Meta today unveiled a groundbreaking advancement in neuroimaging by harnessing magnetoencephalography (MEG) to decode visual representations within the human brain with unparalleled precision.
Meta’s AI system, utilizing a unique three-part architecture, can reconstruct images perceived by the brain in real-time. This remarkable achievement promises to deepen our understanding of how the brain processes images and may even pave the way for clinical applications, including aiding individuals who have lost their ability to speak due to brain lesions.
The three-part system, consisting of an image encoder, brain encoder, and image decoder, aligns MEG signals to deep image representations, producing a continuous flow of images decoded from brain activity.
Meta: “MEG recordings are continuously aligned to the deep representation of the images, which can then condition the generation of images at each instant.”
The researchers trained their AI system using a publicly available dataset of MEG recordings from healthy volunteers. This dataset was made accessible by “Things,” an international consortium of academic researchers who share experimental data from the same image database.
They evaluated the AI’s performance by comparing it with various pre-trained image modules and found that the brain signals closely aligned with modern computer vision AI systems like DINOv2. This demonstrated that self-supervised learning in AI systems can result in brain-like representations, with artificial neurons responding similarly to the brain’s neurons when presented with the same images.
“This research strengthens Meta’s long-term research initiative to understand the foundations of human intelligence, identify its similarities as well as differences compared to current machine learning algorithms, and ultimately guide the development of AI systems designed to learn and reason like humans,” the company wrote.
While the generated images are imperfect, this research is a significant step forward in understanding human intelligence, highlighting the convergence between AI systems and the brain and guiding the development of more human-like AI systems.
Related: AI model MinD-Vis can read people’s minds based on brain activity
The post Meta Achieves AI Breakthrough in Deciphering Visual Brain Activity via MEG appeared first on Metaverse Post.