Researchers at NYU Langone Health have identified the brain region responsible for “one-shot learning,” a phenomenon where seeing something once can dramatically improve the ability to recognize it later. The study, published in Nature Communications, focuses on perceptual learning—the moment when a person first recognizes a blurry object.
The research team pinpointed the high-level visual cortex (HLVC) as the area where “priors”—images seen and stored from past experiences—are accessed to enable this rapid form of learning. “Our work revealed, not just where priors are stored, but also the brain computations involved,” said Biyu J. He, PhD, associate professor in the Departments of Neurology, Neuroscience, and Radiology at NYU Grossman School of Medicine.
Previous studies had shown that patients with schizophrenia and Parkinson’s disease experience abnormal one-shot learning, sometimes leading to hallucinations when previously stored priors overpower current perception. “This study yielded a directly testable theory on how priors act up during hallucinations, and we are now investigating the related brain mechanisms in patients with neurological disorders to reveal what goes wrong,” Dr. He added.
The team is also examining connections between these mechanisms and other types of insight or “aha moments.” Their research used Mooney images—faded pictures of animals and objects—to study changes in recognition before and after participants saw clear versions of the same images. Earlier work by Dr. He found that subjects doubled their ability to recognize such images after exposure to clear versions.
To investigate further, researchers combined functional magnetic resonance imaging (fMRI), behavioral tests using Mooney images, electroencephalography (EEG), and machine learning models. They manipulated image size, position, and orientation to see how each affected recognition rates. Results showed that while changing an image’s size did not affect one-shot learning, rotating or moving it partially reduced recognition. This suggests that perceptual priors encode specific patterns rather than abstract concepts.
Statistical models based on fMRI data indicated that only neural coding patterns in the HLVC matched those revealed by behavioral studies. Intracranial EEG (iEEG) was also used with patients undergoing neurosurgical monitoring; HLVC showed early changes in signaling strength during prior-guided object recognition.
Finally, the researchers built an artificial intelligence model called a vision transformer to simulate this process. Like the HLVC in humans, one module stored accumulated image information as priors and another used this data for improved recognition. After training on enough images, this AI model achieved human-like one-shot learning capabilities.
“Although AI has made great progress in object recognition over the past decade, no tool has yet been capable of one-shot learning like humans,” said Eric K. Oermann, MD, assistant professor in the Departments of Neurosurgery and Radiology at NYU Langone. “We now anticipate the development of AI models with humanlike perceptual mechanisms that classify new objects or learn new tasks with few or no training examples. This is more evidence of a growing convergence between computational neuroscience and advances in AI.”
Other contributors included Ayaka Hachisuka and Jonathan Shor from NYU Langone’s Institute for Translational Neuroscience; Xujin Chris Liu from NYU Tandon School of Engineering; Daniel Friedman, MD; Patricia C. Dugan, MD; Orrin Devinsky, MD; Werner Doyle, MD; Yao Wang from NYU Tandon School of Engineering; Ignacio Saez from Icahn School of Medicine at Mount Sinai; and Fedor Panov from Mount Sinai’s Department of Neurosurgery.
The project received support from a W.M. Keck Foundation medical research grant, National Science Foundation grant BCS-1926780, and NYU Grossman School of Medicine funding. Disclosures noted Dr. Oermann holds equity in several companies and consults for Google and Sofinnova Partners under New York University policies.
NYU Langone Health operates as an integrated health system known for its focus on quality care outcomes across multiple locations including two tuition-free medical schools in Manhattan and Long Island as well as extensive outpatient services in New York and Florida regions.
