Friday, March 29, 2024
HomeSECURITYResearchers have learned to generate images based on human brain activity

Researchers have learned to generate images based on human brain activity

-


Researchers have learned to generate images based on human brain activity

In such a difficult process, artificial intelligence models once again came to the rescue.

Japanese scientists at Osaka University’s Graduate School of Advanced Biological Sciences have been able to reconstruct high-fidelity images of human brain activity using a popular artificial intelligence (AI) imaging platform — Stable Diffusion.

If we discard all scientific terminology, such a complex process can be briefly described as follows: scientists showed subjects a specific image, and then asked them to imagine this image and “hold it in their head” during functional magnetic resonance imaging (fMRI). Then, based on the data collected, the scientists processed the received fMRI signals several times using complex algorithms, and as a result, they used them as input data from Stable Diffusion to generate the final picture.

The neural network was able to generate very close to the original images. This is a great achievement, given the possible data loss in the process of reading brain activity.



Top – original images, bottom – obtained from fMRI signals using a neural network

The researchers reported that they were able to obtain high-resolution image reconstructions, but only after extensive training and fine-tuning of the generative model. This leads to certain limitations, since the process of training generative models is extremely difficult, and in neuroscience there are not many samples to work with. Prior to this study, no other scientist had attempted to use neural network models to visually reconstruct brain activity.

With the development of generative AI, more and more researchers are studying how neural network models can work with the human brain. Recent research shows that artificial intelligence can do a truly impressive job of reconstructing brain signals. This is largely due to the fact that neural network models themselves find algorithms that allow them to arrive at the desired result. And far from always these algorithms are understandable for human perception.

Who knows, maybe in the near future scientists will be able to use neural networks to read human dreams and decode them into videos? We live in an amazing time, no doubt about it.



Source link

www.securitylab.ru

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Most Popular