- New study uses brain waves to reconstruct images using EEG data
- Subjects were shown facial images and a machine learning algorithm was used to simultaneously recreate the image
- Far ranging applications including helping verbally challenged people communicate and police investigations
Machine learning continues to help us take a step closer to the AI powered worlds we have seen previously in movies like The Matrix and Blade Runner. We have previously covered a study by Japanese scientists on an algorithm that could potentially read out minds.
Now, researchers from the University of Toronto have developed a new algorithm that reconstructs images of what people perceive based on their brain activity. The team, led by Dan Nemrodov, has developed this technique based on electroencephalography (EEG) data.
A few people were chosen for a test run of the algorithm. They were hooked up to the EEG equipment and were shown images of a few faces. Simultaneously, their brain activity was recorded and used to digitally recreate the image that was in the subject’s mind.
While the results have been impressive, the researchers are more focused on the fact that EEG data can, in fact, be used for this type of image reconstruction. There have been major doubts about EEG in the past but this new study has put them to rest.
EEG can capture activity in the millisecond scale. In this study, the team was able to estimate that it takes our brain approximately 0.17 seconds to form a passable imitation of a face we saw earlier.
Our take on this
This is not the first study in the field but what makes this stand out is the use of EEG data. Because it’s relatively inexpensive and portable, it has the potential to reach far ranging industries. It could provide a means of communication for verbally challenged folks and help the police is reconstructing a suspect’s face, rather than relying on verbal descriptions from eyewitnesses.
Deep learning using neural networks continues to assist in more ways than we could possible have imagined a few years ago.
Subscribe to AVBytes here to get regular data science, machine learning and AI updates in your inbox!