Summary: Researchers propose that the brain estimates and corrects for the expected sensory-driven activation from an input—such as image contrast—so that what remains is a clearer neural signal of familiarity.
Source: University of Pennsylvania
When someone looks at an image they have seen before, even if it was only a single brief exposure, the brain produces a distinct signature that signals familiarity.
For years, neuroscientists explained this phenomenon with the repetition suppression hypothesis: neurons in a high-level visual area called the inferotemporal (IT) cortex respond less strongly to familiar images than to novel ones. High firing rates were taken to indicate novelty; lower firing rates, familiarity.
Nicole Rust, an associate professor in the Department of Psychology at the University of Pennsylvania, and colleagues found that this simple account fails to explain important observations. “Different images elicit very different levels of activation even when all are novel,” Rust explains. Similarly, basic sensory properties such as image brightness or contrast also change overall IT activation, which complicates interpreting reduced activity as a memory signal.
In a paper published in the Proceedings of the National Academy of Sciences, Rust, postdoctoral fellow Vahid Mehrpour, Penn research associate Travis Meyer, and Eero Simoncelli of New York University introduce an alternative interpretation. They propose that the brain first estimates the level of activation expected from the sensory input—taking into account factors like contrast—and subtracts that expected component. The residual signal reflects familiarity. The team calls this idea sensory referenced suppression.
The visual system and neural coding of memory
Rust’s lab combines precise neural recordings with computational modeling to understand how the visual system converts sensory inputs into interpretable patterns. Light enters the eye through rods and cones and is processed through a cascade of visual areas, culminating in the inferotemporal cortex. The IT contains millions of neurons whose firing patterns vary with what is being viewed: a particular face, a coffee cup, or a pencil produce distinct population responses that the brain interprets.
Beyond encoding object identity, IT activity has been linked to visual memory. Repetition suppression posits a literal threshold in firing rate: more spikes signal novelty, fewer spikes signal prior exposure. But total spike count is influenced by sensory factors beyond memory—contrast being a primary example—so the brain faces an ambiguity. Mehrpour summarizes the problem: “We propose that the brain corrects for sensory-driven changes, in our experiments primarily contrast, and what remains is the memory-related signal.”
Experimental approach
To test this idea, the researchers presented sequences of grayscale images to two adult male rhesus macaques, with each image shown exactly twice: once as novel and then again as familiar. The images were displayed at high and low contrasts in various combinations, and each presentation lasted 500 milliseconds. The animals were trained to report whether an image was new or repeated by making specific eye movements, and they were instructed to ignore contrast differences when making that judgment.
While the macaques performed the task, the team recorded spiking activity from hundreds of individual neurons in IT. This single-neuron resolution differs from approaches that average indirect measures over thousands of cells; it enabled the researchers to examine the neural code that supports behavior.
Using quantitative, linear decoding analyses, the investigators showed that a decoder based only on total spike count—consistent with classic repetition suppression—could not separate the effects of contrast from memory. In contrast, a decoder that first corrected the total spike count for contrast modulation accurately predicted the animals’ behavior. These results indicate that the neural pattern best aligned with single-exposure visual recognition memory is not raw repetition suppression but a sensory-referenced suppression: reductions in IT population response magnitude after correcting for sensory modulation.
Implications for neuroscience, AI, and medicine
The sensory referenced suppression framework reframes how researchers think about neural signatures of familiarity and recognition memory. By demonstrating that the brain can factor out sensory-driven changes—such as contrast—this work clarifies how a stable memory signal can be read out despite variable sensory inputs.
Vahid Mehrpour notes potential applications beyond basic neuroscience: “If we can understand how the brain represents and reconstructs memories in the face of changing sensory conditions, we can design artificial intelligence systems that mirror this robustness.” Translating these principles into AI could help build machines that recognize previously encountered inputs reliably across changes in lighting, contrast, or other sensory factors.
Nicole Rust highlights clinical relevance as well. A clearer understanding of how memory signals are generated and isolated in a healthy brain may inform strategies for diagnosing, preventing, or treating memory impairments, including conditions that affect aging populations.
Further research will be needed to map how sensory referenced suppression operates across different brain areas, sensory modalities, and time scales, and to explore its role in more complex forms of memory and perception. This study, the authors say, brings researchers a step closer to understanding the precise neural computations that underlie single-exposure visual recognition.
Funding: This research was supported by the Simons Foundation (grants 543033 and 543047), the National Eye Institute of the National Institutes of Health (Grant R01EY020851), the National Science Foundation (CAREER Award 1265480), and the Howard Hughes Medical Institute.
Vahid Mehrpour is a postdoctoral fellow in the Visual Memory Lab at the University of Pennsylvania.
Travis Meyer is a research associate in the Visual Memory Lab at the University of Pennsylvania.
Nicole Rust is an associate professor in the Department of Psychology at the University of Pennsylvania. She directs the Visual Memory Lab, co-directs the Computational Neuroscience Initiative, and serves as MindCORE’s executive director for research.
Eero Simoncelli is a professor of neural science, mathematics, data science, and psychology at New York University and founding director of the Center for Computational Neuroscience at the Simons Foundation’s Flatiron Institute.
About this neuroscience research news
Source: University of Pennsylvania
Contact: Michele Berger – University of Pennsylvania
Image: The image is in the public domain
Original Research: Closed access. “Pinpointing the neural signatures of single-exposure visual recognition memory” by Nicole Rust et al., PNAS.
Abstract
Pinpointing the neural signatures of single-exposure visual recognition memory
Memories of images we have seen are often associated with reduced neural responses in high-level visual areas such as inferotemporal cortex, a phenomenon previously framed as repetition suppression (RS).
The authors tested this hypothesis with a task requiring rhesus monkeys to report whether images were novel or repeated while ignoring variations in contrast, a stimulus attribute that also modulates IT responses.
The monkeys’ behavior remained largely invariant to contrast, contrary to the predictions of an RS-based decoder that could not distinguish repeated images from lower-contrast novel images. Instead, behavioral patterns matched a linearly decodable model in which total spike count was corrected for contrast modulation.
These results suggest that the IT neural activity pattern most aligned with single-exposure visual recognition memory is not raw repetition suppression but sensory referenced suppression: reductions in population response magnitude after correcting for sensory modulation.