[…] a team of researchers from the École Polytechnique Fédérale de Lausanne (EPFL) successfully developed a machine-learning algorithm that can decode a mouse’s brain signals and reproduce images of what it’s seeing.
The mice were shown a black and white movie clip from the 1960s of a man running to a car and then opening its trunk. While the mice were watching the clip, scientists measured and recorded their brain activity using two approaches: electrode probes inserted into their brains’ visual cortex region, as well as optical probes for mice that had been genetically engineered so that the neurons in their brains glow green when firing and transmitting information. That data was then used to train a new machine learning algorithm called CEBRA.
When then applied to the captured brain signals of a new mouse watching the black and white movie clip for the first time, the CEBRA algorithm was able to correctly identify specific frames the mouse was seeing as it watched. Because CEBRA was also trained on that clip, it was also able to generate matching frames that were a near perfect match, but with the occasional telltale distortions of AI-generated imagery.
This research involved a very specific (and short) piece of footage that the machine learning algorithm was also familiar with. In its current form, CEBRA also really only takes into account the activity from about 1% of the neurons in a mouse’s brain, so there’s definitely room for its accuracy and capabilities to improve. The research also isn’t just about decoding what a brain sees. A study, published in the journal, Nature, shows that CEBRA can also be used to “predict the movements of the arm in primates,” and “reconstruct the positions of rats as they freely run around an arena.” It’s a potentially far more accurate way to peer into the brain, and understand how all the neural activity correlates to what is being processed.
Source: Researchers See Through a Mouse’s Eyes by Decoding Brain Signals