The new system uses electrical signals from the brain as its sole source of data.
To demonstrate device learning,the researchers selected 30 volunteers. Each of them was wearing an EEG-tracking helmet. All participants were asked to look at the same images. In total, there were about 200 different photos in the set.
Participants were asked to complete various tasksfor example, find female faces, elderly people, or blonde hair. At the same time, they did not have to do anything, it was enough to look at each image for 0.5 seconds.
The scheme of the system. Source: Keith M. Davis III et al, CVPR 2022
Based on brain activity, the machinefirst displays the preference data and then edits the images accordingly. For example, if the participant was tasked with looking for older people, the computer modified the portraits of the young people, making them older.
Remarkably, the computer knows nothing aboutface recognition and has no idea about gender, hair color or any other relevant characteristics. However, he only edited the feature in question, leaving other facial features unchanged.
Keith Davis, study participant from the University of Helsinki
The researchers note that all previousAI systems that can edit images have relied on learning from labeled inputs. For example, to teach the system to “age” people in an image, you need to show it thousands of pictures, indicating who is old and who is young.
The new approach uses the brain activity of the subjects as the only input parameter. This is a completely new paradigm of artificial intelligence, the researchers note.
Read more:
Scientists uncover how vitamins affect the incidence of cancer
The space probe flew 200 km from Mercury. Look what he saw
The satellite of Jupiter was looked at in a new light: what scientists saw there