A model of the brain’s visual cortex suggests how neural activity can remain highly variable without making perception unstable. Researchers found that mechanisms resembling excitation, inhibition and slower self-regulation can prevent tiny differences in initial brain activity from producing radically different outcomes.
The team, led by Mehdi Borjkhani, PhD, from the International Centre for Translational Eye Research (ICTER) at the Institute of Physical Chemistry of the Polish Academy of Sciences, modelled activity in the primary visual cortex, an early stage of vision that responds to features such as edge orientation. The researchers collaborated with institutions in Iran and Germany, and their findings were published in the Journal of Computational Neuroscience.
Instead of modelling facial recognition, the researchers described the functioning of the visual cortex using three variables. The first represented the activity of excitatory neurons, which facilitate the activation of subsequent cells. The second described inhibitory neurons, which hinder such responses and prevent excitation from spreading uncontrollably. Rather than shutting down vision, they regulate the network’s operation. The third represented the network’s overall readiness to respond, a state influenced by factors such as signals arriving from the thalamus. This variable changed more slowly, on a longer timescale than the processes of excitation and inhibition.
The researchers first examined what would happen if the three components of the model were left without stabilizing mechanisms. They tested 225 different settings for excitation strength, inhibition strength and the network’s overall readiness. In nearly nine out of ten cases, the model behaved chaotically: two simulations starting from almost identical conditions increasingly diverged over time.
This can be compared with two paper airplanes launched in seemingly identical ways that ultimately land in entirely different places. In a chaotic system, an initially minute difference can amplify until it leads to a completely different result.
The researchers then equipped the model with two mechanisms resembling regulators found in the actual cortex. The first responded immediately: as excitation increased, the activity of inhibitory neurons also increased, limiting its further spread. The second acted gradually, restoring the network’s overall readiness to a normal level.
Once these safeguards were in place, most activity patterns ceased to diverge unpredictably. In one of the tested configurations, variability in excitatory neuron activity dropped by 93 percent. However, the model was neither silenced nor immobilized. Its activity continued to fluctuate, but instead of wandering along a different path each time, it settled into a stable, repeatable rhythm.
The crucial question was whether the same stimulus could drive initially different network states toward a similar response. To test this, the researchers ran the model 30 times, slightly altering its initial state each time. Before the stimulus appeared, activity unfolded in various ways. Once the stimulus was applied, the divergent patterns began to converge. The model thus behaved similarly to the actual cortex, where the appearance of a specific image constrains the prior variability of neuronal activity.
This reflects a familiar feature of perception. In a well-known drawing, one can see either a white vase or two black profiles. The image itself does not change, yet we perceive first one thing and then another. Perception, therefore, is not merely an unambiguous description of the image on the retina; the outcome can also depend on emotional state, level of focus and past experiences. Sometimes, the same image can trigger processes leading to vastly different conclusions or effects.
An illusion is not a strict example of chaos in the mathematical sense, but it helps illustrate the concept. Similarly, two glances can trigger increasingly divergent responses.
Yet we can recognize the face of a loved one in bright sunlight or dim light, from the front or in profile, and even when it is partially obscured. Such retinal images differ to such an extent that, were one to attempt a computer-based comparison, it would be difficult to devise an algorithm capable of identifying that they depict the same object. Yet our perception quickly and reliably leads to the same interpretation; the resulting conclusions do not diverge but rather converge.
The model was not designed to recognize faces. It was assigned a much simpler task corresponding to an early stage of vision: responding to lines oriented at various angles. Individual virtual neurons responded more strongly to specific orientations, and the degree of these preferences fell within the range observed in the visual cortex of macaques.
Interestingly, in the chaotic variant, the model distinguished the preferred line orientation from the others somewhat more clearly than in the regular variant. This does not prove that chaos improves human vision. It does, however, suggest that the brain’s goal is likely not the elimination of all irregularity. A certain degree of variability may enhance sensitivity and responsiveness, provided it does not spiral out of control.
The study suggests that the brain should be viewed neither as a perfectly regular clock nor as a system plunged into total chaos. It may possess a natural tendency toward a wide range of activity patterns, yet excitation, inhibition and slower self-regulatory processes keep this tendency in check. When a stimulus appears, these mechanisms ensure that varying initial states lead to a similar response.
While the model does not yet explain how we recognize a familiar face, it demonstrates a possible mechanism for a more fundamental ability: transforming variable brain activity into a stable response to what we see. (PAP)
kmp/ agt/
tr. RL
