We come up with the new AI advancement, Deep Image Reconstruction. In a study led by researchers from the Advanced Telecommunications Research Institute International (ATR) and the University of Kyoto in Japan, it shows how they have built a neural network. That study shows, in addition to reading what we think, the new technology is also capable of recreating it.
How Deep Image Reconstruction works?
They have created an algorithm that can accurately interpret and reproduce images seen or imagined by a person. That is a system of deep image reconstruction based on human brain activity. That is a co-work of the scientists Guohua Shen, Tomoyasu Horikawa, Kei Majima and Yukiyasu Kamitani.
So, the new algorithm can generate unique and recognizable images from scratch. They only need to observe the brain activity of the person. And keep in mind that an image does not consist merely in pixels or simple forms. There exist different levels of characteristics or components of various complexities. That is obtained hierarchically to obtain the results.
Three healthy subjects with normal vision participated in the study and saw images in three categories. The visual stimuli consisted of natural images, artificial geometric shapes and alphabetic letters. This activity of the visual cortex feeds a neural network. That is then trained to produce a result that matches the visual information that the subject is seeing, as seen in these videos:
Also, the activity in the visual cortex measured using functional magnetic resonance imaging. Also, that results in hierarchical features of a broad neural network. The subjects looked at more than 1000 images, which included a fish, an airplane, and pure color forms, and the computer managed to interpret it correctly