Mammalian brains can continuously scale up their capacity to process sensory data despite widespread neural noise. Published in the journal Science Advances, the study overturns a three-decade-old mathematical consensus that previously limited how much information large neural networks could handle.
The Three-Decade Mathematical Consensus
For thirty years, mathematical models suggested a strict upper limit on how many details neural circuits could represent.
Scientists reasoned that because biological neurons are tightly interconnected within complex chemical environments, any random fluctuation or noise in a single cell’s activity would spread directly to its neighbors. Researchers assumed that as neural populations grew larger, this accumulated noise would eventually cancel out any gains in information processing.
Mapping Twenty Thousand Neurons in Mice
To test this long-standing assumption, a research team analyzed precise experimental records covering roughly 20,000 neurons across the primary visual cortices of five mice.
Trial-to-Trial Variation and Visual Signals
When researchers repeatedly displayed identical images to the test subjects, individual neurons in the visual cortex did not fire in the exact same pattern every time. This trial-to-trial variation is what neuroscientists define as neural noise.
When an image changes slightly, the average activity of the neural population shifts, providing the signal the brain uses to distinguish between different inputs.
Overcoming Shared Noise Through Scale Invariance
Mathematical analysis of these patterns revealed that shared fluctuations across cells do not form a chaotic block that obstructs vision. Shimazaki noted that all five mice in the study fell squarely into a category where information continually increases. Shared noise slows down the rate at which information grows, but it never halts the process entirely.
When minor shifts in visual lines act as directional signals, the associated noise disperses across the activity space of the neurons through a mathematical framework known as scale invariance. Shimazaki pointed out that scale invariance means a phenomenon maintains its structural properties regardless of differences in scale.
Bypassing Traditional Computational Bottlenecks
These findings show that biological brains bypass traditional computational bottlenecks.
By using a scale-invariant noise distribution, mammalian neural networks are able to continue extracting meaningful data from noisy environments as population sizes expand.
