monkey sea
Researchers from the National University of Singapore and the Chinese University of Hong Kong claim to have developed an AI that can reconstruct “high-quality” videos from brain signals.
As the researchers describe in their non-peer-reviewed paper, the AI model, dubbed MinD-Video, uses publicly available data from fMRI measurements, specifically, the are “co-trained” on the basis of data obtained from cases that have undergone A video of their brain activity being recorded and an augmented model of the AI image generator Stable Diffusion.
Using this “two-module pipeline designed to bridge the gap between image brain decoding and video brain decoding”, “high quality” AI generation of the video first shown to the participants I was able to purely generate a reconstruction of . their brain readings.
According to the researchers, their model was able to reconstruct these videos with an average accuracy of 85% based on “various semantic and pixel-level metrics.”
“Understanding the information hidden within our complex brain activity is a great mystery in cognitive neuroscience,” the paper reads. “We showed that Mind-Video using adversarial induction can reconstruct high-quality video at arbitrary frame rates.”
input/output
The new paper builds on the researchers’ previous efforts to use AI to analyze only brain waves to reconstruct images.
Overall, the AI’s new video rendering is very impressive, as demonstrated by a head-to-head comparison of the original and “reconstructed” videos on the researchers’ website.
For example, a video of a crowd walking down a busy street translated into an equally crowded scene, albeit with more vibrant colors. The underwater scene of colorful fish has become a more vivid underwater scene.
However, the effect is far from perfect. For example, a jellyfish video was inexplicably converted into a clip of swimming fish, and a sea turtle video was reinterpreted as fish footage.
brain reading helmet
The researchers argue that these AI generations can also provide neurological insights, for example demonstrating the dominance of the visual cortex in the process of visual perception.
While this research is interesting, we are still a long way from the future where we can put on a helmet and get a perfectly accurate AI-generated video stream of what is floating around our skulls.
Frankly, considering the data privacy implications, this is probably a good thing.
Learn more about AI: AI Expert Says ChatGPT Is Stupid Than People Think
