Summary: Researchers show that recurrent, wave-generating cortical circuits continuously tune their synaptic weights to internalize the statistical structure of the external world. Functionally resembling biological generative models, these traveling neural waves help the brain infer sensory causes, form internal representations, produce short-term predictions, and replay sequential memories of events unfolding in time.
Key Facts
- The Visual Computational Engine: Recurrent cortical circuits generate traveling waves that modulate moment-to-moment perception, shape sensory representations, predict imminent inputs, and replay temporal sequences of events.
- Biological Generative Modeling: Similar to how artificial language models learn statistical patterns to predict and generate text, traveling waves encode environmental regularities into synaptic networks through experience, creating a biological form of generative modeling.
- Explaining Perceptual Mysteries: Building on the discovery of traveling waves in awake visual systems, this framework explains why an object directly in front of an observer can sometimes be missed depending on the instantaneous phase of a passing wave.
- Adaptive Synaptic Weighting: Waves are not passive feedforward relays; the recurrent connections that generate them actively adapt their synaptic weights in response to sensory input and experience.
- Predictive Inference in Noisy Environments: By internalizing 3D spatial structure, physics, and patterns of body and eye movement, traveling wave dynamics enable the brain to infer likely causes of noisy sensory inputs.
Source: Salk Institute
Your brain has something surprising in common with the ocean: waves. Electrical activity sweeps across the brain’s surface as traveling neural waves. These waves influence behavior and attention and can arise from internal network dynamics or from external sensory events.
A new review by neuroscientists at the Salk Institute synthesizes physiological and computational findings about these neural traveling waves and presents a unifying hypothesis: traveling waves act as a core computational mechanism in the visual cortex. Through these dynamics, the visual cortex—and likely other cortical areas—builds internal models of the world, enabling perception, prediction, reconstruction, and memory replay.
The review was published in Neuron on July 21, 2026.
What are traveling brain waves, and why do they matter?
Dr. John Reynolds and colleagues first described traveling waves in the visual systems of awake animals in 2020. Their work showed that whether an animal perceives a visual object can depend on where and when a wave passes through visual cortex. That discovery raised a central question: what computational function do these waves serve?
According to the review, neural traveling waves allow the visual cortex to: (1) modulate perception on short timescales, (2) turn recent sensory input into evolving internal representations, (3) generate short-term predictions about what will be seen next, and (4) store and replay sequences that represent temporal memories.
How do these waves change our view of brain function?
Traditionally, rhythmic cortical activity was sometimes treated as background noise or as a simple timing signal. This review reframes traveling waves as active computational engines. The recurrent circuits that produce waves do more than pass signals forward: their synaptic strengths are shaped by experience, so the waves themselves reflect learned regularities of the world.
Every sensory encounter—sights, sounds, movements—adjusts the synaptic wiring that gives rise to these waves. Over time, that wiring encodes stable patterns of the environment, allowing waves to embed sensory history into the brain’s ongoing activity. In this way the cortex constructs internal models that make sensory interpretation more robust and efficient.
“This is, in a meaningful sense, analogous to what large language models do,” Reynolds explains. “They learn statistical structure from data and use that knowledge to generate structured outputs. The brain may implement a comparable, biologically grounded generative model through synaptic plasticity and wave dynamics.”
When the brain receives noisy, ambiguous input it needs to answer: what am I most likely sensing now? The world is complex but constrained: objects exist in three dimensions, retinal images shift with eye and body movements, and physical laws constrain how sensory signals evolve. The review proposes that traveling waves encode these regularities so the cortex can infer probable causes and produce coherent percepts and predictions.
Other authors and funding
Additional authors include Lyle Muller, Alexandra Busch, and Zachary Davis.
Funding: National Institutes of Health (grants including R01 EY028723, U01 NS131914, U01 NS139877, EY014800), Research to Prevent Blindness, Natural Sciences and Engineering Research Council of Canada, Western University, Compute Ontario, and the Digital Research Alliance of Canada.
Key Questions Answered
Q: What are neural traveling waves and how were they discovered in awake brains?
A: Neural traveling waves are coordinated sweeps of electrical activity that propagate across the cortical surface. Dr. John Reynolds first reported them in the visual systems of awake animals in 2020 and showed that perception of a visual object can depend on the timing and position of a passing wave.
Q: How do traveling brain waves act like a “biological generative model”?
A: Recurrent cortical circuits learn statistical and physical regularities of the visual environment through synaptic changes. Traveling waves use those learned patterns to fill in missing details, anticipate upcoming changes, and generate internal representations of the world, much like artificial generative models do for language or images.
Q: Why is shifting from “electrical noise” to “computational engine” significant?
A: Viewing traveling waves as active computational mechanisms highlights their role in transforming messy sensory inputs into structured, predictive percepts. Rather than passive rhythms, waves become a low-energy, spatiotemporal substrate for complex cortical computations.
Editorial Notes
- This article was edited by a Neuroscience News editor.
- Journal paper reviewed in full.
- Additional context added by staff.
About this visual neuroscience research news
Author: Salk Communications
Source: Salk Institute
Contact: Salk Communications – Salk Institute
Image: Image credit: Neuroscience News
Original Research: Open access. “Neural traveling waves in cortex: network mechanisms and potential roles in neural computation” by John Reynolds, Lyle Muller, Alexandra Busch, Zachary Davis. Neuron. DOI: 10.1016/j.neuron.2026.06.019
Abstract
Neural traveling waves in cortex: network mechanisms and potential roles in neural computation
First observed in anesthetized animals, neural traveling waves (nTWs) are now widely reported across awake brains, where they influence excitability and behavior. nTWs can emerge intrinsically from ongoing network dynamics or be evoked by sensory inputs and actions.
By structuring activity within cortical regions, nTWs introduce spatiotemporal dependencies across sensory maps that are not captured by strictly feedforward or feedback models.
This review synthesizes physiological and computational evidence around two themes, with a focus on the visual system and connections to other cortical areas. First, it defines nTWs and outlines circuit mechanisms capable of generating them. Second, it highlights how nTWs can implement spatiotemporal computations—such as predicting upcoming sensory inputs—by embedding sensory history into evolving patterns of cortical activity.
The review concludes by proposing a conceptual framework for spatiotemporal generative processing implemented by traveling waves moving over sensory maps.