebookresearch/algonauts-2025
https://arxiv.org/abs/2507.22229
@science