built and trained a 218-million-parameter model on fMRI data from 2,000 peoples brains as they watched films, TV, and images.
Cajal-1.
On the films Meta held out to test TRIBE v2, Cajal-1 is 25% more accurate and closer to the real brain in 27 of 32 recordings.
Give it any video, podcast, or song, and it predicts fMRI responses at 20,484 points across the human cortex every 1.49 seconds. Same speed, no scanner needed.
It's named after Santiago Ramón y Cajal, the first Spaniard to win a Nobel Prize in science. In the 1890s, he used Golgi's silver stain (of Golgi Body fame) to draw neurons by hand and showed that the brain is built from separate cells. A century later, we can predict how these cells respond brain-wide to any stimulus we come across.
built and trained a 218-million-parameter model on fMRI data from 2,000 peoples brains as they watched films, TV, and images.
Cajal-1.
On the films Meta held out to test TRIBE v2, Cajal-1 is 25% more accurate and closer to the real brain in 27 of 32 recordings.
Give it any video, podcast, or song, and it predicts fMRI responses at 20,484 points across the human cortex every 1.49 seconds. Same speed, no scanner needed.
It's named after Santiago Ramón y Cajal, the first Spaniard to win a Nobel Prize in science. In the 1890s, he used Golgi's silver stain (of Golgi Body fame) to draw neurons by hand and showed that the brain is built from separate cells. A century later, we can predict how these cells respond brain-wide to any stimulus we come across.
The model and a Colab demo are now open: https://huggingface.co/ffh92r32rm0/Cajal-1