Scientists at the Salk Institute have determined that neural waves in the visual cortex function as a computational mechanism that enables the brain to construct internal representations of the three‑dimensional external world. The researchers demonstrated that these electrical impulses, which spread across the cortical surface, are not mere background noise but perform essential spatiotemporal computations needed to predict and perceive reality.
The brain uses electrical activity to communicate between synapses. Back in 2020, a team led by neuroscientist John Reynolds identified a direct correlation between running waves in the visual system of awake animals and their ability to detect objects right in front of them. This helped explain the classic phenomenon where a subject struggles to find an item that is clearly visible. However, until now, it remained unclear why these waves have such a pronounced effect on perception.
In their new work, Reynolds and his colleagues propose a concept whereby the neural circuits generating these running waves constantly adjust their physiological structure — or synaptic weights — to adapt to the external world. The study authors explicitly compare this process to the operation of large language models (LLMs). Just as neural networks learn the statistical structure of data to generate meaningful responses, the brain uses running waves as a biological generative model built from scratch based on sensory experience.
These dynamic patterns arise from the intrinsic connectivity of neural networks and allow the brain to integrate spatial and temporal information more effectively than standard hierarchical models. Research confirms that such waves not only influence neuronal excitability but also directly determine the accuracy of perception and the speed of an organism’s reactions.
In the visual cortex, the waves perform four key functions: they modulate moment‑to‑moment perception, transform incoming visual data into coherent images, generate short‑term predictions about the environment, and maintain and replay memory patterns over time.
This type of dynamics — referred to as neural traveling waves (nTW) — is a fundamental and widespread feature of brain function, observed across various spatial and temporal scales in multiple species, including fish, mammals, and humans. In the visual cortex, both spontaneous and stimulus‑evoked waves modulate neuronal excitability and directly affect perceptual thresholds, thereby determining sensitivity to external signals. In the motor cortex, beta‑range waves predict the timing of movement initiation, while similar wave processes in the prefrontal and parietal cortices are linked to working memory mechanisms. Beyond the neocortex, these waves play an equally important role in memory encoding and in organizing the signals required for reinforcement learning. At the whole‑brain level, this dynamics is most prominent during sleep, highlighting the global role of wave processes in supporting cognitive functions.
John Reynolds’ team recorded this dynamics across multiple spatial and temporal scales throughout the brain. The impulses can propagate continuously or discretely via long horizontal fibres that connect neurons within a given region.
Traditional views depicted the visual system as a strict hierarchy for signal transmission. Neural traveling waves expand this understanding by adding a “third dimension” to conventional hierarchical models: spatiotemporal computation within individual brain areas. Experimental evidence confirms that these wave processes directly govern perception and behaviour by modulating neuronal excitability in real time.
Unlike classical systems focused solely on straightforward signal relay, these processes retain a “history” of stimulation, enabling the brain to build short‑term forecasts based on the relationship between the time and location of prior events. This forms the basis for generative data processing: neural networks leverage space‑time dependencies to extract features from dynamic scenes and create a flexible internal representation of the external world — a functional parallel between biological processes and artificial intelligence models.
About the Author
