The integration between stimuli received in the classical receptive field (CRF) and in surrounding regions in the primary visual cortex (V1) remains a central issue in computational neuroscience. The recent study by Obeid and Miller (2025) proposes an innovative model based on a Stabilized Supralinear Network (SSN), demonstrating its ability to reproduce fundamental properties of surround suppression observed in V1.
Unlike traditional linear models, SSN employs transfer functions with supralinear growth that, when coupled with biologically plausible lateral connectivity, generate nontrivial behaviors consistent with the physiology of the visual cortex. The proposal is ambitious: to replicate three crucial phenomena of surround suppression—(1) the simultaneous suppression of excitatory and inhibitory currents; (2) maximal suppression occurring when the orientation of the peripheral stimulus matches that of the central stimulus, regardless of the neuron’s preferred orientation; and (3) selective suppression of orientational components in compound stimuli such as plaids (crossed grating patterns).
The first point is supported by experimental data demonstrating that suppression by surrounding stimuli is not due solely to an increase in inhibition but to a joint reduction of synaptic excitation and inhibition—a result that requires a local connectivity architecture strong enough for the network to become unstable without feedback inhibition. Such a network then becomes an inhibition-stabilized network (ISN), as suggested by Ozeki et al. (2009). The model developed by Obeid and Miller confirms that, to replicate this behavior, it is necessary that local excitatory connections are not only strong but also relatively homogeneous with respect to orientation within the spatial boundaries of the immediate surroundings. This observation provides an important refinement to previous models, which either did not consider such features or were limited to one-dimensional or simplified versions of cortical circuits.
The second aspect addressed by the model—that suppression is more effective when there is correspondence between the orientations of the central and peripheral stimuli—was shown to depend on a specific anatomical arrangement: local connections broadly tuned to orientation. This means that the recorded cell, when receiving inputs preferentially from other cells tuned to the orientation of the central stimulus, will be more sensitive to suppression coming from a peripheral stimulus that shares the same orientation as the center. The model therefore shows that this suppression does not depend on the preferred orientation of the neuron, but rather on the functional profile of the microcircuit to which it belongs.
The third phenomenon—selective suppression of specific orientational components in compound stimuli—was replicated in the model through specific connectivity between neurons with similar orientations. The presentation of a central plaid stimulus, followed by a peripheral stimulus with the orientation of one of the plaid components, led to the selective suppression of that component, evidencing an “input-gain” mechanism, as proposed by Trott and Born (2015). It is important to note that, according to the authors, this behavior does not require the same connectivity structure necessary for the observation of the second phenomenon—a distinction that grants functional autonomy to each mechanism and questions the hypothesis of a common origin.
Another notable point of the study is the verification that these properties emerge in both rate-based models and more biologically realistic models based on membrane potentials and conductance synapses (spiking models). Furthermore, the model is consistent with the rapid decay of cortical activity observed when thalamic input is silenced—a crucial finding, since models with strong recurrent excitation tend to prolong activity. Balancing with inhibition—even in strongly connected regimes—seems to be essential to avoid slow dynamics that do not match the physiological data of V1.
In summary, Obeid and Miller’s work represents a robust contribution to the understanding of the dynamics of sensory cortical circuits, suggesting that local mechanisms, structured in a biologically plausible way, are sufficient to generate complex patterns of visual response. In my reading, especially when compared with previous work that relies heavily on modular architectures or descending inputs from higher areas, this model suggests a surprising sufficiency of local circuits in V1. This finding reinforces the role of intracortical self-organization in the emergence of the functional properties of visual perception.
Reference:
OBEID, Dina; MILLER, Kenneth D. Stabilized Supralinear Network Model of Responses to Surround Stimuli in Primary Visual Cortex. eNeuro, v. 12, no. 5, 2025. DOI: https://doi.org/10.1523/ENEURO.0459-24.2025.
