Join / sign up Change ME
Researchers from the Moscow Institute of Physics and Technology successfully build an image classifier using only EEG data. The experiments used the (natural) brain data captured from the subject, which was then fed into an artificial neural network to classify the image observed. Remarkably, the resulting neural network not only gets 90%+ accuracy, but also also achieves modest results in reconstructing the visual image the subject observer (see image)
While still nowhere near perfect, the qualitative impression of the model’s ability to reconstruct subjects’ visual impressions here is notable. Research like this speaks to the increasing abilities for direct brain interface and has important implications for in applications to prosthetics. In contrast to prosthetic applications in current state of the art, which require implantation and raise concerns of deterioration of implanted technologies, the EEG data needed here can be captured through the non-invasive electrodes, bearing significantly less risk.