A study in mice suggests the brain may start making sense of the visual world much earlier in the perception process than neuroscientists realized.

Researchers in South Korea found evidence that the primary visual cortex does more than provide a simple “first pass” at vision — it extracts meaningful statistical patterns from chaotic scenes, a task long associated with higher brain regions.

“This early statistical compression is what allows us to walk into a crowded room, glance at a busy highway, or watch a flock of birds and instantly ‘get’ what is happening without being overwhelmed by the details,” says Young-Beom Lee, a post-doctoral researcher at the Institute For Basic Science’s Center for Memory and Glioscience.

Testing ensemble perception in mice

The research team explored a phenomenon known as “ensemble perception,” in which the brain pools many visual elements into a compact summary. “I wanted to understand how the brain throws away unnecessary detail while still preserving the most useful structure of the scene,” says Prof. Yee-Joon Kim, who co-led the study with Prof. Doyun Lee, together at the Institute for Basic Science’s Center for Memory and Glioscience.

To investigate this phenomenon, the researchers showed head-fixed mice screens displaying hundreds of randomly moving dots and trained them to categorize overall motion direction by turning a wheel left or right.

“In our stimuli, every single dot moved in its own direction, spread across a range,” Lee explains. “The average direction existed only as a kind of mathematical summary across all the dots.” 

This experimental design prevented mice from solving the task by simply tracking a single moving dot. Successfully identifying the motion category (left or right) required the brain to compute the average motion direction across many dots simultaneously. “The mice performed this feat accurately, even with stimuli so noisy that even humans had difficulty identifying the average direction,” says Lee.

To understand how the brain achieved this feat, the researchers examined neural activity in brain regions linked to vision.   

Finding order in chaos

Using miniature calcium-imaging microscopes, the researchers recorded activity from neurons in the primary visual cortex (V1), associated with early visual processing, and the posterior parietal cortex (PPC), which is linked to higher-order decision-making and spatial reasoning. 

They found that populations of neurons in V1 encoded not only the average motion direction but also variability in the visual scene. 

“We showed that V1 doesn’t just relay raw signals — it already computes statistical summaries,” says Lee.  

Kim cautions that the findings may reflect important differences between how mice and primates process visual information. Unlike primates, mice lack foveal vision — the high-resolution vision humans use to focus on fine details rather than broad peripheral information.

“For a nocturnal animal such as a mouse, the early visual cortex may be adapted not only to represent fine details, but also to extract useful statistical structure from a broad visual field,” Kim explains.

He says future studies comparing nocturnal and diurnal animals could help determine whether this kind of early statistical processing is especially prominent in animals adapted to low-light environments.

The study also revealed an important difference between individual neurons and large neural populations. Single neurons were often unreliable or inconsistent when predicting the average motion direction. But when many neurons across V1 worked together as a group, their combined activity accurately represented the average direction.

Kim compares the phenomenon to a jury. “A single juror might be biased or miss something, but a large group of jurors, taken together, tends to reach a more reliable verdict,” he says.

Learning reshapes early vision

The researchers also found that learning shaped very early visual processing. After mice learned to distinguish between leftward and rightward motion, neural activity in V1 shifted slightly toward the learned category during the task. 

“That suggests early visual processing is not completely fixed,” Kim says. “Higher-level learning or task demands can influence even early stages of sensory representation.” 

“The brain appears to work like a compression pipeline — each step discards some detail but extracts what matters most for guiding behavior,” Lee adds.

The work may also carry implications beyond neuroscience. Modern artificial intelligence systems often process visual data in great detail and at high computational cost. The brain’s strategy — rapidly extracting statistical summaries before refining details — could inspire more efficient machine-vision systems.

“Rather than building AI that simply processes more data or runs more complex calculations,” Lee says, “we might instead build systems that are smarter about what they pay attention to — just as the brain is.”

Reference: Young-Beom Lee et al., Hierarchical Summary Statistics Encoding Across Primary Visual and Posterior Parietal Cortices, Advanced Science (2026), DOI: 10.1002/advs.202512369.

Featured Image Credit: Anni Roenkae via Pexels

Corrections to this article after publication: We have corrected Young-Beom Lee’s academic position and made the roles of Prof. Yee-Joon Kim and Prof. Doyun Lee explicit, including minor text corrections to accommodate these changes. Our apologies for these errors. The explanations of the research are unaffected.